Every marketing stack eventually hits the same wall: the two tools you need to talk to each other don't have a native integration. That is where APIs stop being a developer topic and become a marketing capability. This guide is for marketers who need to scope integrations intelligently — and for the developers who get handed the ticket.

When to build vs when to buy

Default order of preference: native integration first (cheapest to own), middleware second (Zapier/Make — days not weeks, with per-task costs at scale), custom API third. Custom builds are justified when data volume makes per-task pricing absurd, when logic exceeds what middleware expresses (multi-branch transformations, custom retry semantics), or when data residency and security requirements forbid third-party processors.

The patterns you will actually use

  • REST polling on a schedule ("every 15 minutes, pull new leads") — simple, laggy, fine for reporting flows.
  • Webhooks ("tell me when X happens") — real-time, near-zero lag, the right default for lead routing and event sync. Your endpoint must handle retries and out-of-order delivery.
  • Batch ETL for analytics: nightly exports into your warehouse; the honest path for reporting that doesn't need to be live.
  • Queue-based pipelines with a worker: the scalable pattern once volume grows — messages land in a queue, workers process with retry and dead-letter handling.

Auth reality

OAuth 2.0 with refresh tokens is the standard you will meet (Meta, Google, HubSpot...). Operational truths: tokens expire and must be refreshed and stored securely; scopes should be minimal (read what you need, not admin); and every platform has rate limits — Meta's API is the strictest of the major ad platforms, and batching calls is not optional at volume.

Failure modes and maintenance

  • Silent failures: a webhook endpoint returning 200 for partial processing. Log payloads; alert on anomalies.
  • Schema drift: the platform adds a field, your mapping breaks six months later. Version APIs where offered and monitor changelogs.
  • Retry storms: a downstream outage meets a naive retry loop. Use exponential backoff and dead-letter queues.
  • The unpriced maintenance: custom integrations are software with a lifecycle — budget 10-20% of build cost annually for upkeep.

A scoping checklist for the marketing side

  1. Define the business event in plain language ("when a lead is marked qualified in the CRM, tell Google Ads within 5 minutes").
  2. List systems, direction of data, required fields, and failure tolerance (minutes? hours? next day?).
  3. Decide the source of truth per field — integration arguments are usually data-governance arguments in disguise.
  4. Include monitoring and an owner in the spec, not as afterthoughts.

Do marketers need to learn to code for this?

No — but marketers who can read API docs, JSON payloads, and webhook logs get integrations built faster and broken ones fixed sooner. Literacy beats programming.

Zapier or custom — where is the line?

Per-task pricing at volume, branching logic beyond middleware's comfort, or compliance requirements. Until one of those triggers, middleware is the rational choice.

Our marketing technology service designs and builds these integrations. Start a project.