Guide · July 2026

What an ERP integration actually costs

Quotes for the same integration routinely differ by a factor of five. That is not always vendor greed — it usually means the two firms are pricing different amounts of unknown. Here is what actually moves the number.

The short answer

$45k–$75k

One system, one direction, documented API.

$75k–$200k

Bidirectional, with business rules and real error handling.

$200k–$400k

Multi-system program with EDI, WMS and commerce in scope.

The four cost drivers

1. Direction

Reading from an ERP is straightforward. Writing into one is where the cost lives, because a write can fail halfway, be applied twice, or be rejected by a validation nobody documented. Bidirectional integrations routinely cost three times a read-only equivalent, and that ratio surprises people every time.

2. Data quality in the source

This is the most underestimated driver by a wide margin. The field exists, the documentation describes it, and in production it is populated for 60% of records — the rest carry a convention somebody adopted in 2011. Every such discovery is unplanned work, and they are never found during a sales conversation.

3. Business rules leaving the ERP

If the portal must show a price, someone must decide whether the ERP calculates it or the new system reimplements the calculation. Reimplementation is faster to build and permanently more expensive to own, because from that day forward two systems must agree about pricing forever.

4. Whether a real API exists

A modern documented API is the cheap case. Without one, the work becomes change-data-capture against the database plus writes through whatever supported interface exists — typically adding 30% to 60%, and adding upgrade risk that has to be managed rather than ignored.

What is usually missing from the quote

  • Reconciliation. Something will drift. Without a daily job that compares both sides and reports differences, you find out from a customer.
  • Observability. Alerting that names the record and the failed step, not a dashboard that says the service is up.
  • Replay. The ability to re-run a failed window without manual data entry — the difference between a ten-minute fix and a lost weekend.
  • The second environment. Testing against production because no sandbox was budgeted is how integrations damage real orders.

How to get a number you can trust

Buy a two-week integration spike before you buy the build. It runs against your real instance, proves the read and write paths, and surfaces the data-quality surprises while they are still cheap. It costs a fraction of the project and is the single highest-return purchase in this category.

A firm that quotes a fixed price for the full build without a spike is either carrying a large risk premium or planning to recover the difference through change orders. Both cost you more than the spike would have.

Get a range for your integration in two minutes.

Six guided questions, reviewed by an engineer. No sales sequence.