Compare the Market cuts report delivery from 6 weeks to under 1 week

From reporting factory to governed models

Emanuela Ciotti, head of enterprise analytics at Compare the Market, presented a framework drawn from 7 years of working with dbt across 3 different organizations. She began with a warning about AI: it does not validate the data it consumes, it trusts it. A weak data foundation is therefore not exposed by an AI layer on top, but amplified by it.

She described the starting point as a "reporting factory": a BI team acting as a service desk, SQL logic duplicated across reports and tools, and metrics defined differently from one dashboard to the next. An internal audit of the company's Power BI estate found 433 reports in total. Of these, 41% (179) were already abandoned. Each had been built as a separate, one-off answer to a separate question, with no shared foundation underneath.

Three stages: discovering, scaling, transforming

Ciotti's framework splits the change into 3 stages.

• Discovering: the shift from analyst to analytics engineer. Version control, modular design and testing are introduced before any new tooling.
• Scaling: addresses the problems of a single monorepo with more than 4,000 interdependent models. Part of the staging layer depended on Google Sheets seeds, which occasionally broke financial pipelines right before month-end reporting.
• Transforming: the stage her team is in now. It redefines what the analytics function produces, from dashboards to tested and documented data models, and positions that data layer as the foundation AI needs to be trustworthy.

To fix the monorepo problem, the team adopted a multi-repo architecture aligned to domain ownership. It draws on dbt Mesh principles and gives each business tribe ownership of its own models.

The results

The figures Ciotti shared, as reported by dbt Labs, were these: the time to deliver a new report fell from 6 weeks to under 1 week, test coverage passed 75%, and model reuse climbed to 3.0x.

The source does not give a baseline for test coverage or model reuse, so the scale of those improvements is not stated.

What it means for analytics teams

The dbt Labs write-up frames the story as "fundamentally an operating-model change, not just a stack change." In its reading, a team moves from service desk to platform only when what it hands the business changes from a dashboard to a governed model. That, it says, requires designing ownership in from the start rather than assuming it will emerge on its own.

For analytics teams under pressure to add AI, the case points to the data layer as the first job. The source argues that governed, tested models are what make AI output trustworthy, while dashboards built one at a time leave metrics inconsistent. Note that the account comes from dbt Labs, a vendor, and relays results reported by Compare the Market's own team.