I've spent the past seven years inside the programmatic supply chain — first at AppNexus building it, then at Xandr operating it, now at Microsoft trying to make it work for publishers in a world that's changing faster than most of the industry's infrastructure can keep up with.
Programmatic isn't going away. The core value proposition — using data and automation to match supply with demand efficiently — is sound. But the specific mechanisms, business models, and power dynamics are shifting significantly. Here's where I think things are heading.
The identity layer is being rebuilt
Third-party cookies were the plumbing that made cross-site targeting, frequency capping, and attribution work. With that plumbing being ripped out, the industry is building replacements: UID2, first-party data graphs, contextual enrichment, Privacy Sandbox APIs.
None of these individually replicate what cookies did. That's partly the point. The future identity layer will be a patchwork — deterministic where users authenticate, probabilistic where they don't, contextual where neither applies. At Prebid.org, we've been building the open-source infrastructure to support this heterogeneity. It's messy, but it's more honest about the reality of user consent than cookies ever were.
The supply chain needs to get shorter
The programmatic supply chain has too many intermediaries. A buyer's dollar passes through a DSP, an exchange, an SSP, maybe a reseller, before reaching the publisher. Each hop takes a cut and adds latency. I've seen publisher take rates below 40% of gross spend, which is unsustainable.
SPO (supply-path optimisation) is the industry's attempt to fix this, but it's only a partial answer. The structural fix is consolidation: fewer, more capable partners doing more of the work. Publishers working directly with DSPs. Buyers cutting out redundant exchanges. This is already happening, and it'll accelerate.
AI will change what gets automated
AI in programmatic currently means bid optimisation and audience prediction. That's table stakes now. The next wave is more interesting: LLMs enabling sophisticated contextual understanding (page-level semantic analysis rather than keyword matching), generative AI producing creative variations at scale, and AI agents negotiating deals programmatically.
The risk is that AI further concentrates power with the largest platforms who have the most training data. Open-source alternatives — the kind of thing Prebid.org champions — become even more important as a counterweight.
Retail media's gravitational pull
Retail media networks are the fastest-growing segment because they have what everyone else is losing: authenticated, first-party purchase data. Amazon, Walmart, Tesco — they can close the loop between ad exposure and purchase in a way that's increasingly difficult on the open web.
For publishers, this is a competitive threat. Advertising budgets are finite, and retail media is capturing an outsized share. The response has to be demonstrating unique value: premium brand environments, editorial authority, audience engagement metrics that retail sites can't match.
Measurement is the real battleground
Whoever controls measurement controls budget allocation. Post-cookie measurement is harder, noisier, and more fragmented. Attention metrics, incrementality testing, media mix modelling — these are replacing the deterministic attribution that cookies enabled.
I actually think this is a positive development. Cookie-based last-click attribution was precise but misleading — it over-credited bottom-funnel tactics and under-credited brand building. The new measurement paradigm is less precise but arguably more truthful about what advertising actually does.
What I tell publishers
Invest in first-party data relationships. Build direct connections with buyers who value your audience. Don't rely on a single SSP or identity solution. And start thinking about what happens when AI agents — not human users — become a significant source of your traffic, because that's coming faster than most publishers realise.