From Data Chaos to Confident AI: How Overdose Built a Semantic Foundation on Google Cloud

Most enterprises don't have a data problem. They have a meaning problem.

That's the framing Mike Ni, analyst at Constellation Research, brought to an interview with Paul Pritchard, CEO of Overdose, at Google Cloud Next. Throughout their conversation, Pritchard unpacked exactly what it takes to get from scattered data sources to AI that actually moves the business forward.

The answer, it turns out, starts well before the AI.

The Real Problem: Everyone's Fighting Over the Truth

Overdose works with merchants globally as a high-growth digital commerce agency operating across dozens of client relationships simultaneously. The challenge they kept running into wasn't a lack of data. It was that every team had a different version of it.

"With those disparate data sources comes a whole bunch of confusion," Pritchard explained. "Every channel claiming the sale meant that there was a fight every day to figure out what the truth was."

CFOs were looking at one number. Marketing managers at another. Board members at a third. And the agency was spending enormous energy just reconciling the differences before anyone could get to the actual work of growing the business.

The turning point came when Overdose built a semantic layer powered by Looker and Google BigQuery. Not another dashboard. Not another BI tool. A single, grounded source of truth that every team, client, and agency alike could work from.

"The ability to now sit down with a client and understand what the imperative metrics are (profit, effective movement of stock, customer acquisition that turns into lifetime value) and give everyone the same information at the same time has been so impactful," Pritchard said.

From Month-Long Dashboards to Real-Time Action

Before Looker, Overdose's process looked like this: spend a month building a dashboard, present it at the end of the month, and spend the meeting looking backward at what had already happened.

"You can imagine what you're doing at that stage," Pritchard said. "Looking back on the performance of that month and comparing it year on year."

Now, that same insight is available in real time. Teams don't wait for a monthly review to act. They build action-oriented plans based on what's happening now. And when an insight surfaces, generative AI is there to do something with it immediately.

Overdose built a tool that takes the semantic layer's output and runs it through an autonomous action loop: insight → creative brief → creative assets → published to ad channels → performance tracked back through Looker → new assets created when ready.

"Our merchants can move faster to engage with their customers," Pritchard said. "That's the best thing that can come out of it."

AI That's Actually Grounded

This is where the conversation gets pointed. Pritchard is not shy about his view on AI deployed without a solid data foundation.

"If it's not grounded in data, you get confident hallucinations," he said. "And if you get confident hallucinations, your brand is at risk."

It's a line worth sitting with. Confident hallucinations (i.e. AI that sounds certain while being wrong) are arguably more dangerous than obvious errors, because they're harder to catch and easier to act on. For a commerce agency making real-time decisions about campaigns, creative, and spend, that's not a theoretical risk.

This is also, Pritchard argued, why so many AI proofs of concept never make it to production. The semantic foundation wasn't there from the start. The AI had no solid ground to stand on.

"Intent is everything," he said. "A POC gives you some sort of intention. But the reality is, unless it's in the real world, connecting up all of those parts of your business to allow you to act on the fly in real time, that is the magic."

The People Question

One of the most common concerns when AI enters a business is what it means for the people doing the work. Pritchard's framing is worth noting.

"Instead of thinking about the reduction that everyone talks about, we think about the empowerment it brings to our teams," he said. "Our teams are now talking to clients about impactful metrics, rather than thinking about operational toolsets."

The shift isn't fewer people, but better conversations. Teams that used to spend time wrangling data are now spending time on strategy. And as the AI handles more of the execution layer, the human value moves further up the stack.

The Advice for Leaders Starting This Journey

Ni asked Pritchard what data and commerce leaders need to know before embarking on this path. His answer was direct: Start with the truth.

"What we knew is that disparate data sources leading to inconsistent reporting. Every channel claiming the sale meant there was a fight every day. When you get to that truth, what's beyond it is absolutely exponentially beneficial. Start by trying to get to the truth. Once you've got that, anything's possible."

For Overdose, the partnership with Google Cloud and the GCP team was central to getting there. Looker's price point made it accessible to the mid-market merchants it serves. And the consistency it created across every client relationship became the foundation for something bigger: an agency whose institutional knowledge and context compound over time.

"The information we retain over time should be able to be used to continually inform, elevate, and create innovation for those businesses," Pritchard said. "Through our partnership with Google Cloud, we've been able to do that in a consistent way — generating more revenue, more margin, and more customer acquisition for our merchants."

Watch the Full Interview

Catch the full conversation between Mike Ni and Paul Pritchard from Google Cloud Next below.

👉 https://youtu.be/w-0yDAaqyXQ