A vendor demonstrates an integration by copying one new customer into your CRM. Your billing team asks what happens when the customer already exists or changes its address. The demonstration does not answer either question.
Test the connection against the operations your business needs, including failed writes and disputed records. A visible connector proves that a connection exists; it does not prove that it can preserve the right data through the cases your staff will encounter.
Use a trial worksheet with explicit expectations
Consider a hypothetical booking-to-billing integration. The team wants completed jobs to create draft invoices, with staff acceptance before final billing. Use invented records or an approved test environment until the relevant owners accept the behavior.
| Trial case | Expected behavior | Evidence to retain |
|---|---|---|
| Known customer | Match the stable customer ID rather than name alone. | Source ID, destination ID, and resulting record. |
| Repeated completion event | A second event does not create a second invoice. | Both event attempts and the single resulting draft. |
| Permission limit | The connection performs the agreed action without unrelated account access. | Account permissions and demonstration of allowed/blocked operations. |
| Billing system unavailable | Record the failure and provide a controlled retry or staff recovery route. | Failure record, recovery action, and final record count. |
| Changed customer field | Respect the agreed field owner; hold unresolved disagreement. | Competing values, source, and the accepted correction. |
| Partial success | Expose what completed before the failure so staff can avoid repeating it. | Step results and a recovery attempt without duplicated work. |
A vendor may use different implementation methods. Accept the observable behavior and evidence you need rather than prescribing a design you cannot support.
Test recovery as part of the connection
In the worked example, the invoice system times out after receiving a write. Staff do not know whether the invoice exists. Ask the integration owner to show how they determine the result before retrying. Sending the same write again without checking could create a duplicate.
Agree on who receives a failure notice and what they can do with it. “Contact the developer” does not help a billing team close the day unless the support arrangement and response expectations fit that dependency.
Run the recovery case with the person who will handle it. They should find the affected job, determine what happened, and follow the accepted recovery route. If they need access they do not have, resolve that requirement before connecting live work.
Confirm which system owns changing fields
A booking contact and billing contact may differ for valid reasons. Decide which fields the integration copies, which it leaves alone, and which changes need human acceptance. Avoid a broad “sync customer data” requirement that gives every field the same rule.
Include a late-arriving update in the trial. A recent import timestamp does not necessarily mean the customer information is current. Ask how the connection handles stale source values and keeps enough history to explain a dispute.
Our native automation versus orchestration guide covers where a workflow should run. This worksheet checks whether the proposed connection can deliver the required behavior, whichever platform the team chooses.
Decide what can go live
After the trial, record each case as accepted, failed, or unresolved. Link the result to the evidence and name the person who accepts it. A missing capability may justify a narrower first release instead of rejecting the entire connection.
Ask about limits that affect the real workflow
Use the proposed scope to ask about supported operations, volume limits, account permissions, and the history available for investigation. Do not assume the connector can write a field because it can read it, or recover an event because it can show a successful demo. If the vendor documents a limitation, test the corresponding case or narrow the release. Retain the relevant documentation version with the trial results, so the maintainer can tell when a provider change affects an accepted assumption.
Document the supported volume, access owner, and recovery arrangement in the actual scope. Our maintenance guide covers ongoing responsibilities. For help designing and testing a connection around your business rules, see custom software development.

