Included in a typical proposal
- Design and build of the named interface.
- Mapping documentation.
- A test plan for success and failure cases you listed.
SOFTWARE DEVELOPMENT
Connect systems so information can move through a defined, dependable process. Agree data ownership, mapping and failure handling before building the connection.
WHAT THIS SERVICE IS
API development and integration is the work of moving information between systems on purpose: a documented interface, a mapping of fields, authentication, and a plan for when a call fails. It is how a website, a warehouse tool and an accounting product stop needing a person to copy values. It is not a Zapier account with no owner, and it is not scraping a page that forbids it.
The important decisions are ownership (which system is allowed to win if two disagree), frequency, and what a human should do when the other party is down. Retries without a dead-letter path create duplicate orders. We would rather a visible failure than a quiet double charge.
Provider limits, sandbox quality and the other vendor’s change notices belong in discovery. An “API” that is three undocumented XML files on an FTP server is still integrable; it should just be named honestly.
A CLEAR PICTURE
Work cannot start on a promise that “IT will send them”.
Then in a spreadsheet of fields.
Until a signed sample matches.
They will happen; the path must exist.
WHO IT IS FOR
Teams connecting a store to fulfilment, a form to a CRM, or a custom app to Microsoft 365. Also businesses whose previous integrator left no diagrams.
TYPICAL SCOPE
Read the real docs, auth method, rate limits and whether a sandbox exists. If the vendor must enable a paid API tier, that is on the critical path.
Each field, direction, and the winner on conflict. Time zones, currencies and Arabic versus English fields are written down. Silent truncation is not a mapping.
The agreed authentication, jobs or webhooks, and idempotency where orders are involved. Secrets live in a vault, not in a repository.
Logging you can read, alerts if you bought them, and a manual replay or skip path. We will not call a queue “self-healing” if nobody can see it.
INCLUDED AND QUOTED SEPARATELY
AFTER HANDOVER
Sequence diagrams, field maps, and how to rotate keys. If we leave a worker process, you get the repo and the host. A laptop cron job is not an architecture we will hand over as production.
When the other vendor changes a field, someone must read their mail. Put that on a calendar. Integrations rot at the other party’s pace.
PRACTICAL NOTES
Idempotency keys and a stored external id are how you avoid double orders when a retry happens. If the other vendor does not support them, we will say what the fallback is — a human queue, a fingerprint of the payload, or a slow manual check. There is no universal magic header.
Versioning is neglected until the vendor ships v2. Put their status page and a mail filter on a named inbox. The integration owner is a role, even if it is one person. When they leave, the mapping document should still be in your wiki, not in a chat export.
PLANNING THE WORK
Sometimes via files or a supported middleware. Sometimes no. Screen-scraping is a last resort with a short expected life, only if you accept that in the proposal.
You, on accounts we document. We will not hide a critical path inside an unpaid personal automation account.
If both support events and you pay for that tier, closer to it. Many useful integrations are every five minutes. Latency is a requirement you set.
CONNECTED SERVICES
LET’S BUILD WHAT’S NEXT
Tell us what you want to improve, build or simplify. We’ll help you define a practical way forward.
Discuss your project