Cascadia BI Migration
A 24-week program to move an organization off legacy BI and onto a governed, AI-assisted platform — with a certified metric layer at the center.
BI Leadership / Metric Governance
Cascadia BI Migration
How I’d run a Tableau → AI-native BI migration: retire self-serve reporting for the general population, migrate the dashboards that actually matter, and stand up a certified metric layer so the numbers the business runs on have one owned definition. An illustrative program design — company-agnostic, no confidential information.
The Scenario
A high-growth company runs ~500 dashboards on a legacy BI tool — but fewer than 100 are actually used, and only ~20–25 are heavily used. A license renewal is the forcing function: renew for the whole population at full cost, or migrate the reporting that matters onto a modern, AI-assisted BI platform (sitting on the existing cloud data warehouse) and renew lean for a handful of admins.
The trap in a migration like this is treating it as a copy job. Rebuilding 500 dashboards on uncertified metrics just moves the technical debt to a new tool. The program below does the opposite: certify the numbers first, migrate only what’s used, and make parity QA — proving the new number equals the trusted old number — the actual deliverable.
Guiding Principles
- Certified layer first, dashboards second. A migrated dashboard built on an undefined metric just relocates the problem. Define the number, then build on it.
- Migrate what’s used; retire the rest on a published date. Most of the ~500 dashboards should be retired, not moved.
- Two tiers, always labeled. Certified metrics for decisions; exploratory clearly marked as not-yet-locked.
- Parity QA is the deliverable. With AI-assisted, prompt-driven rebuilds, construction is cheap — the value is proving the new number matches the trusted old one, or is correctly different and documented why.
- No big-bang. Wave-by-wave cutover, each with hypercare.
- Partner, don’t turf-war. BI owns definitions, the context layer, and QA; platform engineering owns the warehouse and pipelines; data science owns the AI-platform depth.
- Governance, not gatekeeping. Enable self-service on a trusted base rather than becoming the bottleneck everyone routes around.
The Plan
Sequential phase (has an exit gate) Pilot (proof of the motion) Continuous track Decision gate (go/no-go)
Front-loaded discovery and governance; migration waves through the middle; sunset lands on the renewal deadline. Governance and QA run continuously once stood up, and bars overlap by design.
Phases & Exit Gates
| Phase | Weeks | Objective | Exit gate |
|---|---|---|---|
| 0 · Mobilize & Align | 1–2 | Confirm scope, mandate, and the real state of the platform + warehouse | Charter signed; BI / platform-eng / data-science seam agreed in writing |
| 1 · Discover & Inventory | 2–5 | Know what exists, what’s used, and what the business actually runs on | Prioritized backlog + retire list approved; top-15 metric shortlist |
| 2 · Govern & Certify | 4–8 | Stand up the trusted metric layer the platform consumes | Certified Metric Catalog v1; parity-QA process ratified |
| 3 · Pilot | 7–10 | Prove the motion end-to-end with a friendly, high-value area | Pilot adopted; parity signed off; per-dashboard runbook validated |
| 4 · Wave Migration | 9–20 | Migrate the used dashboards in prioritized waves | All used dashboards migrated & certified; only long-tail remains |
| 5 · Sunset & Renewal | 18–24 | Turn legacy BI off for the general population; renew lean | General population off legacy BI; ~5-license renewal executed |
Risk Register
| Risk | Impact | Mitigation |
|---|---|---|
| Stakeholder outreach never gets prioritized | High — discovery stalls the whole plan | Make it the Phase-1 headline; short structured interviews; executive air cover to secure time |
| “Same number, different answer” after rebuild | High — kills trust in the new platform | Certify the metric before rebuild; parity QA every dashboard; label board-vs-operating differences explicitly |
| Refresh-lag surprises users expecting real-time | Medium | Set as-of expectations in the semantic layer; match cadence to the decision; escalate true real-time needs to engineering |
| AI skeptics / “the tool is wrong” | Medium | Most “wrong” is an undefined metric — show the certified definition + parity, and win skeptics one number at a time |
| Turf friction over the context layer | Medium | Agree the seam explicitly at G0; co-own the context layer — BI brings business definitions, data science brings platform depth |
| Hidden legacy tech debt (“nobody knows how it was built”) | Medium | Treat undocumented dashboards as retire-candidates unless a user defends them; don’t faithfully reproduce debt |
| Scope creep (migrate all 500 / gold-plate) | Medium | Retire-by-default; migrate only used; keep exploratory cheap and clearly labeled |
Success Metrics
- License reduction: ~150 → ~5 by renewal (the headline).
- Dashboards: # migrated & certified vs. # retired vs. # remaining (target: 100% of used migrated).
- Certified metrics: count in the catalog, each with a definition, source, and owner.
- Adoption: active users on the new versions; legacy usage trending to zero for the general population.
- Trust: definitional disputes raised vs. resolved; parity pass-rate at cutover.
- Cost: license spend avoided; warehouse-refresh savings from governance.
- Cycle: time-to-migrate per dashboard, trending down as the runbook matures.
Skills Demonstrated
BI program leadership and delivery · metric governance and certification (one definition, source, and owner per number) · stakeholder management and change enablement across non-technical teams (sales, clinical ops, finance, RevOps) · parity QA for AI-generated queries · cross-functional partnership with platform engineering and data science · phased delivery with explicit go/no-go gates.
An illustrative program design demonstrating how I approach a BI platform migration and metric-governance program. Tools referenced are generic; no proprietary or confidential information is included.