Every ERP integration eventually reduces to one question: when the storefront and the ERP hold different values for the same field, which one wins?

Decide ownership first

Sync patterns that hold up

Real-time for inventory checks at cart and checkout, scheduled batches for catalog and pricing, and queued retries for orders so a brief ERP outage does not lose a sale. Anything that requires the ERP to be reachable for a customer to browse will eventually take your storefront down with it.

Questions worth asking

Ask how they handle a failed order push, how they reconcile inventory drift, and whether they have integrated your specific ERP version. Version matters more than vendor.

B2B buyers are not shopping. They are executing a purchase decision that was already made, usually under a contract, often on someone else’s budget.

Features that actually get used

• Company accounts with multiple users, roles and permissions.
• Contract pricing visible per account, including volume breaks.
• Reorder from order history and saved lists — often the most-used screen in the entire portal.
• Quote requests for non-standard configurations or quantities.
• Approval workflows where orders above a threshold route to a manager.
• Self-service invoices and payment status, which quietly removes a lot of support load.

What to prioritise

Reorder and account management first. Approval workflows second. Elaborate product discovery last — B2B buyers usually know exactly what they want and search by SKU.

Measure success by reduction in phone and email orders, not by time on site.

Three systems, each convinced it owns the product description, is a support ticket generator. The fix is boring and organisational rather than technical.

The ownership matrix

Produce a table listing every shared object — product, price, stock, customer, order, invoice — and for each one record the owning system, the direction of sync, the frequency, and the conflict rule. Get it signed off by the people who own each system, not just by IT.

One direction wherever possible

Bi-directional sync is occasionally unavoidable, but each bi-directional flow roughly doubles the debugging surface. Where you cannot avoid it, define the conflict rule explicitly: last-write-wins, or system-of-record-wins, and never “whichever ran most recently”.

Build for observability

Log every sync with a correlation ID, keep failed payloads, and give the operations team a screen where they can see what failed and retry it without a developer. Integrations that can only be debugged by reading server logs become the first thing nobody wants to touch.