The Onboarding Avengers

Go-live should not depend on discovering broken customer data at the last minute. Procurement onboarding pulls together suppliers, categories, users, pricebooks, purchase orders, data fields, and workflow rules, but the issues are usually hidden until manual review or production setup. We built the Onboarding Data Quality Agent: a readiness check that turns onboarding data into a clear, tenant-scoped go-live report. It gathers data from samples, payloads, or legacy self-service, runs deterministic validation rules, scores readiness, recommends exact next actions, and uses AI only where it helps humans understand and fix the findings. The result is simple: implementation teams know what is ready, what is blocking, why it matters, and what to do next. The problem. Customer onboarding data is messy, fragmented, and risky. Suppliers can be missing external IDs, categories can point to missing parents, workflows can reference inactive users, pricebooks and purchase orders can reference unknown suppliers or categories, and required data fields can be missing defaults. These problems are expensive because they surface late, after teams have already spent time configuring a customer. Manual review works, but it is repetitive, inconsistent, and hard to scale across large legacy datasets. The solution. The Onboarding Data Quality Agent gives onboarding teams a repeatable pre-go-live gate. It converts customer onboarding inputs into a normalized snapshot, validates that snapshot with backend rules, computes a readiness status and health score, persists the report, and returns schema-backed recommended actions that a frontend or implementation team can act on. The LLM is not the source of truth. Rules create the issues, scores, blockers, and action targets; AI explains those facts in customer-friendly language, investigates likely causes, and powers report-grounded chat. How it works. 1. Data gathering. The agent can run against demo samples, direct payloads, or legacy self-service data. For large legacy datasets, compact mode summarizes evidence instead of stuffing huge raw purchase-order files into a report. 2. Deterministic validation. The rule engine checks supplier IDs, duplicate suppliers, category mappings and hierarchy, user readiness, pricebook references, purchase order references, workflow users, and required data-field defaults. 3. Readiness scoring. Every report receives a clear status — READY, READY_WITH_WARNINGS, or NOT_READY — plus a 0-100 health score, issue counts, and top blockers. 4. Recommended next actions. The backend turns issues into structured action objects such as fixing supplier external IDs, reviewing duplicate suppliers, repairing category mappings, or reviewing purchase-order references. Each action can include a schema so the UI knows exactly what remediation form to render. 5. AI summary and investigation. When requested, the report agent writes customer-facing language from the deterministic findings, and the investigation agent groups symptoms into likely root causes, mapping hypotheses, customer questions, and remediation steps. 6. Report-grounded chat. A chat agent can answer questions about a specific report using capped backend context and dataset tools, so admins can ask why something failed without handing the LLM the entire dataset. 7. Operational readiness. Reports are tenant-scoped, persisted, observable, and designed for scheduled or manual workflow runs. The system is built to support real onboarding operations, not just a one-off demo. Why it wins. Onboarding teams get a reliable go-live signal before the expensive part of implementation. Customers get clearer explanations instead of vague import failures. Engineers keep deterministic validation as the source of truth, while AI handles the human-facing explanation layer. The product turns late-stage onboarding surprises into an early, repeatable, actionable preflight report. Resources. Main https://fairmarkit.atlassian.net/wiki/spaces/FUSE/pages/4648271875/Onboarding+Data+Quality+Agent Live App https://uat.fairmarkit.com/services/preflight-agent/ App documentation https://uat.fairmarkit.com/services/preflight-agent/docs/ GIT Backend https://gitlab.fmdev.io/integrations/preflight-agent/ GIT Frontend https://gitlab.fmdev.io/fairmarkit/frontend/-/tree/hack27/data-quality-agent