Data before automation: Climbing the AI maturity curve with Adomik’s Jeff Goldstein
BY ROB BEELER + JEFF GOLDSTEIN, MANAGING DIRECTOR (NORTH AMERICA) OF ADOMIK
AI has arrived in ad ops with some big promises. Smarter analytics, faster teams, and data that finally pulls its weight. But from where we’re sitting, publishers are finding that plugging AI into the ad stack is the easy part. The hard part is trusting that what comes out the other end is accurate, especially when every platform describes the same campaign, buyer, or revenue event in its own way. And with pressure on everybody to automate more, and faster, it’s easy to lose sight of whether your team is actually ready.
In this Q&A, Rob joins Jeff Goldstein, Managing Director, North America at Adomik, to explore the gap between the promise and the outcome. For instance, what makes data decision-ready? How can you tell where you sit on the data and analytics maturity curve? And where do you draw the line between what you automate and what you leave to human judgement?
Jeff also shares how Adomik’s AI teammate, JP, fits into a normal working week.
Rob: Publishers have heard a lot about what AI can do for data, analytics, and monetization. Where’s the gap between that promise and what teams can actually trust enough to use?
Jeff: The biggest gap is between giving AI access to data and giving it data it can actually understand. A publisher may connect an LLM to its ad server, SSPs, OMS, CRM and analytics tools, but those systems often describe the same buyer, campaign, or revenue event differently. An LLM can summarize what it receives, but it can’t decide which definitions are correct. That’s how you get an answer that sounds convincing but rests on inconsistent assumptions.
Rob: If a publisher is making revenue decisions from AI-generated analysis, what has to be true about the data before that analysis is actually useful?
The data has to be decision-ready. Sources need to be cleaned and normalized, entities mapped across systems, and metrics like revenue, CPM, fill, and delivery defined consistently.
That’s where our experience in publisher monetization matters. We’ve codified that complexity into a trusted data foundation built around publisher-specific business logic. Controlled analytics and data-science layers handle the calculations, while generative AI interprets the validated results and adds context.
Rob: You talk about a data and analytics maturity curve. What does it measure, and how can a publisher tell where they are on it?
Jeff: It’s a way to understand what a publisher can reliably do with its data today. It starts with making raw data available, then preparing it through cleansing, normalization, mapping, and enrichment. From there, publishers can connect datasets for consistent analysis, use AI to generate trusted recommendations, and ultimately turn those into controlled, automated actions. Each stage builds on the one before.
Publishers often sit at different points across the business. Yield might already use intelligent recommendations while sales still reconciles reports by hand. So the best approach is to assess maturity by team or workflow. How much preparation does the data still need? Does the team trust the recommendations? Is it ready to automate approved actions?
Rob: How should knowing where you are on the curve change what you automate now versus what you should hold off on?
Jeff: I’d start with the biggest constraint. If a team spends hours reconciling data, automating that preparation creates the most immediate value. From there, teams can move into reporting, anomaly detection, and diagnosis. Then they can move onto recommendations, and then execution, once the right permissions and controls are in place.
Rob: There’s a lot of pressure right now to automate more, faster. Where should publishers keep human judgment or workflow control in the loop?
Jeff: Human judgment remains essential wherever a decision has a meaningful effect on revenue, clients, inventory quality, or user experience. AI might spot a pricing opportunity, but the team understands the client commitments and market conditions that should shape the response. Routine, reversible workflows can run more independently, while consequential decisions keep clear approval and accountability.
Rob: AI vendors are making some very big promises. What should publishers ask before trusting a platform to analyze performance or act on their behalf?
Jeff: Start with the data. How is it collected, cleaned, and connected, and what happens when two systems define the same metric differently? Then ask what’s doing the analytical work, from business rules to machine learning or generative AI, and how results are validated and monitored over time. Before a platform acts, understand who approves what, what audit history exists, and how actions can be reversed. A vendor should be able to show the full path from source data to outcome.
Rob: Adomik’s Jean-Pierre (JP) is designed to work more like a teammate than another reporting tool. What does that look like during a normal week?
Jeff: JP works from the foundation created by Adomik Data Engine, which prepares and connects publisher data before JP’s specialized agents use it. A team might ask JP for its weekly revenue recap. If something unexpected appears, JP can investigate what changed, identify the contributing factors, and add market context, then turn that into a leadership summary, client deck, or optimization recommendation. Once a workflow is established, Autopilot can run that analysis on a schedule or when defined conditions are met.
Rob: You can also bring JP’s agents into tools teams already use. Why does that help adoption?
Jeff: Teams can reach the same intelligence wherever they work best, whether that’s the JP application, an enterprise AI environment, or an internal automated workflow. The agents rely on the same prepared data, business logic, and access controls throughout, so teams using JP gain flexibility without creating different versions of the analysis.
Rob: Once a publisher has a trusted data foundation, what’s the smartest next step for turning that readiness into better monetization?
Jeff: Start with one high-value, repeatable workflow, such as explaining weekly revenue movement, detecting anomalies, or monitoring deal performance. Map that workflow from the underlying data to the business outcome and identify where it breaks down. Publishers become more data-driven by building this progression one use case at a time and applying what they learn across the business.
Ready to make your data AI-ready?
We’re seeing plenty of publishers chase AI’s big promises, but the ones getting real value started with data they could trust. As Jeff explains, that means making data decision-ready, knowing where each team sits on the maturity curve, automating only what it can support, and keeping people in charge of the bigger calls. Ask vendors the right questions, and what comes out the other end becomes something you can act on with confidence.
Find out how Adomik’s JP can help your team move up the maturity curve. Book a demo today at adomik.com/request-a-demo.
👉Want to stay in the loop with what’s happening in the Beeler.Tech community? Subscribe to our newsletter. If you’re interested in attending or sponsoring a future event, you can explore our upcoming events here.