DATADATADATADATA · Methodology
180 Days to a Data-Driven Company
A playbook for organizations with scattered data, conflicting KPIs, and the ambition to run on company-wide governance — and to be genuinely ready for AI.
The warehouse is the last problem to solve, not the first.
Most data orgs have a definition problem, an ownership problem, and a discipline problem before they have a technology problem. This playbook fixes them in that order — and ships a visible trust anchor early, so leadership has something to stand on while the deeper rebuild happens underneath. Every phase is held to a business test, not a tidiness one.
And it matters more now than ever. AI is only as trustworthy as the data beneath it — a copilot answering from contradictory definitions just produces confident nonsense faster. The same governance that fixes BI is exactly what makes a company AI-ready.
01The situation this solves for
If most of these are true, this playbook is written for you.
- No single source of truth. The same KPI disagrees with itself across reports.
- A stack that grew by accident. A legacy database plus a cloud warehouse that became a dumping ground. No transformation layer, no governance.
- A team built as a service queue, not a capability. Order-takers, not analysts.
- One function already works. Proof of what discipline could look like everywhere else.
- AI ambitions on shaky ground. Copilots and agents on the roadmap, but no clean, defined data for them to stand on.
- The goal is governance the business can trust — company-wide, and AI-ready.
02Operating principles
The dictionary comes before the warehouse
Technology is the last part of the problem, not the first. Most rebuilds fail because they migrate the chaos into a more expensive system.
Buy political room with an early trust anchor
Most data leaders disappear into a long rebuild. Ship one dashboard leadership can trust before touching any infra. That's what makes the deeper rebuild possible.
Ad-hoc is a registry, not a refusal
Ad-hoc questions don't go away. The difference is whether they're logged, sorted, and turned into permanent dashboards when they repeat.
Self-serve is the destination; the team is the road
The function should be working itself out of the bottleneck position. If a small team still answers every question by year-end, the design is wrong.
AI is only as good as the data under it
Clean, governed, well-defined data is the precondition for trustworthy AI. The dictionary and semantic layer aren't just BI hygiene — they're the grounding that lets a copilot answer in plain language without making things up, and the same work that gets you there is what makes the company AI-ready almost for free.
03The 180-day plan
Click any phase to expand. The plan is a starting position, designed to be revised against evidence from the first weeks.
The first weeks build nothing. They produce four artifacts everything else depends on.
Artifact 1 — The Chaos Audit
An inventory of every dashboard, every source, every KPI definition in use — cross-referenced so the contradictions become visible on paper. It doubles as an AI-readiness baseline: you can't ground a model on data you can't define.
Artifact 2 — KPI Dictionary v1
One shared doc: every metric, its canonical formula, owner, and source. The social contract. The end-state is a governed semantic layer where each formula is defined once, in code, and every dashboard — and every AI copilot — inherits it by default.
Artifact 3 — Team Assessment
An honest read on the people and where institutional knowledge lives, so none of it is lost as the function evolves. Understanding, not verdicts.
Artifact 4 — Hiring Plan
The founding team sequenced, roles ready to post. The first hire matters most — a second pair of eyes on every later decision.
04What this requires from leadership
The playbook only works if a few non-technical conditions hold.
- Alignment behind the canonical definitions. Disputes get resolved by updating the shared definition, not relitigating it each time.
- A clear boundary on instrumentation. The analytical surface and product instrumentation, agreed once — not a turf negotiation every few weeks.
- Latitude to defer infrastructure decisions. Vendors will pitch early; the call comes when the data supports it, not before.
05The questions every rebuild has to answer
No two companies answer these the same way. They’re shaped by what the first weeks find.
- The actual cloud footprint. Map it before recommending anything. A strategic preference, or whatever's cheaper and fits the team?
- Event tracking and instrumentation. How do you standardize event design across teams?
- BI tool consolidation. Usually several tools in use; the default is one, so the team goes deep instead of broad.
- The AI layer. Where copilots and natural-language querying sit — and what has to be true (governed definitions) before switching them on.
- Central vs. embedded analysts. Start central while the silo problem is live; embed once the dictionary holds.