Every ad platform retires old API versions each year, and field names, enums and metric definitions all move. If the pull logic is hard-coded in one place, a single upgrade leaves days of missing data.
The first layer is version isolation: keep the platform adapter separate from the business model so API changes touch only the adapter while the stored schema stays stable.
The second is backfill. Any failed sync must be re-runnable for a date range, and the re-run must overwrite rather than duplicate. Without idempotent backfill, gaps only get messier.
The third is rate-limit handling. Bulk account pulls trigger limits most often, so you need queues, backoff and priority: today's data first, historical restatement during off-peak hours.
The fourth is validation and alerting. Sample-check totals against the platform back office daily and alert when the gap crosses a threshold, instead of waiting for someone to notice a wrong report.