From Fragmented MMM to Decision AI [Video]

When the Advertising Research Foundation invited our founder Bharath Gaddam to present at its Insights Studio, he opened with a question to the room: how many analytics models does your organization use to plan growth? The most common answer was four to six. That is the problem in a single number.

In this 48-minute session, From Fragmented Marketing Mix Modeling to On-Demand Decision AI for Enterprise Growth, Bharath walks through why enterprise growth decisions break down, and what it takes to fix them.

The argument

Large brands measure everything and still cannot see one truth. Marketing, sales, trade, shopper, and forecasting each run their own model, and each reports a "silo truth" that never reconciles. The deeper cause is not the data. It is the learning system. Correlation-based models handle one KPI at a time, cannot manage hundreds of interacting drivers, and miss the synergies between them. They describe what happened. They do not tell you what to do.

Bharath makes the case for moving from correlation to causation at scale: a transformer-based causal architecture, built on Judea Pearl's causal ladder, that learns every KPI, driver, and time horizon together in one model. That architecture is FOUNT, and the model built on it, POEM365, is pretrained on $5 trillion of transactions across roughly 15,000 brands.

The case study

The center of the session is a real Fortune 500 story. A $2 billion CPG brand called Data Poem after losing 10% of sales, with $400 million spread across product, media, shopper, and trade. They had been planning against a single blended ROI. Learning the drivers synergistically rather than in silos surfaced 11% of incremental sales they could not previously see, with the strongest effect running from media into trade. The model reached 94% forecast accuracy while accounting for external forces from GLP-1 drugs to weather and inflation.

Also covered

• Why Data Poem's model beat the M5 forecasting benchmark, a result still unbeaten in the published rankings, and outperformed time-series models from Google, Salesforce, and Amazon
• How "think as one, act as one, grow as one" works in practice
• A live audience Q&A on data onboarding, regional granularity, new channels, and how teams actually adopt this

Who should watch

Anyone who plans or measures growth at an enterprise, and anyone tired of leaving the planning meeting with five answers and a gut call.