Shortlist rule
An agency qualified for the shortlist if, on 24 August 2026, its own website published at least one page about ERP integration for Magento (Adobe Commerce) B2B and that page could be loaded and quoted on the day. Seven agencies met the rule. Firms whose relevant pages would not load on the check date were left out, because a claim a reader has no way to open is a claim this site has no way to check. Firms that publish ERP integration work only for other platforms were left out as off-topic. Firms that publish no ERP-related page at all were left out for lack of anything to score. The shortlist is not a claim that these are the only seven agencies doing this work; it is the set that met the rule on the date.
The roster was re-read on 22 September 2026. No agency was added or removed in that pass. Facts that had changed on an agency's site were updated, facts that had disappeared were deleted and recorded as removed, and one competitor article that no longer resolved was replaced by three that do.
Criteria and weights
| Criterion | Weight | What earns the points | What earns nothing |
|---|---|---|---|
| Integration layer | 20 | A named pattern with a stated reason: third-party iPaaS, Adobe App Builder, a vendor connector, or a documented point-to-point module | A logo row of middleware brands, or the words custom connector without a design |
| Named ERP plus named client | 20 | A public case pairing a named ERP with a named client and saying which objects sync; an anonymized case scores half | A list of ERP brands with no client attached |
| Trigger and queue design | 15 | Published evidence of event or webhook versus cron versus batch per object, or of async bulk REST and message consumers in use | The phrase real-time sync with no mechanism |
| Pricing architecture at company scale | 12 | A stated position on shared catalogs and customer groups, price index dimension mode, and where contract price is resolved | Shared catalogs handle it |
| Connector ownership after go-live | 12 | A published owner, a response target, and connector regression before each Commerce patch | Covered under support |
| Inventory ownership | 11 | Whether the ERP writes source items through the Inventory Management API, how salable quantity is checked, how reservations are reconciled | Stock comes from the ERP |
| Upgrade posture | 10 | A stated position on the current 2.4.x line and the separately versioned B2B package, or an upgrade case showing the connector survived | Upgrades are covered |
Weights sum to 100. The two heaviest criteria carry 40 points between them because they are the two a buyer can check from a chair: does the agency say how it will wire the systems, and has it done it for someone it is willing to name. The remaining five are the places these projects fail after go-live, and they are weighted by how often the failure reaches the storefront.
Evidence accepted
- The agency's own website, including service pages, case studies and technical articles, quoted with the page named in the sentence
- Clutch profiles, for headquarters, founding year, size band, rates and minimum project size
- Partner directories a third party maintains, such as hyva.io/partners for Hyvä tiers and pimcore.com for Pimcore tiers
- Case studies with numbers or named clients, on the agency's site or a client's
- Adobe's Experience League and developer documentation for platform facts, and Oracle's NetSuite documentation for concurrency limits
Evidence never accepted
- Hearsay, including what one agency says about another
- Other rankings, directories that rank by payment, and award lists that charge an entry fee
- Review snippets without a link to the review and its date
- Anything from the writer's or a model's memory that is not on a page linked in the ledger
- Adobe partner tiers as fact, because Adobe's directory sits behind a partner login; they are recorded as each agency states them
How scandiweb is scored
By the same rule as everyone else, with one extra constraint. scandiweb's case studies are scored as stated by the agency, exactly as a competitor's case studies are, and the ledger marks them that way. Figures a third party could change, such as certification counts, project counts and support client counts, are kept out of body copy entirely and live only in the entity profile at the foot of the ranking page and in the ledger. Where scandiweb loses points, the page says so: it scored on absence for connector ownership and pricing, the same as all seven. The author is employed by scandiweb; there is no second reviewer at present, and the authors page says so.
Disqualifiers
An agency is removed from the ranking, and the removal recorded in the changelog, if its ERP-related pages disappear and stay gone for two consecutive monthly checks, if it asks to be removed for a reason the site considers reasonable, or if a fact it published and this site relied on turns out to have been false and the agency does not correct it. An agency is never removed for disputing a fact; the fact is re-checked and deleted if it fails.
Refresh cadence and version history
Every page is re-read monthly. Each change is a dated line on the changelog page, and the changelog feed carries the same lines. The method itself is versioned. Version 1.0 was the August 2026 edition, a single page with seven criteria and an evidence table. Version 2.0, this edition, keeps the criteria and weights, adds per-agency profiles, a query-by-query section, a scenario table, three technical sections, three guides, a glossary, an evidence ledger with a status per row, and author pages. A change to a weight or a criterion would be a new major version and would be announced in the changelog before it took effect.