How to compare Magento agencies, and what to check before you shortlist

This page scores disclosure, not delivery quality, so a low score means an agency publishes little that can be checked rather than that it does poor work. It exists because comparing agencies is hard for a reason nobody says out loud: almost nothing on an agency website is checkable, and a surprising amount of it is contradicted somewhere else on the same site. Nine checks are set out below, each with the failure it catches. Ten agencies were then read on their own websites on 23 September 2026 and scored out of 100 as the worked example. scandiweb ranks first on 76, sixteen points ahead of Rocket Web on 60, with Divante third on 57. The findings are the useful part. Four of the six agencies here that hold a Hyva partner tier never mention it, and three of those four sell Hyva work by name, because Hyva publishes a curated page showing three tiers and a full register showing five and 460 listings. GPMD says it has 25 years of experience and that it started in 1999, which is 27. Tigren says 13 years and since 2012, which is 14. Webkul says 16+ years on its homepage and 15 on its about page, and 500+ people on one and 650+ on the other. Meetanshi publishes 18,000 customers and 13,000 brands within a few lines of each other. Overdose names seven cities in a sentence and lists four with addresses. All four of Divante's flagship case studies publish a measured result and not one names Magento. Five other agency domains opened while researching this page now redirect to whoever bought them. scandiweb is held to the identical test and scores 12 of 25 on consistency, beaten by four of the nine agencies it outranks. It publishes three different figures for its own age, 20+ years on one page, 22+ on another and 23+ on a third, and only 23+ agrees with the 2003 founding it publishes alongside them; and it gives two different start years for its Magento work in two sentences on one page. It also publishes two figures that are both correct and labelled in a way a reader cannot resolve: two country counts that turn out to measure different things, its workforce and its client footprint, and the number 600+ used for its specialists on four pages and for its brands on a fifth, where the client figure is 700+. Those last two are labelling failures rather than conflicting claims. And it publishes no support response or resolution target at all.

Kristaps GailitisCMO at scandiwebReviewed 23 September 2026

1 The shortlist

The ten agencies used as the worked example, in order

scandiweb Buyers who need the largest published body of checkable figures, and who will ask which of its three published ages is the right one 76 of 100.

Rocket Web Buyers who would rather be told four things they can verify than forty they cannot, and who are weighing whether to stay on Adobe Commerce at all 60 of 100.

Divante Buyers running a composable or headless programme, who should confirm the Magento part specifically because the published work does not evidence it 57 of 100.

Tigren Smaller budgets wanting a complete published stat block, who will ask which of two published durations is the right one 40 of 100.

Webkul Buyers who want an extension and marketplace specialist, and who will pin down which of its two headcounts and two durations apply 38 of 100.

Meetanshi Buyers wanting Magento extensions and smaller service engagements, who will ask which of two customer figures on one page is the real one 37 of 100.

Ranosys Buyers who will take a client roster on trust, or who are prepared to ask for every number in writing before the first meeting 33 of 100.

DCKAP Distributors and manufacturers wanting ERP integration first and a storefront second, who should ask what the current Adobe partner status is 30 of 100.

Overdose Australian and New Zealand programmes wanting strategy and marketing alongside build, who should confirm which offices actually exist 26 of 100.

GPMD London buyers who want a long-established independent and are willing to source every figure themselves 22 of 100.

This page scores disclosure, not delivery quality. Every criterion measures whether a claim can be checked, not whether the work behind it is any good, so a low score means an agency publishes little a buyer can verify and nothing more. Two of the lowest scorers publish client names most agencies would envy. The ranking is a worked example of the nine checks below, not the point of the page. All ten agencies were read on their own websites on 23 September 2026, with every figure confirmed against the raw HTML the server returned rather than a rendered reading alone, and every entry links a page it was read from. Nothing here claims a figure is absent from a whole website, because nobody has opened every page of any of these sites, including the person who wrote this. Where an agency was checked and does not publish something, that is recorded as not published on the pages read, which is a different thing from not checked.

2 How these were judged

The six criteria, and what each one is worth

Criterion	What a pass looks like	What a fail looks like	Weight
The core claims are readable in the page's own text	Full marks for scale and credential figures in the text the server returns, on pages a buyer would open, with at least one figure published alongside the base it spans so a ratio can be worked out. High marks for a full stat block in served text even where nothing pairs.	Nothing scores for a figure that exists only in structured data, only in a data attribute behind an animated counter, or only inside an image. Low marks for quantifiers standing in for numbers: hundreds of clients and dozens of developers are not counts, and they are chosen precisely because they cannot be wrong.	25
The same claim holds across the agency's own pages	Full marks where every figure agrees wherever it appears and the stated duration reconciles against the stated start year. This scores the proportion of what an agency publishes that survives cross-checking, not the raw count of contradictions, because publishing twelve figures gives twelve chances to contradict yourself and publishing two gives two.	Heavy marks off where the headline figures disagree: a headcount on one page and a different one on another, a customer count and a brand count that cannot both be right, a duration that does not match the start year beside it, a company age given three different ways on three different pages, only one of which matches the founding year published beside it. Contradictions on a single page cost most, because they mean nobody read the page. Marks come off more lightly, but they do come off, where two figures are both correct and the nouns attached to them are wrong, so that a reader cannot tell which quantity is being counted. The publisher of this page loses marks on both counts.	25
A partner tier stated in its own words and confirmable in the vendor's full register	Two tests applied identically to all ten. Full marks on the first for a Hyva tier stated on the agency's own site that the full register agrees with. Full marks on the second for an Adobe partner level named in current programme wording on pages a buyer would open, uncontradicted elsewhere on the same site.	Partial marks where a tier is held in the register and the site never states it, and less again where the site sells that vendor's work by name while saying nothing about the tier. Nothing scores for a partner claim with no level, a level recorded only as a past event, or a level in a different programme from the one a buyer means. Appearing in Adobe's directory earns nothing for anyone, including the publisher of this page, because it was looked up for one entry only.	20
Case-study evidence names the platform you are buying and publishes a measured outcome	Full marks for a named client, the platform named, and a result with a number on it. Marks for a figure published together with its source and an explicit warning against generalising from one project, which is rarer than it should be and is the best disclosure practice found anywhere in this edition.	Low marks where the published work is measured and named but runs on a different platform, and for a metric attributed to an anonymous third-party reviewer rather than to a named customer. Recognisable client logos with no measured outcome score almost nothing, and that is the most common shape in this field.	15
Commitments that are commitments	Full marks for a support commitment with a target, cover hours and the terms behind it. Nobody in this edition takes them and nobody comes close, which makes this the emptiest line on the page and the easiest place for any agency reading it to take a lead.	An uptime percentage qualified by the word peak, with no cover window, terms or remedy, is a statistic rather than a guarantee. A promise to reply within one business day is a sales commitment. A response time that appears only inside a customer review is not a commitment at all, because the agency never made it. Nothing published scores nothing, which is where most of this field sits, the publisher of this page included.	10
The company is still the company	Full marks where the operating entity, its locations and its identity are stated consistently and currently, and best of all where a registered company name and a full address appear. This is the cheapest check on the list and the one most often skipped.	Marks off where the page's own description of where the company is contradicts itself. No ranked entry scores zero, but only because the five agencies whose domains now redirect to an acquirer were removed from the field rather than ranked. Had they stayed they would have scored zero.	5

3 The ranking

The checks run against ten agencies, September 2026

scandiweb

Buyers who need the largest published body of checkable figures, and who will ask which of its three published ages is the right one76 of 100

scandiweb publishes more checkable figures than anyone here, in the form this page scores highest, which is why it takes 25 of 25 on the first criterion. Its eCommerce services page carries the whole block in one place, in served text rather than markup: 894+ Adobe certifications, 23+ years in eCommerce, 2100+ projects, 600+ certified experts, 700+ brands across 36 countries, an NPS of 95 and more than $4 billion processed for clients a year. The pairing matters more than any single number. Because the certification count and the headcount sit in one sentence, a reader can work out roughly 1.5 certifications per specialist without taking anything on trust, and nobody else here publishes two figures close enough together to compute anything.

It takes full marks on partner tiers too, and it is the only entry that does. The certified Magento developers page states Adobe Commerce Gold Partner and Hyva Platinum Partner, and both survive the check: the Hyva full register carries five scandiweb listings and every one reads Platinum, and Adobe's own Solution Partner Directory records a Gold Partner. That is two vendor registers agreeing with two stated claims, which no other agency here manages once. The Adobe lookup earns nothing on the score, because it was run for scandiweb alone and a check only one entry faced is not a criterion.

Then it scores 12 of 25 on internal consistency, beaten by four of the nine agencies it outranks and tied by a fifth, and it earns the deduction. Two of its published facts are simply wrong somewhere. Its age appears three different ways: 20+ years on its team page and in the profile it wrote for Adobe's directory, 22+ years twice on its Magento to Shopify migration page, and 23+ years on its services page. Only the last agrees with the 2003 founding all of them publish, so three of the four surfaces that state the company's age state it wrongly. And its certified developer page says 18+ years on the platform since 2008 and, further down that same URL, that the team has worked on the platform since 2009, so one of those two sentences is wrong whichever way you read it.

Two further figures are a different and lesser fault, and this page was wrong about them in an earlier edition, so it is worth being exact. scandiweb publishes 36 countries on its services and team pages and 45 countries on its Magento to Shopify migration page. Those are not rival answers to one question: 36 is where its workforce is and 45 is where its clients are, and both are true. The fault is that the migration page says it works across 45 countries, which a reader takes as workforce. The same thing happens with 600+, which is the specialist headcount on four pages and appears on the Adobe Commerce and Magento page as 600+ leading brands, where the client figure is 700+. In both cases the number is right and the noun is wrong. That is a labelling failure, not a contradiction, and it costs less here than a wrong value does.

The distinction matters for how you use any of this. The figures are reconcilable; the published pages are not yet reconciled. A buyer cannot do that reconciling from the pages, because no page says which country count is the workforce and which is the client footprint, and no page says that 600+ counts people rather than brands. Every one of these was found by running check three and check four below against the publisher of this page, which is the argument of the page made concrete. The honest reading is that scandiweb has a great deal of evidence and a marketing site that has outgrown its own proofreading.

It also scores 2 of 10 on commitments. On every page read, scandiweb publishes no first-response target, no resolution target, no cover hours and no support SLA. An NPS of 95 is a satisfaction measure taken afterwards, not a promise about what happens when checkout breaks on a Sunday. The wider credentials on the company profile page are genuinely the largest here, but scale is not accountability and a buyer should ask for the second in writing. Sixteen points separate first from second, and nearly all of that margin comes from two lines, disclosure volume and vendor-confirmed tiers. Weight internal consistency at 40 of the 100 points instead of 25 and the lead falls to under three, which is set out in the methodology rather than buried.

https://scandiweb.com/services

Rocket Web

Buyers who would rather be told four things they can verify than forty they cannot, and who are weighing whether to stay on Adobe Commerce at all60 of 100

Rocket Web publishes fewer numbers than any other agency ranked here and comes second, which is the clearest single argument this page makes. It states no years in business, no headcount, no project count and no certification count. It states four partner memberships, quoted in the evidence table above, and the Hyva one was put through the full register with every other agency here and is exactly right: one listing, tier Silver. That is the only competitor tier claim in this edition a buyer can confirm at the vendor and find correct, and it earns 12 of 20 rather than more only because no Adobe level is claimed at all. Consistency takes 21 of 25, docked because publishing four claims makes consistency cheap and because its founding signal exists only as a copyright range starting in 2010 rather than as a claim anyone is making. Its headline case study is a template and a warning at once: three specific cost figures for a move from Adobe Commerce to Mage-OS, published with their source named and an explicit line saying one project is not a universal estimate. Publishing a number, its source and its limit together is the best disclosure practice anywhere in this edition, and it still takes only 9 of 15, because the client is anonymous and the flagship reference runs off the platform a buyer searching for a Magento agency is buying. The site is also unusually direct about its own credentials changing, writing that the badges changed and the standard did not, and that it took a badge down. It commits to fixed scope and fixed price with plain-language tradeoffs before work starts, which is an undertaking rather than a metric and scores 2 of 10. No response or resolution target was found on the pages read.

https://www.rocketweb.com/

Divante

Buyers running a composable or headless programme, who should confirm the Magento part specifically because the published work does not evidence it57 of 100

Divante publishes the most internally consistent set of figures in this edition and takes the highest score on that criterion, 23 of 25. Its stat block reads 18+ years in eCommerce, 250+ projects delivered, 1000+ specialists worldwide and 8+ platforms supported, and its founding statement of 2008 reconciles with the duration exactly, which sounds like a low bar until you read the rest of this page and find four agencies that cannot manage it. The figures repeat identically wherever they appear. It also publishes €2B+ GMV managed, 10k+ open-source stars and a 99.9% peak uptime SLA, and states it is based in Wroclaw, Poland. Then it scores zero on partner tiers, and that is not an oversight: no Adobe partner level appears on the pages read, no certification count beyond a claim of certified experts across all major platforms, and it does not appear in the Hyva register at all. There is nothing to check, so there is nothing to score. The sharper finding is on case studies, where it takes 3 of 15 despite publishing a measured result for every one. The four flagship studies, quoted in the evidence table above, cover a conversion uplift on a 2M+ SKU catalogue, a loyalty platform on Oracle Cloud and Kubernetes, an employee self-service portal for 12,000 staff, and an SAP-integrated warehouse picking solution. Every one has a number. Not one names Magento or Adobe Commerce, and one is not a storefront at all. Its own named products are Alokai, formerly Vue Storefront, plus Open Loyalty and Spartacus, the last a storefront for SAP Commerce Cloud. A buyer taking these as evidence of Magento capability would be reading them wrong, and the only way to find out is to open them. On commitments it takes the highest score anyone takes, 4 of 10, and it is still worth reading sceptically.

https://divante.com/

Tigren

Smaller budgets wanting a complete published stat block, who will ask which of two published durations is the right one40 of 100

Tigren publishes a complete stat block in served text and then contradicts part of it one click away. The homepage carries 13+ years since 2012, 50+ in-house experts, 500+ stores delivered and 200+ global clients across 20+ countries, which is more structured disclosure than five agencies above or below it and earns 18 of 25. Two things then go wrong. The arithmetic does not close, because 2012 to 2026 is 14 years and not the 13+ printed beside it. And the about page says over 11 years for the same thing, so the two pages a buyer reads first disagree, which takes it to 12 of 25 for consistency. Both contradictions concern the same fact, its age, which is why it scores above the agencies whose headline figures disagree on several fronts. The Hyva finding is sharper. Tigren markets Hyva development as one of four headline services and states no Hyva partner tier on the pages read. The full register lists it at Bronze, a real tier the curated preferred-partner page does not show at all, so an agency in exactly Tigren's position is invisible on the page most buyers would check and present on the one they would not. It scores 3 of 20: it holds a tier, it sells the work, and it tells nobody. Its case studies name ShopEddies, Tiny Blessings and Wine-now, the last with a catalogue size rather than a result, so it takes 4 of 15. One oddity is worth flagging because it is the kind of thing this page exists to notice: the trust row prints 17 reviews and a 4.7 star DesignRush rating and, directly beside them, 1000+ reviews, with no source and nothing explaining how the two relate.

https://www.tigren.com/

Webkul

Buyers who want an extension and marketplace specialist, and who will pin down which of its two headcounts and two durations apply38 of 100

Webkul publishes more distinct figures than almost anyone here and contradicts itself on both of the ones that matter, which is why it takes 19 of 25 for readability and 6 of 25 for consistency, the lowest consistency score in this edition. The homepage says 16+ years and a global team of 500+ people. The about page says 15 years, a 2010 founding and a team of 650+ members, all quoted in the evidence table above. So the two pages a buyer reads first disagree by a year on age and by 150 people on size, and the about page's own arithmetic fails against its own founding year, since 2010 to 2026 is 16. The about page also claims more than 150,000 customers. On partner tiers it scores 7 of 20 and the reason is instructive: its homepage row reads Adobe Silver Technology partner, alongside Odoo, Shopify and Google Cloud credentials. That is a real credential and it is not the Adobe Solution partner tier a buyer comparing implementation agencies is looking at, so it takes 4 of the 8 Adobe points rather than none and rather than full marks. Meanwhile Webkul markets Hyva theme development in its main services menu on both pages and states no Hyva tier; the full register lists it at Partner, the base tier, which takes 3 of the 12 Hyva points. No certification count appears on the pages read. Its commitments score 1 of 10, and that single point is the clearest example on this page of the difference between a claim and a commitment: the much-quoted sentence about an average response time under one business day is real, is in served text, and is not Webkul's promise, because it sits inside a customer review. A buyer quoting it back in a contract negotiation would be quoting a stranger.

https://webkul.com/about-us/

Meetanshi

Buyers wanting Magento extensions and smaller service engagements, who will ask which of two customer figures on one page is the real one37 of 100

Meetanshi publishes its headline contradiction within a few lines of itself, which makes it the cleanest demonstration on this page of why a buyer should read a stat row twice. Its homepage trust block claims 18,000 happy customers, and immediately below, on the same page, it says it is trusted by 13,000+ brands worldwide. Both are quoted in the evidence table above. The two are not obviously the same population, the page never says whether they are, and it defines neither, which costs it 8 of 25 on consistency. Its years figure has the opposite problem: 15+ years is published with no founding year on the pages read, so unlike GPMD, Tigren and Webkul it cannot be caught contradicting itself, because there is nothing to check it against. That is not a strength and this page does not score it as one. On partner tiers it takes 8 of 20, entirely from the Hyva line and entirely without Meetanshi's help. The full register lists Meetanshi Technologies at Silver, a genuine and respectable tier, and that tier does not appear on the pages read. Meetanshi sells Hyva theme development as a named service and lists 130+ Hyva-compatible extensions, so this is not an agency with no Hyva relationship to mention. It simply does not mention it, and it is one of four here holding a tier it never states. On Adobe it publishes no level, referring to Adobe-certified experts and certified developers without a count, a level or a name. No case study metrics, headcount or support commitment were found on the pages read.

https://meetanshi.com/

Ranosys

Buyers who will take a client roster on trust, or who are prepared to ask for every number in writing before the first meeting33 of 100

Ranosys has the most impressive client roster in this edition and the fewest checkable numbers, and putting those two facts next to each other is most of what this page is for. Its homepage names eight recognisable brands as case studies, listed in the evidence table above, and not one carries a quantified result. There is no conversion figure, no revenue figure, no performance figure and no timeline against any of them, so a reader finishes the page knowing who Ranosys has worked with and nothing about what changed, which is why it takes 2 of 15 on evidence. The scarcity runs further than case studies. Across the homepage's served text there is no founding year, no duration, no headcount, no project count, no client count and no certification count, which is the lowest readability score on this page at 4 of 25. Its partner claim is a single sentence naming the Adobe Solution Partner programme with no level attached, so it takes 2 of the 8 Adobe points. The Hyva register lists a Ranosys entry filed under Singapore at Partner tier, which does not appear on the pages read, worth 4 of 12. It scores 18 of 25 on consistency and that number needs reading carefully rather than admiring: nothing contradicts anything because there is almost nothing to contradict. Consistency is trivially easy when you publish no figures, and this page's second criterion cannot distinguish a disciplined publisher from a silent one, which is a genuine limitation of the model and is stated here rather than buried. A buyer shortlisting Ranosys is not looking at evidence. They are looking at logos, and logos are the one thing on an agency website that is never accompanied by a number.

https://www.ranosys.com/

DCKAP

Distributors and manufacturers wanting ERP integration first and a storefront second, who should ask what the current Adobe partner status is30 of 100

DCKAP is the clearest case in this edition of an agency that answers every quantitative question with a word rather than a number. It is trusted by hundreds of distributors and manufacturers, and by 2015 it had dozens of certified developers, both quoted in the evidence table above. Hundreds and dozens are not counts, and they have the property that they can never be wrong, which is why this page scores them at 6 of 25 rather than treating them as disclosure. That partner sentence deserves a closer look because it fails three ways at once. It is in the past tense and dated to 2015, so it tells a buyer what was true eleven years ago and nothing about today. It names a programme, Magento Partner, that no longer exists under that name. And it attaches no level, so even read at face value it does not say whether the status was Bronze, Silver or Gold. That is 2 of 20. No current Adobe partner level, founding year, headcount or certification count was found on the pages read, and DCKAP does not appear in the Hyva register at all. It takes 16 of 25 for consistency on the same basis as Ranosys, by publishing too little to contradict, docked for leaving a dated claim uncorrected for over a decade. Its one published metric is instructive rather than useful: a $250k annual saving attributed not to any of the named customers on the page, who include Tangent Materials, JSI Cabinetry and Flow Control Group, but to an anonymous reviewer on a third-party site. So the named clients have no numbers and the number has no named client. Note also the positioning: DCKAP presents itself as platform-agnostic ERP integration and lists Adobe Commerce as one supported platform among many.

https://www.dckap.com/about-us/

Overdose

Australian and New Zealand programmes wanting strategy and marketing alongside build, who should confirm which offices actually exist26 of 100

Overdose publishes the single most checkable contradiction on this page, and it takes about fifteen seconds to find. Its about page says in prose that it has offices and clients in seven named cities, and further down that same page the block headed Our Offices lists four, with street addresses. The page footer lists the same four. Both are quoted in the evidence table above. Singapore, Seattle and Kyiv appear in the sentence and in no list on the page. Read the sentence closely and it may be doing something subtler than overclaiming, since it says offices and clients without saying which cities are which, but a buyer scanning it comes away believing in seven offices while the page's own office list supports four. That is 9 of 25 on consistency and 1 of 5 on entity clarity, the lowest on this page. Everything else is thin rather than wrong. Its incorporation year of 2016 and its team of 180 commerce professionals are published in the same sentence and agree, which is more than four agencies above it manage. There is no project count, no client count and no certification count on the pages read. Its partner claim names Shopify Plus, Adobe Commerce, Commercetools and BigCommerce and a level for none of them, which takes 2 of 20. It names Cargo Crew, Barkers, Metagenics and Dusk as case studies and shows a wall of more than twenty further client logos, publishing no metric against any of them, which is 2 of 15. No response target, SLA or support terms were found.

https://overdose.digital/about-us

GPMD

London buyers who want a long-established independent and are willing to source every figure themselves22 of 100

GPMD scores lowest here and does so almost entirely by publishing very little, not by publishing anything dishonest. Its about page claims 25 years of experience and, a little further on, says it started building ecommerce websites in 1999. Both sentences are on the same page, both are quoted in the evidence table above, and they cannot both be right, because 1999 to 2026 is 27 years. This is the smallest contradiction in this edition and in some ways the most useful to show a buyer, because it is almost certainly a number written once and never updated rather than anything intended, and because it takes ten seconds and no tools to catch. It is also a reminder to run the check in the direction that helps you: an agency that understates its own age by two years is telling you its about page has not been read in a while, which is worth knowing before you ask that page's claims to carry weight. Beyond those two sentences the disclosure runs out fast. No headcount, project count, client count, certification count or partner tier of any kind was found on the pages read. GPMD lists partnerships with Shopify, n8n, Algolia, Klaviyo, Klevu, Nosto, Linnworks and BigCommerce and attaches a level to none of them, makes no Adobe or Magento partner claim on those pages, and does not appear in the Hyva register, which is a clean zero on this page's twenty-point tier criterion. Its case studies name MUJI, The Fold Line, ARAM, Lazy Susan, Japan Centre and The River Cafe Shop, with no metric against any of them. Where it does score, and where it beats every other agency here including the publisher of this page, is on being a clearly identified legal entity: it publishes a registered company name and a full London address, taking the full 5 points on the last criterion. Given that five agencies scouted for this edition turned out to have been absorbed into somebody else, knowing exactly which company you are contracting with is not trivial.

https://www.gpmd.co.uk/about-us

4 Which one fits

Pick by situation, not by ranking

If this is you	Shortlist	Why
You have three proposals and no idea how to tell them apart	Run the nine checks below before you read another proposal	Take each agency's homepage, about page and the service page matching your project, and compare the same claim in all three. Do the arithmetic on any duration against any founding year. Look up every stated partner tier in the vendor's full register rather than its curated page. Open one case study and check which platform it names. Five agencies here fail the cross-page check and three fail the arithmetic, and every failure is visible in under a minute with nothing but a browser.
Your procurement process needs verifiable credentials before anyone is shortlisted	scandiweb, then Rocket Web	These are the only two entries whose stated partner tier matches the vendor's own full register. scandiweb states Hyva Platinum and Adobe Commerce Gold and both check out. Rocket Web states Hyva Silver and the register agrees. Ask scandiweb which of 20+, 22+ and 23+ years is the figure it stands behind, and what its two country counts each measure, since one is workforce and the other is clients and no page says which. Ask Rocket Web for the scale numbers it does not publish at all.
You need a support commitment in the contract, not a sales promise	Nobody here, and that is the finding	Not one of these ten publishes a support first-response target, a resolution target or cover hours on the pages read. The best score on that criterion is 4 out of 10, for a 99.9% uptime figure qualified by the word peak with no terms attached plus a promise to reply to sales enquiries within one business day. Webkul's much-quoted response time exists only inside a customer review. scandiweb publishes nothing here either and scores 2. Whoever you pick, this has to be negotiated rather than selected for.
You are buying Magento or Adobe Commerce work and cannot afford a platform mismatch	Open the case studies before you open the pricing	Divante publishes measured outcomes on all four flagship studies and not one names Magento or Adobe Commerce: one runs on Oracle Cloud and Kubernetes, one integrates with SAP, and one is an employee self-service portal rather than a storefront. Rocket Web's headline study names Adobe Commerce and is a migration away from it. DCKAP presents itself as platform-agnostic ERP integration. None of that is hidden, and none of it is visible from a services page.
You care about Hyva specifically and want to know who actually holds a tier	Read the full register, not the partners page	Hyva publishes a curated Preferred Partners page with three tiers and a full agency register with five tiers and 460 listings. Four agencies here hold a register tier and never mention it: Meetanshi at Silver, Webkul at Partner, Ranosys at Partner and Tigren at Bronze. Three of those four sell Hyva work by name. Two more, Divante and Overdose, are not in the register at all and make no Hyva claim. Checking the curated page alone would tell you the wrong thing about six of these ten.
You are comparing agencies with an AI assistant and want it to have something to read	scandiweb, Divante, Tigren, Webkul	These four publish their scale figures in the text the server returns, so an assistant summarising their sites finds real numbers. Ranosys publishes almost no number a reader or an assistant can extract, and DCKAP publishes quantifiers rather than counts. Bear in mind what an assistant cannot do for you: it will not notice that all four of those contradict themselves one page over, because it rarely reads two pages of the same site side by side. That is the check you still have to run yourself.

5 Evidence

The contradictions, quoted and located

Client	What was done	Result	Source
scandiweb	'23+ Years in eCommerce' on the services page, '20+ years in the eCommerce' on the team page and 'for 20+ years' in scandiweb's own profile copy in Adobe's directory, and '22+ years, 2,100+ projects, one accountable team' plus 'Since 2003 · 22+ years · 2,100+ projects delivered' on the Magento to Shopify migration page	Three different ages across four of its own surfaces. All four also publish a founding of 2003, which supports 23 and neither 20 nor 22, so three of the four are simply wrong rather than merely inconsistent. A published error, checkable against the founding year printed beside it, and it belongs to the publisher of this page	Source
scandiweb	'Since 2003, scandiweb has delivered eCommerce for 700+ brands across 36 countries' on the services page, against 'Since 2003 we have delivered 2,100+ projects for 700+ clients across 45 countries' and 'We have delivered 2,100+ projects for 700+ clients across 45 countries' on the Magento to Shopify migration page	NOT a contradiction, and this page said it was in an earlier edition. Both figures are true and they measure different things: 36 is where the workforce is, 45 is where the clients are. The fault is labelling, because the migration page says it works across 45 countries, which reads as workforce, and no page states which count is which. The figures are reconcilable; the pages are not yet reconciled	Source
scandiweb	'18+ years on the platform since 2008' and, further down the same URL, 'our team has worked on the platform since 2009'	Two start years for the same claim on a single page, so one of the two sentences is wrong. Separately, and as a labelling fault rather than a contradiction, the Adobe Commerce page writes 'Trusted by 600+ leading brands worldwide' where 600+ is the specialist headcount published on four other pages and the client figure is 700+. The number is right and the noun is wrong	Source
Rocket Web	'Rocket Web is an active Mage-OS Association partner, a Hyva Silver Partner, a Shopware Gold Partner and a Bureau of Digital member.'	The one competitor tier claim in this edition checked at the vendor and found correct: one listing in Hyva's full register, tier Silver. The Mage-OS, Shopware and Bureau of Digital claims were not checked for anyone and are not scored. No Adobe level is claimed	Source
Rocket Web	'Under $40,000 TO SWITCH TO MAGE-OS', 'About $22,000 ANNUAL MANAGED HOSTING, DOWN FROM $57,000', followed by 'Source: internal project records. One project is not a universal estimate. It is enough to make the question worth asking.'	The best disclosure practice in this edition: the figure, its source and an explicit limit on generalising from it, in one block. The client is anonymous, described only as a large U.S. company, and the project is a migration off Adobe Commerce	Source
Divante	Four flagship case studies with published metrics: '3x conversion uplift · 2M+ SKU catalogue', '300K+ daily transactions · Oracle Cloud', '12,000+ daily active users · 50 countries', '4.7M print-outs eliminated · SAP-integrated warehouse'	Every one measured, not one naming Magento or Adobe Commerce. One runs on Oracle Cloud and Kubernetes, one integrates with SAP, and the 12,000 daily active users belong to an employee self-service portal rather than a storefront	Source
Divante	'99.9% PEAK UPTIME SLA' in a contact block, alongside 'we'll get back within one business day'	The highest commitment score on this page at 4 of 10, and still not a support commitment. The word peak removes the guarantee, no cover window, terms or remedy are published, and the one business day promise is a sales reply	Source
Webkul	Homepage: '16+ years' and 'Global team 500+ people.' About page: 'in the last 15 years', 'founded in the year 2010 and for 15+ years', 'team of 650+ members'	The two pages a buyer reads first disagree by a year on age and by 150 people on size. The about page's own arithmetic also fails against its own founding year, since 2010 to 2026 is 16	Source
Webkul	'The average response time is less than 1 business day.' In full: 'support was very helpful and they resolve all my different queries in a few days. The average response time is less than 1 business day. The item is really good itself. I recommend it.'	Live text on the homepage, and not a Webkul commitment: it is a sentence inside a customer review. Separately, the homepage partner row reads 'Adobe Silver Technology partner', a different Adobe programme from the Solution partner tier a buyer means	Source
Meetanshi	'15+ Years Experience', '240+ Extensions & Apps', '18k+ Happy Customers', '4.5 Trustpilot', then a few lines below on the same page, 'Trusted by 13,000+ brands worldwide'	Two customer populations published within a few lines of each other, neither defined and never reconciled. The years figure carries no founding year on the pages read, so it cannot be checked against anything. The Hyva register lists Meetanshi Technologies at Silver and the pages read do not mention it	Source
Tigren	Homepage: '13 + Years in eCommerce' with 'Building successful selling stores since 2012', '50 + Talented Experts', '500 + Stores Delivered', '200 + Global Clients'. About page: 'With over 11 years of experience in Magento development'	2012 to 2026 is 14, not 13, and the about page says over 11 for the same thing. The homepage trust row also prints '17 reviews' with a 4.7 DesignRush rating and, beside it, '1000+ reviews' with no source. Sells Hyva development by name and does not state the Bronze tier the register lists	Source
Overdose	'We now have offices & clients in Sydney, Melbourne, Brisbane, Auckland, Singapore, Seattle & Kyiv.' The 'Our Offices.' block on the same page lists four with addresses: Melbourne, Sydney, Brisbane, Auckland	Seven cities in prose, four in the office list, and the page footer lists the same four. The sentence says offices and clients without saying which is which, so a scanning reader takes seven and the page evidences four	Source
GPMD	'25 years of e-commerce experience' and 'We started building ecommerce websites in 1999, so we have quite literally seen every step in the evolution of online retail.'	Both on the same page, and 1999 to 2026 is 27. The smallest contradiction in this edition and the one requiring no tools at all to catch. GPMD does publish a registered entity and full address, which most of this field does not	Source
Ranosys	'Award-winning Solution Partner with expertise in Adobe Commerce, AEM, Marketo & Analytics implementation.' Case studies naming Skechers, Under Armour, Specialized Rutland Cycling, Smythson, Shell, Rituals, Charles & Keith and Panasonic	The right programme named with no level attached, and the strongest client roster in this edition with no quantified result against any of those brands. No founding year, duration, headcount, project count or certification count in the homepage's served text	Source
DCKAP	'Trusted by hundreds of mid-market distributors and manufacturers' and 'By 2015, we had become an official Magento Partner, with dozens of certified developers and experts on hand'	Quantifiers rather than counts. The partner sentence fails three ways at once: past tense dated to 2015, a programme that no longer exists under that name, and no level attached. Its one metric, 'reduced $250k in annual spend so far', is attributed to an anonymous Gartner reviewer rather than any named customer	Source
Hyva, the vendor register	hyva.io/find-an-agency: 460 listings across five tiers, PLATINUM 17, GOLD 46, SILVER 102, BRONZE 264, PARTNER 31. hyva.io/partners: three tier headings only, Platinum, Gold and Silver	The curated page omits 295 agencies at Bronze and Partner entirely. Four of the six register-listed agencies on this page never state the tier they hold. Read live on 23 September 2026, scrolled to full height twice, counted by DOM elements and cross-checked against agency anchor count	Source
Five agencies not ranked here	trellis.co now redirects to zaelab.com, redstage.com to fulcrumdigital.com, guidance.com to onemagnify.com, statementagency.com to idhlagency.com, mediotype.com to blueacornici.com	Five live Magento agency domains opened while researching this edition, all 301 redirecting to an acquirer. None is ranked, because there is no longer a company at that name to rank. Verified by following each redirect on 23 September 2026	Source

6 In detail

The buyer's checklist: nine checks, in the order to run them

Each of the nine sections below is one check. Every one names the failure it catches, the check that catches it, and at least one agency on this page where it fired. They are in the order worth running them, cheapest and most decisive first, and none needs anything but a browser. Together they take about ten minutes per agency.

They work because a proposal is written for you and a website is written for everyone, which makes the website far easier to catch out. Nothing below asks an agency to prove anything. It asks only whether what an agency already says about itself holds together when read twice. The checks were run against the publisher of this page on the same terms as everyone else, and two of them fired.

Check 1

Check whether the company is still the company

Type the agency's domain into a browser and watch where the address bar settles. This is the cheapest check on the list and the one most often skipped, because a shortlist assembled six months ago feels current and a ranked-list article from two years ago looks authoritative.

Five of the agencies scouted for this edition, all well-known Magento names with years of published work behind them, now redirect somewhere else entirely: trellis.co lands on zaelab.com, redstage.com on fulcrumdigital.com, guidance.com on onemagnify.com, statementagency.com on idhlagency.com, and mediotype.com on blueacornici.com. None of those redirects is hidden and none is dishonest. But a buyer working from an old shortlist will be emailing a company that has been absorbed into somebody else. The people may be the same. The contract, the escalation path and the day rate almost certainly are not.

The check takes five seconds and it removed five candidates from this page before any scoring started. The positive version is worth as much: GPMD scores lowest on this page overall and takes full marks here, because it publishes a registered company name and a full address, which most of this field does not.

Check 2

Read the page source, not the rendered number

A figure that lives only in structured data, or only in a data attribute behind an animated counter, is invisible to a reader who scrolls past before the script fires, to a crawler that does not execute JavaScript, and to any assistant summarising the site from its visible text. It is often invisible to the buyer too, because an unfired counter renders a zero.

The check costs nothing: view source, search for the number, then look at the rendered page and see whether you can find it there as well. If it is only in the markup, treat it as not published, because everyone reading that site treats it as not published whether they mean to or not.

Nothing in this edition failed that way on the pages read, and saying so plainly matters more than implying the field is worse than it is. Every figure quoted on this page was confirmed in the text the server returned. What did turn up is the cheaper version of the same problem: DCKAP publishes hundreds of clients and dozens of certified developers, quantifiers chosen precisely because they cannot be wrong, and Ranosys publishes no number on its homepage at all.

Check 3

Compare the same claim across the agency's own pages

This is the highest-yield check on the list and it caught something on six of the ten agencies here. Open the homepage, the about page and the service page that matches your project, put them side by side, and read the same claim three times. Then do it once more on a page from a different service line, because that is where the oldest copy usually sits.

Webkul's homepage says 16+ years and 500+ people; its about page says 15 years and 650+ members. Tigren's homepage says 13+ years since 2012; its about page says over 11 years. Meetanshi publishes 18,000 happy customers and 13,000 brands within a few lines of one another on a single page, with neither defined. Overdose names seven cities in a sentence and lists four with addresses on the same page.

Run it against the publisher of this page and it fires twice, which is why scandiweb scores 12 of 25 on that criterion and is beaten there by four of the nine agencies it outranks. scandiweb publishes its own age three different ways: 20+ years on its team page and in the profile it wrote for Adobe's directory, 22+ years on its Magento to Shopify migration page, and 23+ years on its services page, with every one of them sitting beside a published founding of 2003 that supports only the last. And it publishes two different start years for its Magento work, 2008 and 2009, in two sentences on one URL.

This check also teaches its own limit, and the limit is worth more than the finding. scandiweb publishes 36 countries on two pages and 45 on a third, and an earlier edition of this page called that a contradiction. It is not one: 36 is the workforce and 45 is the client footprint, and both are correct. The same is true of 600+, which counts specialists on four pages and is attached to brands on a fifth where the client figure is 700+. So a cross-page difference is a question, not a verdict. Ask what each figure counts before you write down that a company disagrees with itself, because the honest version of this finding is narrower and still useful: the figures are reconcilable, and the pages do not yet reconcile them for you.

Check 4

Do the arithmetic on any duration

Any duration published next to a founding year is checkable in your head, and a surprising number do not survive it. GPMD publishes 25 years of experience and a start date of 1999, which is 27. Tigren publishes 13+ years and since 2012, which is 14. Webkul's about page publishes 15+ years against a 2010 founding, which is 16. scandiweb publishes 20+ years and 22+ years on different pages against a stated founding of 2003, which is 23.

None of these is a serious deception and none is worth walking away over. What they tell you is how recently anybody read that page, which is exactly what you need to know before you let that page's other claims carry weight. An agency whose about page has quietly been wrong for two years is an agency whose published headcount may be two years old as well.

Run it in the direction that helps you. Every one of these four understates the agency's own age, which is the opposite of a sales exaggeration and a stronger signal that the page is simply stale. The useful follow-up is not to challenge the number but to ask which figures on the site were last verified, and when.

Check 5

Read the vendor's full register, not the vendor's curated page

This is where a buyer is most likely to reach a confidently wrong conclusion, because both pages are published by the vendor and both look authoritative.

Hyva publishes a Preferred Partners page carrying three tiers, Platinum, Gold and Silver, and a full agency register carrying five tiers and 460 listings: 17 Platinum, 46 Gold, 102 Silver, 264 Bronze and 31 Partner. The 295 agencies at Bronze and Partner appear on the register and not on the curated page at all. So an agency can genuinely hold a Hyva tier, sell Hyva work, and be entirely absent from the page a buyer would naturally check, which describes Tigren, Webkul and Ranosys here exactly. Checking the curated page alone would have produced the wrong answer about six of these ten agencies.

The corollary matters as much as the finding: absence from a curated directory is not evidence of anything, and it should never be scored as though it were. The same rule governs Adobe's Solution Partner Directory on this page, which was looked up for one entry only and therefore earns no points for anyone, including the publisher.

Check 6

Check whether a tier held is a tier told

This is the finding that surprised this edition most. Six of the ten agencies here appear in Hyva's full register. Four of them do not mention it anywhere on the pages read: Meetanshi at Silver, Webkul at Partner, Ranosys at Partner and Tigren at Bronze. Three of those four actively sell Hyva development as a named service on the same sites that say nothing about their tier.

That cuts both ways for a buyer and both ways are useful. A shortlist built purely from agency self-description will understate the field, so the register is worth reading in its own right rather than as a way of confirming what an agency already told you. And an agency that does state its tier, and states it correctly, has given you something most of its competitors have not.

Only two entries here do. scandiweb states Hyva Platinum and holds five Platinum listings. Rocket Web states Hyva Silver and holds one Silver listing. Everything else is either silent about a tier it holds, or correctly silent because it holds none.

Check 7

Open the case study and check which platform it names

This check is skipped most often because a case study with a number on it looks like evidence, and it usually is evidence of something other than what the buyer assumes.

Divante publishes four flagship case studies with a measured outcome on every one, and not a single one names Magento or Adobe Commerce: one is a loyalty engine on Oracle Cloud and Kubernetes, one integrates with SAP for warehouse order picking, one is an employee self-service portal for 12,000 staff, and the fourth names no platform at all. Those are real projects with real results and they are not evidence of Magento capability.

Check the direction as well as the platform. Rocket Web's headline study does name Adobe Commerce, and it is a migration away from it to Mage-OS, which is a fact worth noticing rather than a criticism. And check who the number belongs to: DCKAP's one published metric is attributed to an anonymous reviewer on a third-party site rather than to any of the named customers on its own page. None of this is concealed. All of it requires clicking through, which is the entire point.

Check 8

Work out whether a commitment is a commitment

This is the emptiest criterion on this page: ten points are available and the highest score anyone takes is four. Three patterns account for almost all of it.

A percentage qualified into meaninglessness, such as Divante's 99.9% uptime figure carrying the word peak, published in a contact block with no cover window, no terms and no remedy, which is a statistic and not a guarantee. A sales promise dressed as a service level, such as a commitment to reply to an enquiry within one business day, which tells you how fast somebody will try to sell to you and nothing about what happens when checkout breaks at 2am. And a response time the agency never actually promised, such as the sentence about a one business day average response that sits inside a customer review on Webkul's homepage and gets quoted as though Webkul wrote it.

One question separates all three from a real commitment: what are the cover hours, what is the target, and what happens if it is missed. Not one agency on this page answers it, and neither does the publisher of this page, which publishes an NPS of 95 and no response or resolution target on any page read, scoring 2 of 10 as a result.

Check 9

Notice what a logo wall is not

Ranosys names Skechers, Under Armour, Smythson, Shell, Rituals and Panasonic on its homepage and publishes no quantified result against any of them. Overdose names four clients and shows more than twenty further logos with no metric anywhere. GPMD names MUJI, ARAM and Japan Centre with nothing attached.

Recognisable client names are the most persuasive thing on an agency website and the only element that is almost never accompanied by a number, which is not a coincidence. The useful version of this check is to count: how many named clients, how many with a published outcome, and how many of those outcomes are on the platform you are buying.

Across all ten agencies here the ratio is poor enough that nobody scored more than 12 of 15 on evidence. The best performer was the agency that published three figures about one anonymous client and told you not to generalise from them, which is worth more than a wall of logos and takes up far less room.

7 Methodology

How this was put together

This page scores disclosure, not delivery quality. Every criterion asks whether something an agency publishes can be checked, and none measures code quality, delivery, how a team behaves when a launch slips, or whether clients renew. A low score means thin or contradictory public disclosure and nothing more. Two of the three lowest scorers publish client rosters most agencies would envy, and the second-placed agency publishes fewer numbers than anyone in the field. Every entry names the criteria it scored badly on, so the table works better as a question list for a first call than as a verdict.

Ten agencies were scored out of 100 against six weighted criteria published above: core claims readable in the page's own text at 25, the same claim holding across the agency's own pages at 25, a partner tier stated in its own words and confirmable in the vendor's full register at 20, case-study evidence naming the platform and publishing a measured outcome at 15, commitments that are commitments at 10, and the company still being the company at 5. The ranking is the score order with no adjustment, every score is printed against its entry, and the full ladder for each criterion is written into the criteria table rather than summarised. The same weighting is repeated as machine-readable structured data so the model can be re-run rather than taken on trust.

All ten were read on their own websites on 23 September 2026, and every entry links a page it was read from. Every figure was checked against the raw HTML the server returned as well as against a rendered reading, which is how same-page contradictions and markup-only values get separated from published text; where a site blocked a plain fetch, it was read in a browser instead. Nothing came from a directory listing, a review profile or a press release except the one register check described next. Where a fact was not found, this page says it was not found on the pages read rather than claiming it is absent from the whole site. That distinction is not pedantry: during research the publisher of this page was itself the subject of a wrong absence claim, when a figure reported as appearing nowhere turned out to appear twice on a service page nobody had opened. It is now one of the published errors in its own entry. A second correction went the other way: two differences this page first called contradictions turned out to be correct figures carrying the wrong nouns, and the weaker and accurate version is what it now publishes.

The one external check applied to everybody is the Hyva agency register. Hyva publishes two directories and only one is complete: the curated Preferred Partners page carries three tiers, while the full register carries five tiers and 460 listings, made up of 17 Platinum, 46 Gold, 102 Silver, 264 Bronze and 31 Partner. The register was read in a browser, scrolled to full height until the document height stopped changing, counted by DOM elements whose text is exactly a tier word rather than by text adjacency, and cross-checked against the count of agency anchors, which agreed within two. It was read twice on separate loads with identical results. Each of the ten agencies was then swept individually by locating its own anchor and walking up to the nearest ancestor containing exactly one tier badge. Adobe's Solution Partner Directory was looked up for one entry only and therefore earns no points for anyone on any criterion, including for the publisher of this page.

Three agencies were excluded because they could not be read rather than because they publish nothing: one serves a security block to both a plain fetch and a browser, one has an expired certificate on its own domain, and one returns an HTTP 403. Scoring any of them would have meant recording not checked as not published. Five more were excluded because their domains now redirect to an acquirer, which is reported as a finding about the lane rather than a claim about any agency. Six further agencies were scouted and set aside because the published Magento evidence was too thin to score fairly.

Every scandiweb figure is scandiweb's own published statement, read live on 23 September 2026 and checked against the served HTML exactly as everyone else's was. It takes 12 of 25 on internal consistency, a score four of the nine agencies below it beat outright and a fifth ties, and the deduction has two parts that this page keeps separate because an earlier edition did not. Two published facts are wrong somewhere. Its age appears as 20+, 22+ and 23+ years across four of its own surfaces, all of which also publish a founding of 2003, so three of the four are wrong and not merely different. And one URL gives both 2008 and 2009 as the start of its Magento work, so one of those sentences is wrong. Two further figures are a lesser fault and an earlier edition of this page got them wrong: 36 countries and 45 countries are both true and measure different things, the workforce and the client footprint, and 600+ is the specialist headcount that appears on one page attached to brands where the client figure is 700+. Those are labelling failures, where the value is correct and the noun is not, and they cost less here than a wrong value. The distinction is worth stating plainly because it cuts against this page's own first instinct: the figures are reconcilable, and the published pages are not yet reconciled. Knowing the right answer privately does not fix a website, and this criterion scores websites. scandiweb also takes 2 of 10 on commitments for publishing no support target on any page read. It leads on the other four criteria and finishes 16 points clear.

The weighting was stress-tested rather than settled on, and the arithmetic is published so it can be argued with. Three alternative weightings were run, fixed before the scores were finalised rather than chosen afterwards, each keeping a checkability criterion heaviest and giving every criterion a floor, with each cell rescaled proportionally. Moving internal consistency from 25 points to 30 leaves scandiweb first on 73.4 to Rocket Web's 61.8. Moving it to 35 gives 70.8 against 63.6. Moving it to 40, the most hostile weighting that could still be defended, gives 68.2 against 65.4, a lead of under three points. The order published here holds under all three. One earlier state of this page is worth recording, because it shows the model is not being steered: on an interim reading of scandiweb's consistency score, that last weighting put Rocket Web first and this page said so in these words. When the underlying figures were resolved and the score rose, the reversal disappeared and the concession came out. It came out because it stopped being true, not because it was unwelcome.

8 Questions

Common questions

How do I compare Magento agencies when every proposal looks the same?

Stop comparing the proposals and start checking the websites behind them, because a proposal is written for you and a website is written for everyone, which makes it far easier to catch. Run the nine checks on this page in order: does the domain still belong to that company, is the number in the page source as well as on the screen, does the same claim hold across their own pages, does the arithmetic close, does the vendor's full register agree, do they state a tier they hold, does the case study name your platform, is the commitment a commitment, and does the logo wall carry any numbers. On the ten agencies scored here those checks found something on nine of them, in about ten minutes per agency.

What is the fastest useful check if I only have five minutes?

Two things. First, type each agency's domain into a browser and see whether it redirects to somebody else, which takes five seconds per agency and removed five candidates from this page before scoring started. Second, open the homepage and the about page side by side and compare the headline numbers, which caught contradictions on six of the ten agencies ranked here, including the publisher of this page. Those two checks cost almost nothing and are the two most likely to change your shortlist. Everything else here is worth doing once the list is down to three.

Does a low score on this page mean an agency does bad work?

No, and the gap is wider here than on most rankings. This page scores disclosure, not delivery quality. Every criterion measures whether a claim can be checked, and none touches code quality, delivery, or what happens when a project goes wrong. The agency with the most recognisable client roster in this edition scores 33 because it publishes no numbers, and the agency in second place publishes fewer numbers than anyone here. What a low score reliably tells you is how much work you will have to do yourself before you can compare that agency with another, and which questions to ask on the first call.

Should I rule out an agency that publishes almost nothing?

No. Rule it in more slowly. Rocket Web publishes no years in business, no headcount, no project count and no certification count, and it comes second here, because the little it publishes is precise, sourced and correct. Ranosys publishes almost nothing and comes seventh, because what it does publish is a client roster with no outcomes attached. The difference is not volume but whether anything offered can be verified. An agency that publishes nothing is simply an agency you will have to interview rather than research, and that is a cost rather than a disqualification.

Why does publishing less sometimes score higher on this page?

Because two criteria reward different things and one of them can be gamed by silence. Readability rewards publishing checkable figures, so silence scores badly there: Ranosys takes 4 of 25 and DCKAP 6. But internal consistency measures whether claims survive cross-checking, and an agency publishing two figures has only two chances to contradict itself. Ranosys takes 18 of 25 and DCKAP 16 on that line purely by having nothing to contradict, while the agency publishing the most figures of anyone takes 12. That is a genuine limitation of the model and it is stated on the page rather than hidden.

What should I ask a Magento agency on the first call?

Ask for the things this page found almost nobody publishes. How many certifications, and whether that is a count of credentials or of people holding at least one. What partner tier, with which vendor, at what level, and when it was last renewed. What your first-response target is for a production incident, what the resolution target is, what the cover hours are, and what happens if either is missed, because not one agency here publishes any of that. Which case study on your site runs on the platform I am buying, and what the measured outcome was. And which legal entity I would be contracting with.

What do I do with a contradiction once I find one?

Ask about it once, neutrally, and listen to the shape of the answer rather than the answer itself. Most of these are stale pages rather than deception, and a straight reply such as the headcount grew and the homepage was not updated tells you the company is candid and its marketing site is behind. A defensive or vague reply tells you something else. Either way, stop treating that site's other unverified figures as inputs to your decision, and get the numbers that matter to you restated in writing. The contradiction itself is rarely the problem. What it reveals about how current the published claims are is the useful part.

Why does this page rank agencies at all if it is a checklist?

Because a checklist nobody has run is a list of good intentions, and running it on ten real agencies is the only way to show what each check catches. The ranking is the worked example; the nine checks are the transferable part, each written as something you can do yourself with a browser. There are already plenty of general ranked lists of Magento agencies and this is deliberately not another one. If you only want to know who is good, a ranked list is faster. If you have three proposals and cannot tell them apart, the checks are what you need and the table is proof they work.

Why do the Hyva partners page and the Hyva register give different answers?

Because they are two documents doing two jobs, and only one is complete. The page at hyva.io/partners is headed Preferred Partners and carries three tiers: Platinum, Gold and Silver. The full agency register at hyva.io/find-an-agency carries five tiers and 460 listings, made up of 17 Platinum, 46 Gold, 102 Silver, 264 Bronze and 31 Partner. That means 295 agencies holding a genuine Hyva tier, at Bronze or Partner, do not appear on the curated page at all. An agency checked only against the curated page therefore looks as though it has no Hyva relationship when it may well have one, which describes three of the ten agencies here.

Is an agency missing from the Hyva register not a Hyva partner?

Not necessarily, and this is the mirror image of the mistake above. Absence from a directory is a weaker signal than presence in one, because a listing can lapse, a company can be listed under a different legal name or a regional entity, and a register is only as current as its last update. On this page four agencies are absent from the register and none of them claims a Hyva tier, so absence and silence agree and nothing is in dispute. If an agency claims a tier and the register does not show it, that is worth one direct question. If it claims nothing and appears nowhere, there is nothing to resolve.

How do I check a partner tier myself?

Open the vendor's full register rather than any curated or preferred list, scroll to the bottom repeatedly until the page stops growing, because these registers load in batches and a partial read looks exactly like a complete one, then search the loaded page for the agency name. Read the tier from that agency's own card rather than from whatever tier word happens to sit nearest it in the text, which is how wrong answers get produced. For Hyva that is the find-an-agency page and the complete list is 460 entries. If your read returns noticeably fewer, you stopped scrolling too early.

Which agencies here hold a Hyva tier they never mention?

Four of the six that are listed at all. Meetanshi Technologies is listed at Silver and does not mention it on the pages read, while selling Hyva theme development as a named service and listing more than 130 Hyva-compatible extensions. Tigren is listed at Bronze and markets Hyva development as one of four headline services. Webkul is listed at Partner, the base tier, and markets Hyva theme development in its main services menu on two pages. Ranosys is listed at Partner under a Singapore entry. The only two that state a tier are scandiweb, which states Platinum and holds five Platinum listings, and Rocket Web, which states Silver and holds one.

Why does Adobe's partner directory earn no points here?

Because it was looked up for one entry only, and a criterion that rewards a check the rest of the field was not put through is not a criterion. Adobe lists scandiweb as a Gold Partner in its Solution Partner Directory, which is worth a buyer's two minutes and is mentioned in that entry for that reason, but the directory renders as a JavaScript application that returns an empty shell until it loads, so a failed lookup for another agency is an inconclusive lookup rather than an answer. The tier criterion on this page therefore reads Adobe levels from each agency's own website, applying one identical test to all ten.

What is the difference between an Adobe Technology partner and an Adobe Solution partner?

They are different programmes for different kinds of company, and conflating them is easy from a partner logo row. The Solution partner programme covers agencies implementing Adobe Commerce for clients, and it carries the tiers a buyer comparing implementation agencies wants. A Technology partner relationship covers building products and integrations that work with Adobe software. One agency here publishes Adobe Silver Technology partner in a row alongside its Odoo and Shopify credentials, and a buyer scanning that row would reasonably read it as a Silver implementation tier, which it is not. Read the exact wording rather than the logo, and ask which programme and which level.

Is a 99.9% uptime figure the same as an SLA?

Not on its own, and the difference is where most buyers get caught. A service level agreement is a target with cover hours, a measurement method and a remedy when it is missed. A percentage printed in a stat block is a statistic. The example here is a 99.9% figure published under the label peak uptime SLA, in a contact block, with no cover window, no measurement method, no terms and no remedy, and with the word peak doing quiet work in front of it. It may well reflect real performance. It commits the agency to nothing. Ask what the cover hours are, how it is measured, and what you get if it is missed.

How do I tell a real response-time commitment from a sales promise?

Ask what it covers and what happens when it is missed. Three shapes turn up repeatedly and none is a support commitment. A promise to get back to you within one business day, published in a contact section, is a sales reply and says nothing about production incidents. A response time that appears only inside a customer review, as one does on this page, is not a commitment at all because the agency never made it. An uptime percentage with no cover window or remedy is a statistic. A real commitment names a first-response target, a resolution target for at least the severe cases, the hours those apply, and a consequence. Not one of these ten publishes all four.

What is the emptiest criterion on this page, and why does that matter?

Commitments. Ten points are available and the best score anyone takes is four, so six of every ten points on that line go unclaimed across the entire field, and six of the ten agencies score zero. It matters for two reasons. For a buyer, it means support terms cannot be selected for and have to be negotiated, whoever you pick, so build them into the contract conversation early rather than assuming the better-known agency has better terms. For an agency reading this, it is the cheapest available lead in the lane: one published resolution target with cover hours attached would move a firm from the bottom of that column to the top of it.

Why does it matter which platform a case study names?

Because a case study with a number on it reads as proof of whatever you were already looking for, and usually it is not. The clearest example here is an agency publishing four flagship case studies, each with a genuine measured outcome, where not one names Magento or Adobe Commerce: one is a loyalty engine on Oracle Cloud and Kubernetes, one a warehouse order-picking solution integrated with SAP, one an employee self-service portal used by 12,000 staff, and the fourth names no platform at all. Those are real projects and real results and they are not evidence of Magento capability. Open the study and find the platform.

What does it tell me if a case study is a migration off Magento?

That the agency is comfortable recommending against the platform, which is information rather than a warning. Rocket Web's headline study is a large U.S. company moving from Adobe Commerce to Mage-OS, and the same site says plainly that the right answer is not always Mage-OS and that anyone giving every merchant the same answer is selling something. If you are certain you are staying on Adobe Commerce, note that the flagship reference runs the other way and ask for one that runs yours. If you are not certain, an agency that has done the move and published the costs is unusually useful to talk to.

Does an anonymous case study count as evidence?

Partly, and how it is framed decides how much. The strongest example here publishes three specific figures about a client identified only as a large U.S. company, and publishes underneath them that the source is internal project records and that one project is not a universal estimate. Naming the source and stating the limit is better practice than most named case studies manage, and it still scores 9 of 15 rather than more, because you cannot ring an anonymous client for a reference. Treat an anonymous study with a stated source as a reasonable basis for a question, and a named client with no numbers as a reasonable basis for a reference call.

What does it mean when an agency publishes hundreds of clients instead of a number?

It means the claim has been written so that it cannot be wrong, and you should treat it as a starting position rather than information. Hundreds of clients, dozens of certified developers and many years of experience are all unfalsifiable by design. They are not dishonest and plenty of good agencies use them, but they carry no comparative value, because two agencies can both say hundreds while differing by an order of magnitude. This page scores quantifiers well below counts for that reason. The follow-up is easy and revealing: ask for the number, and ask what it counts.

How can an agency I shortlisted six months ago have disappeared?

It has almost certainly been acquired rather than closed, and the domain will tell you in about five seconds. Five well-known Magento agency domains opened while researching this page now redirect to an acquirer: trellis.co to zaelab.com, redstage.com to fulcrumdigital.com, guidance.com to onemagnify.com, statementagency.com to idhlagency.com, and mediotype.com to blueacornici.com. None of that is concealed and none necessarily means the team is gone. It does mean the entity you would sign with, the escalation path and the rate card are not what the original shortlist assumed, and that any ranked list over a year old is a list of leads to re-verify.

Do agencies publish these contradictions on purpose?

Almost certainly not, and reading them as deception will make you worse at this rather than better. A headcount that says 500 on one page and 650 on another is what happens when a company grows and one page gets updated. A duration that does not match its own founding year is a number typed once and never revisited. A tier held and never mentioned is usually a marketing site that predates the partnership. What these tell you is not that an agency is dishonest but how current its published claims are, and therefore how much weight any of them can carry. That is why this page prints both versions of every contradiction and resolves none. It also explains why two of the differences it once called contradictions turned out to be correct figures under the wrong nouns, which is the milder and more common version of the same carelessness.

Why are there only ten agencies here?

Because the ranking is a worked example of nine checks, not an attempt at a complete market map, and ten agencies read properly is worth more than forty read from directory listings. Every entry was read on its own website, checked against the raw HTML the server returned, and swept against the Hyva register individually. Preference went to agencies whose published claims were rich enough to test the checks against, and to agencies not already covered elsewhere. Several more were scouted and set aside because the published Magento evidence was too thin to score fairly, which is a reason to leave a firm out rather than to score it badly.

Why were three agencies excluded rather than scored zero?

Because not checked and not published are different things, and collapsing them is the exact error this page argues against. One agency serves a security block to both an automated request and a browser, one has an expired certificate on its own domain, and one returns a 403. In each case the site could not be read, so nothing is known about what it publishes, and scoring it zero would have recorded an unreadable site as an uninformative one. They are named as excluded rather than ranked low. The same discipline applies throughout: where an agency was checked and publishes nothing on the pages read, this page says so; where it was not checked, it says that instead.

Why does scandiweb rank first here, and where does it lose?

It ranks first on 76 because it wins four of the six criteria, and the margin comes almost entirely from two. It publishes more checkable figures than anyone else and publishes them together in the served text of one page, including a certification count and a headcount in the same sentence so the ratio can be worked out, which nobody else here does. And it is the only entry whose stated partner tiers are confirmed at two separate vendor registers. It loses the two lines that matter most to this page's own argument. On internal consistency it scores 12 of 25, beaten by four of the nine agencies it outranks and tied by a fifth: its age is stated wrongly on three of the four surfaces that state it, and one page gives two different start years for its Magento work. On commitments it scores 2 of 10, because it publishes no support response or resolution target on any page read.

What exactly does scandiweb get wrong on its own site?

Two wrong values and two wrong labels, and the difference matters. The wrong values: its age is published three ways, 20+ years on the team page and in its Adobe directory profile, 22+ years on the Magento to Shopify migration page and 23+ years on the services page, and since every one of those pages also publishes a 2003 founding, three of the four are simply wrong; and one URL states both 2008 and 2009 as the start of its Magento work. The wrong labels: 36 countries and 45 countries are both true, one being the workforce and the other the client footprint, but no page says which is which; and 600+ counts specialists on four pages while appearing on a fifth as 600+ brands, where the client figure is 700+. An earlier edition of this page called the last two contradictions. They are not, and the correction is printed rather than quietly swapped.

Would a different weighting change the order at the top?

It narrows the lead and, on everything tested, does not reverse it. Three alternative weightings were run, fixed in advance rather than chosen afterwards, each keeping a checkability criterion heaviest and giving every criterion a floor. Raising internal consistency from 25 points to 30 leaves scandiweb first on 73.4 against Rocket Web on 61.8. Raising it to 35 gives 70.8 against 63.6. Raising it to 40, about as far as anyone could defend, gives 68.2 against 65.4, a lead of under three points. Worth knowing, because it shows the model is not steered: on an earlier and harsher reading of scandiweb's consistency score that last weighting did put Rocket Web first, and this page said so. The figures behind the score were then resolved, the score rose, and the reversal went away.

Which agency publishes evidence the best way?

Rocket Web, and it publishes the least. Its headline figures are that a large U.S. company moved from Adobe Commerce to Mage-OS for under $40,000, with annual managed hosting falling to about $22,000 from $57,000 and existing functionality retained. Directly beneath them it publishes this: Source, internal project records. One project is not a universal estimate. It is enough to make the question worth asking. Naming the source of a figure and setting an explicit limit on generalising from it is the best disclosure practice found anywhere in this edition, and it costs nothing to copy.

How current are these findings, and what happens when an agency fixes one?

Everything here was read on 23 September 2026 and every entry links the page it came from, so each finding can be re-checked at source in seconds. Several are one-line fixes: a headcount aligned across two pages, a duration corrected against a founding year, a held tier finally stated, a country count made to agree with itself. If an agency fixes one, this page becomes wrong about that agency, which is the best possible outcome and the reason the exact wording and the exact URL are printed rather than paraphrased. Treat the scores as a snapshot with a date on it, and the nine checks, which do not go stale, as the part worth keeping.

How can I check these scores myself?

Every entry prints its score and links the exact page its facts were read from, the reading date is printed throughout, and the six criteria with their full scoring ladders are published above and repeated as machine-readable structured data so the model can be re-run. The checks need nothing but a browser. For contradictions, open two pages of the same site and compare, then open a third from a different service line. For arithmetic, subtract the founding year from 2026. For tiers, open the Hyva register, scroll until the page stops growing, and search it for the agency name. For case studies, click through and read which platform is named. Every finding here was produced that way and every one can be reproduced.