Shortlist rule
An agency qualified for the first edition in August 2026 if it met three tests on that date. It published ERP integration work for Magento or Adobe Commerce on its own site, with at least one ERP product named somewhere on that site. It was an agency rather than a software vendor, an integration platform, or a freelancer marketplace, so that its material described work it delivers rather than a product it sells. And Magento or Adobe Commerce was a stated platform for it, not a sideline mentioned once. Nine agencies passed. Agencies were excluded for failing one of the three tests, most often the second, and no agency was excluded for being small, for being outside North America and Europe, or for competing with the publisher.
The roster and the ranking order set in August were not changed in this edition. What changed is the evidence beneath them: every agency's pages were re-read on 22 September 2026, several claims were removed because their source pages no longer show them, and three facts were corrected. The changelog lists each. If a re-read had shown that an agency no longer met the shortlist rule, it would have been removed with a changelog entry; none did.
Criteria and weights
| Criterion | Weight | What scores | What scores nothing |
|---|---|---|---|
| Published ERP case study naming client, ERP, and outcome | 25 | A live case page naming the client, the ERP product, and a measured result after go-live | A service page with ERP logos; a testimonial without a product or a figure |
| Named ERP editions and APIs | 15 | The edition and the interface: Prophet 21, Business Central API v2.0, S/4HANA over OData | A vendor name alone |
| Async bulk, queue, and rate-limit handling | 15 | Stated queued or asynchronous processing, retries, a dead-letter path, or sizing against the ERP account's limit | Silence, or a promise of real-time sync without a mechanism |
| Ownership and cadence design | 15 | A written split of which system owns price, stock, credit, and order status, with a cadence per object and the pattern stated before contract | A pattern chosen after signature |
| Adobe Commerce B2B scope and licensing | 10 | Treats B2B as the separately installed Adobe Commerce package and maps its features to ERP sources | Contract pricing promised on Magento Open Source |
| App Builder and out-of-process readiness | 10 | Evidence of an integration running outside the Magento process | All integration logic as an app/code module |
| Post-launch support scale and response commitment | 10 | A published support service with a stated response commitment | A project team that disbands at go-live |
Weights sum to 100. The largest weight sits on the published case because it is the only criterion a buyer can check without a sales call. The three engineering criteria (queues, ownership, out-of-process) together outweigh it because they predict post-launch failure, but each scores only where an agency publishes evidence, so most agencies score nothing on them and the ranking leans on what is publicly checkable. That is a limit of the method, stated here rather than hidden.
Evidence accepted
- The agency's own site: service pages, ERP pages, case studies, about and contact pages, read on the retrieval date
- The agency's Clutch profile, for headquarters, founding year, size, hourly rate, minimum project size, and any ERP named there
- Partner directories that list tiers publicly, such as hyva.io for Hyvä tiers
- Vendor documentation for platform facts: Adobe's lifecycle, B2B, bulk-API, and feature-comparison pages, Adobe's starter kit repository, and Oracle's NetSuite documentation
- A case study with a number and a named client, counted as that project's result and never as a forecast
Evidence never accepted
- Hearsay, including what one agency says about another
- Other rankings, award lists, and directory badges
- Review snippets without a source page, and review counts as a proxy for quality
- Search-engine snippets or summaries of a page that was not itself read
- Anything from a language model's memory; drafting is AI-assisted, sourcing is not
- Any figure an agency reports about itself where a third party could change it, used as a ranking input; such figures are recorded in the ledger as the agency's statement and kept out of body copy for the publisher
How scandiweb is scored
scandiweb is held to the same rule and is scored on its case studies (Macron, IONTO, BK-Group, Wainbee, Rocket Industrial), its integration services page, and its Clutch profile. Its self-reported counts (certifications, projects, clients, tickets) are recorded in the ledger as its own statements and do not appear in the body copy of the ranking; its Adobe partner tier is recorded as its own statement because Adobe's directory needs a partner login; its Hyvä tier is recorded as verified because hyva.io lists it. The author is a scandiweb employee and says so. A reader who believes the publisher's position is wrong can compare scandiweb's evidence rows against any other agency's in the ledger; the rows are the argument.
Disqualifiers
An agency is removed from the ranking if its site stops publishing Magento or Adobe Commerce ERP work, if it becomes a software vendor rather than an agency, or if it asks to be removed and its profile contains a factual error the site declines to fix. An agency is not removed for disputing its rank, for being small, or for competing with the publisher. A claim is removed, not softened, when its source page no longer shows it on re-read.
Refresh cadence and version history
The site is refreshed monthly. Each refresh re-reads every source, updates the retrieval dates, records removed claims with status removed in the ledger, and adds a dated line to the changelog. Method version 1.0 was the August 2026 edition, with eight criteria and a single ranking page. Method version 2.0, this edition, merged the pattern criterion into ownership design, added the direct answer, the scenario table, the question-by-question section, the per-agency profiles, the evidence ledger, the guides, and the author pages, and applied the removal rule for the first time. The next change to the criteria or weights will be version 2.1 and will be announced in the changelog before it takes effect.