Notes / January 16, 2026

Adobe Commerce: B2B data model and performance habits (FORCE America)

Shared catalogs, company accounts, and the profiling discipline that keeps checkout fast.

B2B program notes—patterns and metrics, not proprietary schema dumps.

Adobe Commerce B2B adds companies, shared catalogs, negotiable quotes, and layered permissions. The product surface looks like toggles; the engineering surface is indexer cardinality, cache invalidation graphs, and price permission matrices that quietly destroy performance if ignored.

1. Shared catalogs and price books

What explodes

  • catalog_product_price (and company-specific dimensions) grows with SKU count × company groups × websites.
  • Reindex jobs that were “fine” in B2C become multi-hour when shared catalogs multiply effective SKUs per customer.

Mitigations

  • Segment companies into catalog groups deliberately—avoid “one company = one catalog” unless revenue justifies ops cost.
  • Partial reindex strategies: know which indexer invalidations your ERP sync triggers nightly.
  • Staging data volume must mirror prod cardinality—10 SKUs in QA will not find the slow query.

2. Permissions and cache

When a sales rep reassigns accounts:

  • Customer group / company association changes can invalidate full page cache and block-level FPC entries.
  • GraphQL caches may need tag-based invalidation if you cache products queries per company.

Documentation
We kept a matrix: event → invalidated tags → expected customer-visible delay. Support stopped guessing “up to 15 minutes” answers.

3. Company user ↔ customer linkage

Treat linking as a migration project:

  • Checksum report: counts of users, linked customers, orphans.
  • Idempotent repair command for drift after bulk HR imports.

Automated tests

  • Parameterized tests: given company X and SKU set Y, assert visible price and add-to-cart permission.

4. Performance habits that paid off

  • Blackfire / New Relic traces on checkout/cart, graphql, and company/users endpoints before peak.
  • MySQL slow log sampling on staging with production-sized dumps (anonymized PII).
  • Bulk APIs for ERP SKU updates instead of admin grid saves—human scale ≠ machine scale.

5. Quote / negotiation flows

Negotiable quotes add state machines—do not reimplement half in custom modules:

  • Subscribe to extension points (quote_submit, custom approval gates) instead of rewriting controllers.
  • Ensure email templates and history comments carry audit IDs for finance.

6. Incident examples (redacted)

  • Indexer backlog caused checkout timeouts—fixed by throttling ERP delta frequency + raising max_connections temporarily while we split jobs.
  • Wrong shared catalog assignment duplicated SKUs in navigation—caught by nightly catalog diff script comparing expected category counts.

Resume framing

“Partnered on Adobe Commerce B2B at scale: shared catalog strategy, company permission invalidation, indexer/reindex discipline, and automated tests around price visibility—closed performance incidents with measurable before/after.”