When AI Changes Who Gets Paid

Agency economics, advertising budgets and the growing value of helping AI agents choose

The next disruption to advertising may arrive without a collapse in advertising expenditure. Brands may continue to spend more while paying less for some of the work that has historically supported the agency holding company model. At the same time, a growing part of the buying decision may take place in systems that conventional media plans barely address.

AI can reduce the labour required for particular marketing tasks, while creating work in review, data preparation and systems integration. Shopping agents introduce another point at which a brand must establish that its offer meets a customer's needs. These developments raise a question that reaches beyond agency efficiency: how should a business divide investment between creating preference and helping turn that preference into a verified buying decision?

The evidence supports changing agency remuneration and pressure on some incumbents. It does not yet establish that agentic commerce is causing broad cuts to traditional media or industry-wide margin compression. The opportunity examined here is a prospective redistribution of marketing budgets—and the conditions under which it would make commercial sense.

Here, agentic commerce means AI systems performing shopping tasks on a customer's behalf: comparing offers, checking suitability, recommending an option or helping execute a purchase. Ordinary AI research and referrals provide adjacent evidence of changing behaviour; they do not establish autonomous purchasing.

The holding companies face a transition, with sharply different results

Describing the holding companies as uniformly distressed misses important evidence. WPP's third-quarter 2026 revenue less pass-through costs declined 2.1% like for like. Its full-year headline operating margin guidance remained 12–13%, with second-half margins expected to fall by up to approximately two percentage points year on year. WPP attributed that expected pressure to investment phasing, new-business ramp-up and incentives. These figures establish pressure; they do not isolate AI as its cause. [1]

Publicis, by contrast, reported 5% organic net revenue growth in the same quarter and maintained guidance for a slight improvement on its 18.2% operating margin in 2025. Omnicom reported 7.2% organic growth in core operations and an 18.2% core adjusted EBITA margin. These are differently defined measures and should not be read as a directly comparable margin league table. [2][3]

The divergence matters. It is consistent with a contest over business models and market share, rather than an inevitable financial outcome for every large agency group. The results alone cannot establish how much of that divergence AI explains.

Nor were these businesses previously just collections of media buyers. Their capabilities already include creative strategy, data, technology, consulting and commerce. The transition concerns the balance between services, labour, proprietary systems and commercial results within those portfolios.

Advertising growth can coexist with agency fee pressure

There is little basis for predicting an immediate collapse in aggregate advertising. WARC's Q3 2026 outlook forecasts global advertising expenditure growing 11.9% to $1.34 trillion this year. This is a forecast, rather than a final expenditure count, but it runs against a simple industry-contraction narrative. [4]

Gartner provides a more revealing picture of the allocation problem. Its 2026 CMO Spend Survey places average marketing budgets at 7.8% of company revenue, against 7.7% in 2025. Its accompanying analysis describes increased allocation to paid media funded by cuts to agency spending. The survey covered 401 marketing leaders, predominantly at companies with more than $1 billion in annual revenue; it is not a census of all advertisers. [5][6]

An advertiser can therefore spend more on distribution while negotiating lower service fees. AI-generated production savings might finance extra media, new commerce capabilities or a reduction in total costs. There is no automatic rule that sends the savings back to an incumbent agency.

Remuneration is changing too. WFA and Agency Mania Solutions report that the use of labour-based remuneration models fell from 33% in 2022 to 19% in their latest research. Fixed-fee and output models have gained ground. Yet performance-linked remuneration still accounts for less than 20% of compensation for most respondents across most disciplines. The industry is moving toward different commercial arrangements; it has not completed a universal shift to payment for outcomes. [7]

This distinction helps explain the margin risk. If clients demand lower prices faster than agencies reduce delivery costs, margins compress. If agencies keep fees while delivering more efficiently, margins can improve. If clients buy proprietary technology and higher-value services, revenue can grow. AI makes all three paths possible.

Above-the-line media faces a higher burden of explanation

Broad-reach television, audio, outdoor, print and sponsorship can be difficult to connect to individual transactions. But difficulty observing a complete customer journey does not establish that the investment failed.

Advertising can create familiarity, trust, memory and a willingness to pay before a customer enters an identifiable purchase process. Those effects may influence what someone later asks an agent to find. “Find this brand in my size” is a different commercial opportunity from “find the cheapest acceptable substitute.”

The Profit Ability 2 research offers a useful corrective to excessive short-termism. Commissioned by television marketing body Thinkbox, it combined econometric analysis covering £1.8 billion in UK media investment, 141 brands and 14 categories. Its reported average profit return rose from £1.87 per pound in the short term to £4.11 when sustained effects were included. Those are historical study averages, shaped by its sample and modelling; they are neither a universal return guarantee nor a forecast for agentic commerce. [8][9]

The implication is that broad-reach media must be assessed over an appropriate horizon. Econometric models, experiments and brand measures remain relevant when a click trail cannot capture the effect.

Nevertheless, the budget contest becomes harder. An investment committee comparing an expensive campaign with a system that records product questions and linked orders may find the latter easier to understand. That can create legitimate pressure to test new allocations. It can also create a measurement bias: spending flows toward what is easiest to count rather than what generates the greatest additional profit.

Traditional media is vulnerable where its incremental contribution is weak, where duplication is high, or where the next pound would work harder elsewhere. Its vulnerability cannot be inferred simply from the absence of transaction-level attribution.

Agents introduce another commercial decision point

Consider a hypothetical customer who discovers a premium appliance through advertising, then asks an AI assistant whether it will fit a particular kitchen, work with an existing accessory and arrive before a deadline. The campaign helped create demand. But the final selection may depend on precise information that the assistant cannot locate or verify.

Spending more on awareness does not necessarily solve that failure. The brand needs to make the deciding facts accessible and answerable.

AI-assisted shopping also spans different behaviours. A person may research with an assistant and complete checkout manually. An agent may compare offers, request clarification or prepare a transaction for approval. Fully autonomous purchasing is a further step. Evidence about one behaviour should not be presented as proof of another.

Adobe's October 2026 report illustrates the growing value of the research-and-referral route. It found that, in September, AI-referred visits to US retailers had a conversion rate 90% higher and generated 83% more revenue per visit than non-AI visits. Its analysis measures shoppers clicking through from AI sources. It does not establish the volume of autonomous purchases, the causal effect of AI on spending, or the effectiveness of any individual merchant tool. Higher purchase intent among those visitors could help explain the results. [10]

BCG likewise describes a market requiring both preparation and proportion: its analysis put external-agent referrals below 1% of visits for leading European brands and retailers in late 2025, with a possible rise to 5% by the end of 2026. That is a conditional projection for a particular sample, not a measured global share of commerce. [11]

Visit share and commercial importance are separate questions. A small stream of high-contribution journeys may justify investment; impressive conversion rates on negligible volume may not. The allocation should follow the merchant's actual opportunity.

From buying exposure to funding a capability

Product feeds, transaction integrations, merchant-owned expertise and measurement each address a different part of agentic commerce. Improving product information, enabling checkout and answering a suitability question should be budgeted against the particular problem each solves.

Buying a sponsored placement is media expenditure. Installing a system that helps visiting agents establish product suitability is an investment in the merchant's selling capability. Treating both as a single new advertising channel obscures the allocation decision.

Disclosure: AIVO Journal is published by AIVO, Inc., which develops AIVO Assist. The product descriptions and testing discussed below come from AIVO and do not constitute independent validation.

AIVO Assist belongs in the latter category. AIVO describes it as an on-site product expert that supplies merchant-approved, sourced answers to visiting buying agents, including when an offer is unsuitable. Its published design gives answers a reference that can connect them to subsequent clicks, bookings or sales in the merchant's analytics. These are AIVO's product claims. [12]

This creates a practical case for funding: a merchant may already have demand, yet lose opportunities because an agent cannot confirm a decisive condition. Improving that exchange could increase conversion or contribution from existing demand. Whether it does so is a testable commercial hypothesis.

The initial budget could come from ecommerce development, conversion improvement, innovation, production efficiencies or marginal media activity. There is no reason to assume that every installation must be financed by cutting television. Nor should a merchant transfer a large budget merely because a new system offers more detailed reporting.

Direct attribution is useful. Additional profit is the investment test.

An answer linked to an order provides an observable commercial connection. It does not, by itself, establish that the order would have been lost without the answer.

The customer might already have decided to buy, perhaps because of earlier advertising. Recording the final exchange does not assign it sole credit for creating demand.

The IAB and IAB Europe explicitly distinguish attribution from incrementality: identifying associated sales is different from estimating additional outcomes caused by an intervention. [13]

For an on-site agent expert, reporting should distinguish recorded questions, successfully linked orders and estimated additional commercial value. Disclose the proportion of interactions successfully matched to outcomes, rather than treating unmatched journeys as either successes or failures.

A credible evaluation would compare eligible journeys with and without access to the capability, ideally through random assignment where feasible. Agents can cache or reuse answers, customers can switch devices, and purchases can move to marketplaces beyond the merchant's measurement. The merchant may also have no control over the referring AI system. These limitations affect both linkage and experimental design; changes in stock, price and delivery must also be accounted for. Where randomisation is impractical, phased or matched comparisons offer evidence with weaker causal certainty.

The financial test is the change in contribution after variable costs, less the additional cost of deploying and running the capability. Use revenue net of discounts and refunds, and account for product costs, payment fees, fulfilment, marketplace commissions and variable service costs where relevant. Deduct additional software, implementation, maintenance and media costs once, without subtracting costs already included in the contribution measure. For a time-limited pilot, distinguish upfront investment from recurring costs.

Value need not come exclusively from extra orders. Correctly identifying an unsuitable product may prevent a return, an expensive support exchange or a disappointed customer. Those avoided costs belong in the evaluation when they can be credibly estimated.

AIVO reports improved recommendation correctness in controlled tests using agents it built and ran, on websites where deciding facts were difficult to find. This is a measure of performance on the tested decision task, rather than a standard, independently audited commercial metric. The cited results do not establish incremental live-merchant sales or profit. Detailed claims should be read alongside the test methodology. [12]

What this means for holding company margins

The risk extends beyond the loss of billable hours. A merchant-owned capability may compete for money that previously passed through an agency's media or production operations. If the agency has no role in that capability, it may lose a share of the client's expanding marketing expenditure.

However, agencies can build a role around it: identifying decision failures, preparing approved product knowledge, implementing systems, designing experiments and connecting agent interactions to broader brand and commercial measurement.

These services may support different fee structures. Recurring software and specialist services could improve economics, while implementation and knowledge maintenance could remain labour intensive. Outcome-linked fees introduce disputes over causality, data access and responsibility for stock, price and fulfilment. None of these arrangements guarantees higher margins.

There is also a potential tension between agency incentives and client allocation. An agency rewarded mainly for managing media volume may have less commercial incentive to recommend an owned capability that reduces that volume. This is an incentive risk to examine in contracts, rather than an allegation about every agency relationship.

Reallocate at the margin, then follow the evidence

The sensible starting point is a defined commercial problem. Are agents visiting? Which facts do they fail to establish? Is the product available through a route the merchant can influence? Can linked outcomes and additional commercial value be measured with useful confidence?

Where those conditions hold, a bounded pilot can be funded from weak marginal activity or operating efficiencies. Protect the ability to assess longer-term brand effects alongside short-term commerce gains. Expand investment when additional contribution or credibly measured avoided costs support it, and investigate cases where improved answers produce no commercial change.

For some businesses, the binding constraint will remain awareness or distribution. For others, it may be the inability to answer a buying agent's question. A good allocation process discovers which constraint matters.

Brands may increasingly fund both the creation of preference and the capability to turn that preference into a verified buying decision. Agencies face pressure when their commercial model captures only the former, or when the work they sell becomes easier to obtain elsewhere. They also have an opportunity to help clients connect the two.

Agentic commerce creates a case for a new allocation within the marketing portfolio. That allocation should begin with a defined decision problem, a bounded test and a clear contribution framework. Its scale should follow the evidence.

The commercial advantage belongs to businesses that can create demand, establish when an offer fits and complete the purchase reliably. Investment earns its place when that capability produces additional profit through better sales, avoided costs or both.

Research and editorial notes

Central finding: Evidence supports agency fee pressure, changing remuneration and an emerging agentic commerce allocation. It does not establish that agentic commerce has already caused broad above-the-line budget cuts or industry-wide holding company margin compression.

Forecast versus evidence: The budget redistribution and incentive analysis above are the author's interpretation. Company results, survey findings and vendor product descriptions are attributed separately. Global ad growth, the BCG referral share and company guidance are forecasts rather than realised outcomes.

Publication safeguards: Retain the disclosure before the first substantive discussion of AIVO Assist. Avoid describing Assist as independently validated causal attribution or presenting recommendation-test improvements as sales uplift. Do not extrapolate Adobe referral performance into autonomous transaction market share. There is no defensible universal percentage of traditional advertising to transfer into agentic commerce on the evidence reviewed.

Further evidence that would strengthen a follow-up: live merchant order-linkage coverage; controlled incremental contribution results; maintenance and implementation costs; category-level agent volumes; longitudinal evidence on changes to media budgets; and agency contract economics. A pilot's duration and sample should follow traffic, baseline conversion and the smallest economically meaningful effect, rather than a fixed calendar promise.

Sources

1. WPP, Third Quarter 2026 Trading Update, 29 October 2026.
2. Publicis Groupe, Third Quarter 2026 Revenue, 13 October 2026. Organic growth is 5%; do not substitute reported growth for this measure.
3. Omnicom, Q3 2026 investor presentation, filed with the SEC, 22 October 2026. Core operations exclude disposed and held-for-sale businesses; comparative figures combine Omnicom and IPG.
4. WARC, Global Ad Spend Forecast: Q3 2026.
5. Gartner, 2026 CMO Spend Survey, 11 May 2026.
6. Gartner, CMO Spend in 2026: Redefining Marketing Investment Under Constraint, 25 June 2026.
7. WFA, Global brands are shifting to paying for outputs not time, 29 September 2026. Findings describe the research sample rather than a market-wide audit of fee revenue.
8. Thinkbox, Profit Ability 2, 24 April 2024. Research commissioner has an interest in television advertising.
9. WARC, Profit Ability 2: The new business case for advertising, 24 January 2025.
10. Adobe, AI traffic now generates 83% higher revenue per visit for US retailers, 29 October 2026.
11. BCG, How Retailers Can Win With AI Shopping Agents in 2026.
12. AIVO, AIVO Assist product page, accessed 7 November 2026. Vendor-reported capabilities and testing; not independent validation.
13. IAB and IAB Europe, Guidelines for Incremental Measurement in Commerce Media, November 2025.