"It just needs to connect to the payment system" is the sentence that most often precedes a three-week delay. The coding is rarely the problem; the account, the approval and the differences between test and production are.
Why an integration is not a small task
Connecting to an external service involves more than calling it:
- Account creation and verification with the provider, often requiring business documents.
- Approval or review before production access is granted — payment providers in particular.
- Credential management, separately for test and production.
- Handling every failure mode the provider can produce: rejection, timeout, partial success, duplicate submission.
- Reconciliation — matching what the provider says happened with what your application recorded.
- Testing against a sandbox that behaves differently from production.
The last two are where the real effort sits, and they are invisible in a feature list.
Lead times to start early
These are calendar items, not development items, and they should begin at project start:
- Payment providers. Merchant account applications require business registration, banking details and sometimes a business model review. Days to weeks, and rejection is possible for some business categories — worth confirming acceptance before designing around a provider.
- Messaging services. SMS in many markets requires sender ID registration and template pre-approval, which is a formal process with its own queue.
- Platform integrations. Connecting to an established ecosystem usually requires an account with verification and may require review of how you use it.
- Enterprise systems. Connecting to a client's existing system means access, documentation and a test environment from a third party whose timeline you do not control. This is routinely the longest pole.
Sandbox is not production
Test environments are simplified. Differences that regularly surface at go-live:
- Rate limits are often absent or generous in sandbox and strict in production.
- Latency is lower in sandbox, so timeouts that never occur in testing occur in production.
- Error variety is narrower — sandboxes produce a handful of scenarios where production produces many.
- Data validation is sometimes looser in sandbox, so real data is rejected for reasons testing never revealed.
The practical mitigation is a production smoke test before launch: a small number of real transactions, with real credentials, verified end to end including reconciliation.
Designing for the failures that will happen
An external dependency will be unavailable at some point. What should happen then is a design decision, not an accident:
- Does the user's action fail, or queue? A payment must fail clearly. A notification can queue and retry.
- What does the user see? A specific message is far better than a generic error, and it reduces support load.
- Is the operation safe to retry? Payment and messaging APIs commonly support an idempotency key so a retried request does not charge twice. Using it is not optional for anything involving money.
- How do you find out? Errors from an integration should be logged and alerted, not silently swallowed.
Versioning and the change you did not make
Providers change their APIs. Versions are deprecated with notice, and the notice arrives by email to whoever registered the account — which is another reason the client should own it.
Two habits prevent the resulting outage: pin to a specific API version rather than tracking the latest automatically, and ensure the account's notification address reaches someone who will act on it. An integration that stops working because a deprecated version was retired on a published schedule is an avoidable failure.
Cost
Third-party services usually charge per use, and this is often the largest recurring cost in a small application. Worth establishing before committing:
- The pricing model — per transaction, per message, per API call, per active user.
- Whether a free tier exists and what happens at its ceiling.
- How cost scales at ten times current usage.
- Whether there are minimums or monthly commitments.
An AI or messaging integration that is inexpensive at pilot volume can become the dominant operating cost at scale, and finding that out in advance may change which provider is chosen.
Related Manboo Services
- Flutter App Development for iOS and Android MVPs.
- Business Software Development for CRM, booking, inventory and workflow systems.
- Admin Panel Development for operations dashboards and backend management.
- AI API Integration when your product is ready for smart search, chat or automation.
- Request a scope review and project estimate when you are ready to plan the first version.