Data Cleansing & Onboarding

Rapid B2B growth had filled Odoo with duplicate, inconsistent and incomplete restaurant records. A commercial rep standing outside a venue could not reliably tell whether it was already in the system. Growth could not match external intelligence to internal records, so segmentation and AI labelling ran on unreliable input.

Meanwhile the signup flow was the source of the mess: free-text addresses and optional-everything fields meant every new account risked becoming another duplicate.

• Duplicates on restaurant name + area, with no reliable de-dupe check
• Free-text location entry producing unmatchable address data
• Segmentation and AI enrichment degraded by dirty input
• Manual support effort spent reconciling records by hand

Three tracks running in parallel: cleanse what exists, harden the front door, and keep it clean automatically. Automation and pragmatism drove coverage — manually correcting every legacy record was an explicit non-goal.

• One-time cleansing: fuzzy match on restaurant name + area, merge, then enrich missing fields.
• Phone verification as the authentication anchor for every new account.
• Instagram handle as the business-identity signal — skippable, but the account stays flagged unverified pending a manual call.
• Searchable multi-select of canonical cities and zones. No free-text location entry anywhere in the onboarding UI.
• Branch count inferred from selected zones rather than demanded at signup.
• Scheduled nightly job scanning for new duplicates, suspicious fields and critical missing data.
• Save-time block: a clear name + area duplicate cannot be written.
• Audit log on every merge and edit, visible to the internal team.

20 tickets in Linear, priority Urgent, sequenced so prevention landed before the bulk merge — otherwise the cleanup would have been re-polluted while it ran.

• Built canonical city and zone master lists, kept in lockstep with platform location changes.
• Rewrote the signup flow: phone verification, then Instagram, then zone multi-select — no address, no documents.
• Added real-time de-dupe checks between the customer app and the Odoo data store via webhooks.
• Shipped a lightweight flag-for-review workflow so commercial and support could correct records without engineering.
• Ran the merge in batches with audit logging, so any bad merge stayed reversible.
• Rolled corrections out without disrupting live internal workflows — a stated project risk.

Data stopped being a hurdle and started being usable input. Field reps can search a name and area and trust the answer; growth can match external intelligence to internal records.

• Target of 99% accuracy and uniqueness for active restaurant records.
• Roughly half the manual support workload on record confusion removed.
• Segmentation by cuisine, location and menu intelligence unblocked.
• Signup friction reduced rather than increased — phone plus Instagram plus zones, in seconds.
• Designed to hold at 10x current data volume with batched scheduled jobs.