Sync Error AI Assist

The component that turns a backlog of failed NetSuite sync records into an AI-diagnosed, DLP-redacted, reasoning-traced, operator-approved queue. Ranked the #1 first-to-bill wedge in the Squire evaluation because it rides an error flow Squire already runs rather than replacing SyncCentral’s production pipeline.

What it is

When a NetSuite record fails to sync, today a consultant has to open it, work out what went wrong (often a field-mapping mismatch), and fix it by hand. Sync Error AI Assist sits beside that existing error-record flow and does the first pass automatically: each failed record becomes a claim the AI diagnoses, explains, and proposes a fix for — which an operator then reviews and approves before anything is written back.

It is deliberately assist, not autopilot. The AI proposes; a human disposes; every step is recorded.

Why it matters — the first-to-bill case

Per 30-suitecentral-for-squire and 33-suitecentral-deployment-options, Sync Error AI Assist is the #1 ranked feature for speed-to-revenue, ahead of even AI Field Mapping. The reason is leverage, not novelty: it enhances the error-handling Squire already operates across its existing SyncCentral deployments, so it can be switched on beside the current flow with human approval around it — the fastest path to billable consultant-hour savings. This is the heart of the recommended Enhance, don’t replace posture in three-review-paths and the pilot-30-60-90 plan.

How it works

The pipeline is webhook-driven with a polling backstop:

1. Ingest. A NetSuite User Event SuiteScript POSTs each sync-error record to /api/sync-error-assist/ingest with an HMAC-signed payload (constant-time verification). The server inserts the claim row synchronously and returns 202 Accepted.
2. Diagnose. A fire-and-forget worker runs the AI suggestion pipeline — a real provider via the provider registry produces a diagnosis and a proposed fix (e.g. normalizing a mismatched phone-number field).
3. Govern. The suggestion is routed through the governance layer before it can be applied: DLP masking of sensitive fields, a reasoning trace, cost tracking, operator approval, and a tenant-attributed audit-log entry. This is the same controls posture as the nl-action-gate and the broader embedded-intelligence governance story.
4. Backstop. A 30-minute polling reconcile runs alongside the webhook path to catch any missed deliveries, and a watermark-recovery sweep re-queues records that stalled mid-processing.

Enablement is per-tenant opt-in (sync_error_assist.enabled + webhook_enabled), so it is off until a tenant explicitly turns it on.

Honest status: beta, demonstrated, pilot-ready

Consistent with production-vs-demo and production-proof, the honest framing is “demonstrated, pilot-ready,” not “in production.”

• The webhook ingest path, HMAC verifier, AI pipeline, governance controls, and operator review UI are all wired in code with ~155 unit + integration tests (PR 17a data/service core, 17b operator UI, 17c real-time webhook).
• The proof card status is beta: there is no production credential test on file yet — a pilot against real NetSuite + a live provider + a deployed SuiteScript is the next milestone.
• The 10-scenario fixture corpus is Claude-drafted, not Squire-empirical. The pilot’s accept-rate target measures the AI on real Squire data; the fixture proves shape, not realism.

So the claim to make is “this rides your existing error flow and is ready to pilot on one SyncCentral deployment” — not “this is running in production today.”

The demo

A 4:39 executive proof clip walks the whole flow on a NetSuite customer-sync phone-number mismatch, through DLP/trace/approval/audit to write-back. See sync-error-assist-video for the chapter summary and where to watch it.

Verify

The reviewer evidence lives in the Sync Error Assist proof card (docs/review/proof-cards/sync-error-assist.md), which carries the source map, the ~155-test list, the live-vs-fixture posture, the known gaps, and a 60-second verification recipe. See also the claim-to-evidence map in claim-proof-matrix.

Related: three-review-paths · pilot-30-60-90 · production-vs-demo · suitecentral-deployment-options