I keep seeing the same value proposition from AI companies in data-heavy verticals: "talk to your data."

Hundreds of billions in capex are flowing into AI with the promise of materially expanding GDP, and the best we can do is reactive report queries? I think it misses the point.

People don't want to talk to their data. They want intelligence. They want an agent that can analyze data at a depth no human could, surface what matters, and act on it autonomously. That's what they're hiring AI for. Not to replace the search bar in their BI tool.

The proliferation of “talk to your data” positioning begs the question of whether these companies actually listened to the problems their customers wanted solved, or whether they simply needed to check a box that they were using AI, and that was the easiest option.

We almost made the same mistake.

Gigi has a full-funnel dashboard built into our product. It's legacy tech: a restrictive dashboard with beautiful data visualizations, powered entirely by Amazon Marketing Cloud. We shipped it as table stakes in 2025 and intentionally haven’t invested much further in it since. Our customers enjoy it. It's valuable. And some agencies even grant their clients read-only access.

But customers kept making the same request: help me make sense of this.

That is true for most advanced measurement and analytics in advertising. The reports look great in demos, they're cool in theory, yet they go underutilized because customers struggle to synthesize insights into actions on their own.

So we started exposing the underlying reports in this dashboard, and the metrics within them, as tools Gigi could call. The intent was to bring these insights into the chats and tasks where customers actually do their work. As we were bringing this to market, two things stood out:

First, our lead account manager flagged that no one is asking to talk to these reports. My response: “Of course they aren't.” No one wants to talk to data, and no one is going to ask to.

Second, the customer comms drafted around the release framed every example prompt as a data query. Things like: "In my overlap report, which combinations of ad units are driving the most new-to-brand?"

This led us to pause the release. The framing was wrong. We weren't supposed to provide customers with a way to talk to a report. We were supposed to give Gigi a new tool to better support strategic recommendations.

We repositioned. Instead of letting customers query reports directly, we incorporated a new set of underlying skills that use these reports as tools that Gigi reviews when providing strategic recommendations. So when a customer asks Gigi a general question, like “how to increase incremental sales?” or “what are the top recommendations for a client where new-to-brand is the stated KPI?” Gigi pulls the relevant report, analyzes it, and returns specific budget allocation, audience, and media planning recommendations based on the insights. The customer never asks to talk to the report. They get the strategic output and the data work happens underneath.

This is where we want Gigi to go. We want customers using Gigi as an autonomous strategist, not a task executor. Central to that strategic layer is powering Gigi with the right tools and skills to take action on what the data actually says.

That's what our customers hire Gigi for. Not to ask her questions about data.

If you're an agency or brand evaluating agentic AI, don't grade vendors on whether their product can talk to your data. Grade them on what they do with that data autonomously.

Note: This was originally published as part of our Cherry Picked newsletter. Subscribe here 🍒