The work was never abstract. It was getting organizations to act on models they had already built and trusted in theory — but kept overriding in practice.
Rohan Talati spent twelve years inside that problem. As Executive AI Leader at Macy's, he led customer and marketing data science across retention, personalization, and experimentation at scale. At Citi, as VP Analytics, the scope covered acquisition, lifecycle, and pricing decisions. Neither role was advisory. Both required building the systems, getting them deployed, and staying close enough to the organization to understand why value wasn't translating — even when the models were accurate.
The pattern was consistent: teams had data. They had models. The analysis was correct. And yet the decisions the business made on Monday morning looked the same as they had before any of it existed. The model output and the business action occupied different worlds with no reliable mechanism connecting them.
Most AI engagements fail at the seam. A strategy consultant shapes the vision and moves on before it's tested. A data scientist builds the system without enough access to the organizational dynamics that will determine whether it gets used. The value that was promised never arrives because nobody held both levels at once.
Audiences AI engagements operate at two levels simultaneously — not alternating between them, but holding both throughout.
Every engagement starts with one decision — a specific, recurring business decision where data exists but the logic connecting it to action does not. The full sprint model is outlined in the Framework page.
The organizations with the most sophisticated AI are rarely the ones getting the most value from it. That observation repeated too consistently, across too many companies, to be explained by model quality or data maturity.
The gap is structural. Inside most organizations, there is no function whose job is to define how a model output becomes a business decision. Data teams build models. Business teams make decisions. The logic in between — given this output, here is what the business does next, and here is how we know if it was right — has no natural home. It sits between disciplines, between planning cycles, acknowledged by both sides and built by neither.
Audiences AI was built around that gap. Not to replace the models or the teams — both exist and both matter. But to design the layer between them: the decision logic, the action structure, and the feedback loop that makes the investment compound over time rather than sit in a dashboard.