Senior agency executives told the Programmatic IO conference in New York on 28 September 2026 that client data, and any AI model trained on it, should belong to the brand and not to the agency. They also expect headcount-based fees to give way to outcome-based pricing and software licences (AdExchanger, 1 October 2026).
Agentic AI is software that can plan and carry out multi-step tasks, such as media planning, buying or reporting, towards a goal it has been given. Agencies now build these tools for individual clients, which raises a practical question for every advertiser: who owns the tool, and who owns the data that feeds it?
Who spoke on the panel?
The session was moderated by Sarah Sluis, Editorial Director of AdExchanger. The four speakers were:
- Sir Martin Sorrell, founder and executive chairman of S4 Capital, the owner of Monks
- Bob Lord, president of Horizon Media
- Maggie Summers, chief experience officer at Dentsu X
- Shane McAndrew, chief data and technology officer at WPP Media
Why are agencies stepping back from owning data?
The speakers broadly agreed that holding data causes an agency more trouble than benefit. Sorrell's argument was that an agency which owns the data can be accused of marking its own work. He also pointed to a Dentsu forecast that algorithms will direct 75% of advertising spend by 2028 (AdExchanger, 1 October 2026), which in his view raises the pressure on agencies to be transparent and neutral in how they choose data partners.
That position runs against the recent pattern of holding companies buying data and technology businesses. Publicis acquired Epsilon in 2019 and plans to acquire LiveRamp. Asked what he would tell a client about LiveRamp once that deal completes, Sorrell said he would point them to the strongest alternative available.
Who owns an AI model built on client data?
At Horizon Media the answer is the brand. Lord said that on the agency's blu. AI platform the data belongs to the client, as does every model built on it. His reasoning is practical: if the relationship ends, the agency has nothing it can take with it.
Lord added that he has seen separations in which an agency carried a brand's data over to a competitor's campaign. He has also seen agencies treat AI tools built on a client's data as their own intellectual property and reuse them elsewhere, a practice he described as unacceptable.
What does AI change for DSPs and for the agency model?
A demand-side platform (DSP) is the software advertisers and agencies use to buy ad inventory automatically. According to McAndrew, the users of DSPs are increasingly machines: agents talk to other agents and return intelligence faster than a human operator could extract it.
He sees the same shift as a risk for DSPs. AI makes it easier for a buyer to take strong data signals directly from SSPs and publishers, bypassing the DSP's data marketplace, and he expects it to speed up the long-running effort to shorten the supply chain.
Sorrell turned to the agencies themselves. He regards Meta's plan for a single platform covering creative, planning, buying and distribution as a threat to the survival of the enterprise agency model. His forecast is that agencies will work closely with the end-to-end systems of the largest technology companies and take on the role of verifying what those systems report.
What could replace FTE pricing?
FTE (full-time equivalent) pricing is the model in which an agency's fee is set by the number of people assigned to a client account. Three models were discussed:
| Model | What the fee depends on | Position on the panel |
|---|---|---|
| FTE | Number of people on the account | The current default. Lord said inertia around this model slows the move to newer ones |
| Outcome-based pricing | Results the campaign delivers | Backed by Sorrell. Summers said Dentsu runs FTE and outcome-based models side by side |
| Software licence | Access to an agentic interface the client logs into | Backed by Lord |
McAndrew cautioned against treating automation as a way to offer savings through smaller teams. High-volume, low-value tasks will be automated, he said, while 95% of WPP's attention goes to the intelligence that drives, measures and validates client growth.
Which five questions should a brand in the Gulf ask its agency?
For an advertiser in the GCC reviewing an agency contract that involves agentic AI, the panel's arguments reduce to five questions:
- Who does the contract name as the owner of the first-party data used in campaigns?
- Who owns a model or agent built on that data, and may it be used for another client?
- How are the data and the model handed over or deleted when the relationship ends?
- Is the data partner the agency recommends owned by its holding group, or independent?
- Is the fee tied to headcount, to outcomes or to software access?
Key points
- Executives from S4 Capital, Horizon Media, Dentsu X and WPP Media described agency ownership of data as a liability at Programmatic IO on 28 September 2026.
- Horizon Media stated that on its blu. platform the brand owns both the data and the AI models built on it.
- WPP Media expects machine-driven use of DSPs to grow and AI to accelerate supply-chain compression.
- Outcome-based pricing and software licences were put forward as alternatives to FTE fees.
Source: Anthony Vargas, Why Agencies Think They Shouldn't Own Agentic AI Tools Or The Data Used To Build Them, AdExchanger, 1 October 2026.