AI Is About to Rewrite Audience Targeting
Last week I wrote on LinkedIn about wrapping five months at Arrivalist, where I built a travel audience business from scratch. This post is what that experience taught me about where the industry is going — and a few things I didn't expect.
A small story about big data
We uploaded 400 travel audience segments into one of the largest DMPs in the world.
Then I went to find them.
A search for "beachgoers in West Palm Beach" returned nothing. Zero. We had a Florida beachgoers segment with West Palm Beach right there in the metadata. Turns out the search was pure Boolean exact match — no fuzzy logic, no inference, no semantic understanding. You had to know exactly what to type, down to the character.
This wasn't a scrappy startup. This was a category leader.
That moment stuck with me. Because if navigation breaks at 400 segments, what happens at 400,000? What happens when you're reconciling behavioral signals across channels, device graphs, offline transactions, and a compliance layer that changes by geography? The data problem isn't volume. It's navigability. And the tools we built for a simpler era weren't designed for this.
This is no longer a segmentation problem. It's a systems problem.
Where I think AI fits — and I say "think" deliberately
I'm not an AI expert. I want to be clear about that, because LinkedIn has roughly ten thousand of those already.
What I am is someone who just spent six months building an audience product and used LLMs practically throughout — generating metadata at scale, filtering for regulatory compliance, speeding up research that would have taken weeks. Not magic. Just useful, in the right spots.
And that experience gave me a specific point of view: AI isn't an add-on to the existing audience workflow. The existing workflow — build a segment, push to activation, measure, repeat — was designed for a world where data was sparse and static. That world is over. The new one needs infrastructure that can reconcile fragmented identity, model intent from behavioral data, and adapt continuously. The old tools were bolted together to approximate that. AI is the first thing that can actually do it.
What changes when that infrastructure works? The human role shifts. Not out — sideways. Instead of tactical assembly, the job becomes strategic inquiry: where is the next marginal dollar most effective — advertising, discounting, loyalty, fulfillment, support — and how does that decision propagate across channels? Those are questions we've never had the operational capacity to ask seriously. That's the real unlock.
The part that keeps me up a little
There's a risk in this future that doesn't get talked about enough.
At my first company — an SEM agency I co-founded — we built increasingly sophisticated bid management algorithms. They worked. Performance improved. And then something quietly broke: we stopped understanding why. Was it the ad copy? The landing page? SERP position? Time-of-day patterns? The algorithm outperformed our manual work, but the transparency evaporated. We couldn't pull the thread anymore.
As AI takes over the operational layer of audience targeting, that dynamic will return at much larger scale. The analyst who used to catch the weird pattern at 11pm in a spreadsheet — the anomaly, the spike, the question nobody thought to ask — isn't in that spreadsheet anymore. The risk isn't replacement. It's losing the human instinct that generated the best insights in the first place.
That won't fix itself. It'll need to be designed for.
Where this goes
The destination is audience systems that actually match the complexity of real human behavior — dynamic, cross-channel, compliance-aware, continuously learning. Not a predefined segment pushed to a DSP. Something that evolves with the customer.
We're not there yet. The gap between where the infrastructure is today and where it needs to go is real — and I saw it up close.
If you're working in this space — audience modeling, identity, activation, clean rooms, or the connective tissue between any of these — I'd love to compare notes. I'm just getting started on this problem, and it's the most interesting one in digital advertising right now.
