Who builds it, who it serves
This way of working is not "AI replaces the analyst". It is close to the opposite: the people who understand the data write down what they know, once, and everyone else gets to stand on it.
The data team builds and owns the context
Metric definitions, warehouse knowledge, connections, conventions. Expert work that stays expert work.
Everyone else works inside it
Product, operations, commercial — answering their own questions within your definitions rather than around them.
A data-autonomous organisation
Answers multiply without a queue, while correctness stays owned and taught by the people who understand the subject.
1. The data team builds and owns the context
Metric definitions, warehouse knowledge, connections, conventions, and the reports that matter are built and maintained by the people who are professionals at it.
This is expert work and it stays expert work. It is the part that does not get delegated to a model, because being wrong here is both invisible and expensive: a subtly incorrect definition of "active customer" does not throw an error, it just quietly misinforms every decision made downstream of it for a year.
2. Everyone else works inside it
Product, operations, commercial — people who are not analysts and never will be — open the same repository as context in their own AI assistant and answer their own questions.
The important part is what they are not doing. They are not guessing which table to use, not inventing a definition of revenue, and not waiting three days for someone to run a query. They are working inside the definitions your team wrote, whether they realise it or not.
Tip
This is the difference from self-service BI. A query builder hands the hard part — knowing which data to trust — to the person least equipped to judge it, and the organisation quietly accumulates wrong numbers. Here the expertise is written down and applied upstream, so a non-specialist's question is answered through your team's judgement rather than around it.
3. It compounds into data autonomy
Most questions get answered without joining a queue. The ones that recur get promoted into reviewed, scheduled reports. Every question that exposes a missing or ambiguous definition sends someone back to improve the context, which makes the next answer better.
Autonomy without a free-for-all: the answers multiply, while correctness stays owned, curated and taught by the people who understand the subject.
What each group needs to know
| Who | What they do | What they need to learn |
|---|---|---|
| Data team / analysts | Build the context; write reports for anything recurring | The framework's data helpers and report components |
| Product, ops, commercial | Ask their own questions against the repository | Nothing about trellum — the context files do the work |
| Whoever owns the platform | Connect the repository to a studio so reports reach everyone | The portal |
Where teams get this wrong
Writing reports before definitions. An assistant with good definitions and no reports will write a good report. One with ten reports and no definitions will confidently reproduce whatever the last report happened to assume.
Treating the context as documentation. It is not a wiki nobody reads — it is the input to every answer produced in the repository. Stale definitions cause wrong numbers, so they deserve the same review as code.
Opening it up before it is ready. Invite the rest of the business in once the definitions cover the questions they actually ask. Too early and you have recreated self-service BI with extra steps.
Next
- The analytics repository — what to write down, and what we actually require