Gus Dantas

Gus Dantas, retail media measurement

Gus Dantas

Eight years in marketing analytics in Australia, most of it agency side, working across media agencies for advertisers who had to justify what they spent. Retail media came after that, and it is the first time I have been on the side producing the numbers rather than checking someone else's.

At PetAds I built the measurement framework the network runs on. Campaign, sale and audience data land in one hub, the frameworks there decide what counts as incremental, and the insights come back out by category, by supplier and by campaign. Off site, on site and in store all report into the same place, so the network gets measured as one system instead of three.

The newer part is agentic. I build agents and MCP integrations that read the advertising platforms directly and assemble the post campaign reporting someone used to put together by hand, so the person arrives where the judgement is. This site was built the same way. The film, the writing and the code were made with an agent working next to me, which seemed like a fairer demonstration than describing it.

  1. 2018 Switched On
  2. 2020 AKQA Media
  3. 2022 The Media Precinct
  4. 2025 oOh!media
  5. 2026 Petbarn PetAds

Work

Microsoft FabricOneLakeIncrementality

The platform the network reports against. Lakehouse architecture on Microsoft Fabric and OneLake, bronze through gold. Building rather than renting, so the method is ours to publish and to defend. Onsite, offsite and in store land in one model against one set of definitions, which is the part that makes an iROAS figure defensible rather than merely produced.

This work involves retailer transaction data, so the figures stay private. The architecture is described above.

ClaudeMCPAgent orchestrationAd platform APIs

Post campaign reporting used to be days of manual pulls across six advertising platforms before anyone could look at a result. It now runs as a supervised pipeline. Claude agents and MCP integrations gather and assemble, a person reviews at the decision point, and the report exists while the campaign is still worth discussing.

The automation is not the interesting part. What matters is what a shorter reporting cycle does to how often a team can afford to run a proper experiment, which is the argument in piece 03 above.

PythonRSQLTableauPower BIRegressionClusteringSegmentation

Five engagements from the agency years, before retail media. Kept for depth, not as current positioning.

BNPL revenue forecast 2021
Historical data and regression producing scenario based recommendations for a buy now pay later platform. The modelling was the small part. Getting the business to trust the output was the rest.
Mortgage customer segmentation 2021
Algorithmic segmentation feeding creative targeting for a home loans broker. The segmentation that ran as live media rather than the one that lived in a slide.
Scenario planning tool 2022
Regression powered planning that extended the segmentation, so the business could test a media investment before committing the spend.
Cart abandonment clustering 2021
Clustering applied to abandonment behaviour at an office and stationery retailer. Four abandoner personas, each with its own recovery lever.
Creative performance, global FMCG portfolio 2020 to 2022
Ongoing campaign reporting across a large SKU portfolio on tight deadlines. The unglamorous work that teaches you to ship.

Email linkedin.com/in/gus-dantas