If In Doubt - Verify! What the WPP case and Sony probe tells us about the operating model!

It seems Everyone in marketing is talking about Sony and WPP. Richard Foster’s amended complaint claims that Sony launched its own investigation into WPP’s media practices after Chinese authorities convicted three former GroupM China executives over a kickback scheme. The filing alleges that Sony’s investigators found hundreds of millions in client rebates had been wrongly retained. WPP says the claims are baseless and is moving to dismiss them. The allegations are serious, but the discussion is already disappearing down familiar rabbit holes: it is China, it is one agency, it is one case, it is simply how things are done. This misses the point, rather than being just another scandal to hit agencies everyone can ignore we need to look at the fundamentals that apply to nearly all enterprise client relationships with hold co agencies and consider if they are fit for the AI era.

If you want to read about the Sony story particularly. The original story by Lara O'Reilly is here https://www.businessinsider.com/lawsuit-alleges-sony-investigation-into-wpp-rebates-2026-8 and Omar Oakes great take here https://www.moreaboutadvertising.com/2026/08/omar-oakes-what-sony-found-when-it-stopped-trusting-wpp/

As enterprise advertisers move into 2027 planning, they need to ask themselves three questions:

If a large proportion of my media spend is going through agency principal media, can I be completely confident it is better for my business than the available alternatives—and can I validate the value and margin? Do I have complete transparency across media planning, buying, delivery and measurement, at the platform and line level my own AI will need to learn and make better decisions? Am I judging success using outcomes, and can I be completely confident those outcomes cannot be selected, measured or gamed to increase agency margin?

If you are not confident about all three, you should read on. The first part of this article explains why the problem becomes more dangerous in an agentic era. The second sets out what advertisers can do about it: the detail they should demand in their plans, the commercial and measurement service levels they should put in place, and the intelligence they need to own. There is something here for publishers and other suppliers too, because opaque buying does not only affect advertisers. It also determines which media gets selected, how its value is represented and who captures the return.

I hate to tell you this, but it is my opinion that the reality is worse than many advertisers think. This was a big part of the conversation up and down the Croisette in Cannes and continues to be discussed in rooms from London to New York, never mind Shanghai. The consistent view is that these practices are widespread, bad for advertisers and ultimately bad for agencies because they destroy trust. Crucially, I believe the problem is getting worse. In the post-pandemic years, agencies have lost some of their traditional trading cash cows in television and out-of-home as digital has surged forward. Some agencies are building the technology, talent and services they need for the future. Others appear more focused on finding new ways to replace lost margin and boost next year’s profits. Take that as opinion, but it is something every enterprise advertiser should verify for itself.

Agencies still make a great deal of money from deal-based media. Proprietary inventory, principal trading, rebates—call it what you like. The fee on the plan is not always where the profit is. The trade is where the profit is. None of this is automatically wrong. An agency can use its scale to secure better prices, take genuine financial risk and create real value. The problem is the information asymmetry. The agency knows what it paid, the margins available, the commitments made to media owners and the incentives attached to different buying decisions. The client knows what it was charged. Those are not necessarily the same number, and the gap between them remains one of the most profitable secrets in the industry. A plan may be brilliantly coordinated and bought, with the agency’s return fully earned. Or it may bear little resemblance to what the advertiser needs while generating the agency a lovely margin. Without transparency, the client cannot tell the difference.

At the same time, some major media plans are becoming remarkably light on detail: large numbers, broad channels and not much underneath. Advertisers should insist on plans that show the named platforms and media owners, buying routes, formats, audiences, geography, flighting, forecast delivery, unit costs, fees, measurement and whether each line is being bought as agent or principal. This does not mean freezing every decision months in advance. Agencies need room to optimise, but that freedom should operate within visible rules, with material changes recorded and actual delivery reconciled back to the plan at the same level of detail. A plan that says “£4 million, AV, H2” is not a serious foundation for accountability, and it is useless to an advertiser building its own AI brain. AI needs granular, accurate information to learn what worked, understand why and make a better decision next time.

That loss of precision becomes particularly dangerous in the agentic era. AI agents will not simply report on media decisions; they will increasingly make them. They will select audiences, move budgets, choose inventory, adjust bids, test creative and judge performance. If the data and commercial incentives beneath those decisions are opaque, AI will not correct the problem. It will automate it. The danger is not necessarily a rogue AI making inexplicable decisions. It is a highly effective AI doing exactly what it has been instructed to do while the advertiser cannot see whose interests its objective function serves. A world in which the answer to “Why did we buy that?” is “The AI said so” is not progress.

We are now seeing commercial deals in which enterprises are effectively asked to exchange transparency for access to agency AI. The pitch is: do not worry about the inputs, the media price or the agency margin; judge us on the outcome. I understand the attraction. Nobody actually wants impressions. They want the additional sales, customers and profit those impressions cause. Agencies are also being pushed towards output- and outcome-based billing because AI makes traditional time-based fees harder to defend. If work that once took days can be completed in minutes, clients will not continue paying for the hours that used to be required. This gives clients real leverage. If an agency wants to be paid for an output, the client can define the output, the evidence and the service levels required to validate it. Outcomes should not become an excuse for less transparency. They should require more.

The problem is that an outcome is not necessarily incremental. A sale recorded after an advert was served is not automatically a sale caused by the advertising. It might have happened anyway because of existing demand, distribution, pricing, promotion, seasonality or another channel. Attribution tells us where a sale was recorded. Incrementality asks whether the advertising created something that would not otherwise have happened. If an agency controls the media, the data, the outcome definition and the measurement, it is marking its own homework with millions in fees riding on the answer. If the deal works, fine: the advertiser gets what it wanted, the agency earns a good return and everyone is happy. If it fails, the advertiser may have no data with which to understand why because it surrendered that data at signing. That is where the comparison with The Big Short becomes relevant. Opacity allows incentives and risk to hide inside something that remains sophisticated and profitable for the seller right up until it stops working.

Solutions. 

Advertisers do not need to reject principal media, agency AI or outcome-based remuneration. They should attach clear requirements and service levels to them:

Plans should show spend at the relevant platform, media-owner and line-item level, including the buying route, format, audience, geography, flighting, forecast delivery and unit cost. Actual delivery should be reported and reconciled against the plan at the same level, including a record of material changes and the reason for them. The agency should disclose when it is acting as an agent and when it is acting as a principal. Media prices, margins, rebates and other incentives should be capable of independent validation. Planning, buying, delivery and performance data should remain available to the advertiser in a usable, machine-readable form. Outcomes, baselines, counterfactuals and measurement methods should be agreed before the money moves. Incrementality testing should be independent when agency remuneration depends on the answer. Important decisions made by people or AI agents should be recorded and open to interrogation. Data and decision histories should be portable, allowing the advertiser to investigate failure or change partners. Where agreed transparency, delivery or measurement standards are missed, there should be a clear commercial remedy.

Publishers and other suppliers also have something important at stake. In an agentic market, their media will increasingly be selected by systems optimising against rules and incentives they may never see. Quality inventory can lose to inventory that produces a better agency margin or satisfies a wider trading commitment. Publishers need to make the quality, provenance, audience, context and performance of their media easy for advertiser and agency systems to interrogate. They should also push for evidence that their value is being represented accurately to the client and resist structures that turn a differentiated audience into anonymous inventory inside somebody else’s bundle. Advertisers need publishers to be fairly rewarded for creating valuable audiences; otherwise the long-term result is poorer media, weaker evidence and an even more concentrated supply chain.

AI has huge potential to be a force for good and a force for ill in marketing. I founded TAU to push the industry towards the right side of that choice: using AI to make marketing better, more transparent and more accountable, rather than worse. Marketers should have their own intelligence—systems that interrogate plans, test assumptions, compare alternatives, identify gaps and hold agencies and platforms to account. The point is not to replace good agencies. It is to give marketers the evidence they need to recognise good work, challenge poor work and make better-informed decisions. If you want help, TAU can independently assess your 2027 media plan or help you create your own with full transparency. Every assumption, data source, recommendation and even the code itself can be interrogated if you want. We can do it at pace, saving your team time without asking you to surrender visibility or control.

If in doubt, verify.