Data strategy as a prerequisite for AI ROI

Cleaning the plumbing before buying another model seat—an analytics-led engagement.

Operations leaders wanted forecasting and assistant features, but joins across ERP, spreadsheets, and the TMS were unreliable. We reframed the program: data contracts first, models second. Leadership accepted a quarter of “boring” data work because prior tool purchases had failed when pipelines could not be trusted.

Challenge

Three sources of truth for shipment status; manual CSV exports for weekly reporting. Vendor demos assumed cleaner data than existed. Ops leads were tired of dashboards that disagreed by Monday afternoon.

Approach

Two-week data lineage sprint: entities, owners, refresh cadence, and known gaps.
Minimum viable metrics layer with agreed definitions before any generative use cases.
AI use-case backlog scored by data readiness and business impact.
Named data owners per entity (shipment, customer, lane) with written refresh SLAs—so fixes stuck.

Outcomes

Single executive dashboard with signed-off definitions—used in ops reviews within the month.
Deferred two vendor purchases until ingestion SLAs were met.
Clear gate: no new AI spend without passing data-readiness checklist.
First assistant use case approved only after two consecutive weeks of matching manual baseline reports.