How do you let analysts work freely without ending up with five conflicting versions of the same number?

What is being tested

Whether you resolve this with tiers rather than with restriction.

Why restriction fails

Restricting access does not remove the need for the analysis; it relocates it. The analyst extracts to a spreadsheet, computes there, and the result has no governance at all — no lineage, no versioning, no access control — and is emailed to people who treat it as authoritative.

The governed platform loses visibility of exactly the analysis it was trying to govern.

The resolution: tiers, not gates

Tier 1 — certified. Owned, tested definitions, monitored freshness and quality. Used for board reporting, regulatory submissions and operational decisions. Changes managed.

Tier 2 — curated. Cleaned, modelled, documented datasets without tier 1 guarantees. Suitable for most analysis.

Tier 3 — raw and exploratory. Anything, for those who understand it. No guarantees, clearly labelled.

Analysts work freely in tiers 2 and 3. The only rule is that anything presented as an official number must come from tier 1 — a labelling requirement rather than an access restriction.

What makes it work

The certified path must be the easiest path for common questions. If a governed number takes three days and computing it yourself takes an hour, people compute it themselves — and they are right to.

A promotion path from exploratory to certified, so a useful analysis can graduate rather than remaining unofficial forever.

Tiers visible in the tools themselves, so a consumer sees which tier a number came from without asking.

Lineage everywhere.

The signal that the balance is wrong

Shadow analysis — spreadsheets circulating with numbers nobody can trace.

That is not a discipline problem. It is evidence that the governed path did not serve a real need — the same signal as teams routing around an internal platform, and it should be treated the same way: as a product finding rather than as non-compliance.