There is a common refrain that AI requires high-quality data to deliver high-quality results. “Garbage in/garbage out” refers to the idea that any AI trained on less than perfect data will not be able to produce valuable outputs.
Before I get accused of dismissing the importance of quality data, high-quality data does produce the best results. Great data is definitely better than spotty data.
But the reality is that between signal loss, missing audience information and aging data, most of the data that companies have is spotty. The good news is that this data is actually far more usable than most people think.
Building around the data you have
Despite what you may have heard, agentic AI provides an effective solution for dealing with spotty data.
Think about how a human analyst handles messy information. Hand them a spreadsheet with inconsistent formatting, a dashboard from a different platform and a CSV that’s missing half of its columns and a good analyst will figure it out. They’ll recognize that “revenue” in one system means the same thing as “net sales” in another. They’ll spot gaps, adjust their confidence accordingly and still deliver a useful recommendation. They don’t need every system to speak the same language or every field to be perfectly populated.
AI agents work the same way. Rather than relying on rigid, programmatic connections that break the moment a schema changes or a field is missing, agents build a semantic understanding of the data they encounter. They grasp what the data means, not just where it sits. And they understand workflow intent (e.g., what you’re actually trying to accomplish), which allows them to reason through imperfections rather than choke on them.
An agent can look across disparate data sets, recognize that two differently formatted tables are describing the same customer behavior and knit them together without requiring a monthslong data engineering project to make the schemas match first.
This is fundamentally different from traditional analytics and automation, which depend on clean, standardized inputs connected through brittle pipelines. When a field changes or a data source degrades, the pipeline breaks. But agents adapt just like a human would.
Even spotty data is worth it
Across nearly every application in our industry, from audience targeting to media buying to closed-loop measurement, AI agents consistently provide a performance boost, even when they’re working with spotty data sets.
The reason is that agents apply sophisticated analytical techniques at a scale and frequency that human teams simply can’t match. They compensate for gaps in the data by being experts in the predictive methods themselves, knowing when to adjust confidence, when to lean on priors and when to flag uncertainty.
Take measurement. Without agentic AI, approaches like MMM, incrementality testing and attribution are expensive, manual and time-consuming. Most brands can’t perform this level of measurement at all, and those that do manage it only a few times a year. They use the data they have, imperfect as it is. The results are still valuable.
Agentic AI allows advertisers to perform that same sophisticated measurement far more frequently. The common fear is that putting AI on top of imperfect data will somehow amplify the imperfections. The fear is that one expensive, carefully managed study per year is safer than frequent automated calculations running on the same flawed inputs.
But this logic is backward. More frequent measurement on imperfect data actually reduces risk, because you’re course-correcting continuously rather than placing one big bet on a single annual snapshot.
Think of it like navigation: Checking your GPS every 30 seconds with some signal noise will keep you on course far more reliably than checking it once an hour with that same noise. Each individual reading might be slightly off, but the pattern of frequent reads keeps you headed in the right direction.
The same principle applies to media measurement. Frequent, AI-driven reads on imperfect data will outperform rare, manual reads on that same data every time.
Rather than wait for impossibly perfect data, start applying agentic AI to the data you’re already dealing with. The brands that get an early start will lay a foundation for better results, and they’ll get those results faster than brands that sit around waiting for the data to be ready.
“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Follow Newton Research and AdExchanger on LinkedIn.
For more articles featuring Matt Emans, click here.
