Can Claude or Copilot bring your agency in-house? — own the system, rent the scarcity.

For marketing leaders, the real question is not what to cut, but which marketing capabilities should become core.

Over the last few months, we spoke with several marketing leaders at global brands and large consumer goods companies. The leaders are considering adding AI strategy and automation specialists. The scope tended to be specific: mapping campaign planning to marketing strategies, execution, optimization and analytics; identifying where AI could remove manual work; building agentic workflows; connecting them to advertising platforms; and reducing reliance on external agencies or bringing an entire team (social media ads, for example) in-house.

That last objective is the one worth examining. Not because agencies are about to disappear. Because this transition has happened before.

01Marketing capabilities have moved in-house before

When social advertising began scaling rapidly in the mid-2010s, running sophisticated campaigns required an unusual combination of skills. You needed people who understood the media, but increasingly also people comfortable with data, experimentation, algorithms and quantitative optimization.

Many advertisers and agencies could not build that capability quickly enough. So they rented it through the adtech platforms I led.

Over time, two things changed. The talent became less scarce as people learned the platforms. Expertise spread globally. Brands and agencies could hire B+ level operators, if not A level, in markets such as Argentina, India and Eastern Europe at substantially different economics.

Then the platforms absorbed more of the complexity. Google and Meta progressively automated bidding, audience selection, placements, creative optimization and measurement. Today, Google's Performance Max applies AI across bidding, budgets, audiences, creative and attribution. Newer agentic capabilities are extending that further.¹

That was not primarily a cost question; it was a core-capability question.

After social media adtech, I led the ecommerce and growth platforms. Performance marketing followed a similar path. As consumer and e-commerce businesses grew more dependent on measurable acquisition economics, many brought performance marketing inside. SaaS companies did the same as product-led growth became strategically important. Product analytics, lifecycle marketing, experimentation and growth stopped looking like outsourced campaign activities and became part of the operating capability.

AI is reopening that question across a much larger portion of marketing.

02AI changes how much capability a small team can own

This is the important part of the current transition. AI does not simply make an existing marketing organization somewhat more productive. It can change the minimum organizational structure required to perform the work.

Consider a simplified paid-media workflow: research, briefing, audience planning, campaign setup, creative adaptation, QA, activation, monitoring, optimization, reporting, learning, and planning for the next campaign. Historically, economics favoured specialization. Different people owned different steps. The client coordinated with the agency. The agency coordinated teams internally. Information moved through email, spreadsheets, decks, dashboards and advertising platforms.

AI can now participate across much of that chain. One agent can ingest campaign history and surface performance patterns. Another can compare results against targets and identify anomalies. Another can generate audience or creative hypotheses. Another can produce platform-ready variants. Another can draft the weekly performance narrative. Advertising platforms themselves automate bidding, placements, targeting and portions of creative optimization. APIs and deterministic software can execute approved changes. Humans retain responsibility for strategy, material budget decisions, brand positioning, exceptions and high-consequence actions.

03Do not start with automation. Start with strategy.

This is where a leader may end up putting the cart in front of the horse. The tempting sequence is: start with AI, look at the agency scope, find tasks that appear automatable, bring them inside, count the savings.

A more useful AI automation sequence runs with the same strategic discipline that marketers know: goals, then strategy, capabilities required for the strategy, execution plan for the prioritized capabilities, and finally measurement to repeat the process.

The exercise did not end with what AI can replace. It ended with what the company needs to become unusually good at itself. That is the more useful question for a CMO.

The same principle holds outside advertising. A European fintech company we worked with was spending nearly €20 million annually on growth across acquisition, activation, retention, geographic expansion and new business models. The company did not lack data. Its teams were analytical and funnels were measured.

The problem was that optimization had become a substitute for deciding where future growth should come from. We helped the Chief Growth Officer and the marketing leadership develop a three-year market strategy, shift investment toward lifecycle value and B2B2C opportunities, and reallocate the €20 million portfolio accordingly across different capabilities, including machine learning and AI.¹¹

That case matters to this discussion because the sequence is identical. A sophisticated AI system can optimize a suboptimal strategic choice faster, leading to work automation the company should not be doing at all. The AI tool can scale an agency workflow whose underlying objective no longer makes sense.

04Most brands are not yet operating this way

The gap between possibility and reality remains large. A 2025 study by MediaSense and the World Federation of Advertisers, covering more than 30 global companies, found that 71% of in-house teams were implementing generative AI in specific areas and 65% were actively experimenting. Only 12% said AI was fully integrated into their workflows.⁵

That may be the most useful statistic in this discussion.

Giving every employee access to Claude, ChatGPT or Copilot is not an AI-native marketing organization. Neither is automating one reporting process.

The difficult work starts one layer above the tools. Which activities should machines perform? Which decisions can safely be delegated? Which require human approval? Which systems need to connect? How do we validate outputs? How does the workflow learn from one campaign to the next? Who ultimately owns the result?

05The cost advantage is real, but it is the secondary benefit

The research supports that distinction. In 2024, the World Federation of Advertisers and MediaSense surveyed more than 80 multinational advertisers representing more than $60 billion in global ad spend. Three out of four wanted to change their agency remuneration model. Only 15% identified cost reduction as the primary motivation. Advertisers cited better alignment to business outcomes, greater accountability and improved access to talent.²

The CMO does not necessarily want a cheaper agency over better performance. A marketing leader wants their organization to own capabilities that have become strategically important, and to stop paying premium rates for work that technology has made increasingly standard.

A company may want to internalize performance measurement because it wants the learning. It may want customer insight inside because that knowledge compounds. It may want direct control of first-party data because every future model and campaign depends on it. It may want campaign operations inside because automation means outsourcing no longer creates the same leverage.

At the same time, it may gladly pay more for world-class creative judgment, category expertise, new-market intelligence, technical specialists or independent strategic challenge.

AI did not create this movement. In 2023, the Association of National Advertisers reported that 82% of surveyed members had an in-house agency, up from 78% in 2018, 58% in 2013 and 42% in 2008.³ The ANA's 2026 research found respondents were five times more likely to say marketers were bringing more work in-house than pulling back. 53% said the primary role of an in-house agency should be as a strategic partner participating upstream in strategy and brand-building work.⁴

That challenges the old idea that in-housing simply means moving production jobs onto the client's payroll. Companies increasingly want internal marketing organizations that understand the business deeply enough to shape strategy. AI adds leverage to that ambition. A brand no longer needs to recreate a 40-person external agency structure simply to acquire meaningful execution capability.

06The agency is not disappearing — it is being unbundled

The evidence does not show brands abandoning agencies. In the same WFA research, 61% of advertisers expected overall agency fees to increase over the following three years, because brands expect to pay more for scarce strategic and technical talent. At the same time, 58% expect to pay less where AI is deployed.²

Those numbers look contradictory only if we assume an agency sells one thing. It does not.

An agency relationship has historically bundled strategy, creative judgment, specialized talent, market knowledge, technology, production, media operations, experimentation, measurement and a substantial amount of labour. People build campaigns, traffic ads, refresh creative, assemble reports, analyze performance, monitor pacing, manage experiments and move information between systems.

Agency compensation takes many forms: labour-based fees, retainers, project fees, fixed-output pricing, commissions, incentives and hybrids. But wherever compensation remains directly connected to media spend, AI creates an interesting economic question.

Consider a company spending $20 million on media at an 8% managed-spend fee. The agency fee is $1.6 million. Raise media investment to $25 million and the fee becomes $2 million. The agency earns another $400,000. The question is: did the work increase by 25%?

Historically, campaign scope and complexity often expanded with spend, making that relationship easier to defend. But the platforms themselves now automate increasingly large portions of campaign execution.¹ As media spend and human effort separate, media spend becomes a weaker proxy for the value of highly automated execution.

This does not mean 8% is too much. An agency generating exceptional incremental value could easily be worth more. It means the conversation needs to move from what percentage to pay, to what economic value is being delivered.

Some of the clearest evidence comes from the agencies themselves. In October 2025, WPP launched WPP Open Pro, a self-service version of its AI-powered marketing platform. The product enables brands to plan campaigns, create content and publish directly to major advertising platforms. WPP explicitly described it as designed in part for brands moving toward in-house execution.⁶

One of the world's largest advertising businesses is productizing activities that historically required agency teams. That is not corporate self-destruction. It is an adaptation. If execution increasingly becomes software, WPP would rather participate in that software layer than defend every manual process.

At the same time, WPP is moving in the opposite direction at the high end. In its February 2026 strategy update, the company described its Jaguar Land Rover relationship as an outcome-led model with agency objectives aligned closely to client business goals.⁷ WPP is therefore moving simultaneously toward more technology-enabled self-service execution, and toward deeper, higher-value strategic relationships. The pressure sits between them.

Publicis has made the same argument from another direction. At Cannes in June 2026, Publicis criticized AI pitches centred on false efficiencies and argued the industry must move from AI demonstrations to proof of business value.⁸ That position has credibility because Publicis reported 5.6% organic growth in 2025 and an 18.2% operating margin, describing AI as a driver of both growth and margin expansion rather than a headwind.⁹

AI does not automatically eliminate the agency. It forces the agency to move higher up the value chain.

07Own the system. Rent the scarcity.

For many brands, this becomes a useful operating principle.

Customer data, campaign history, measurement and institutional learning compound over time. An agency relationship should not leave the advertiser unable to understand its own customers or performance without the agency present.

The brand should maintain direct control over strategically important advertising accounts, measurement architecture and data relationships. Partners can operate them. The company should be able to access, understand and change its own marketing system independently.

High-volume, repeatable, data-rich work — reporting, campaign QA, performance analysis, creative adaptation, brief generation, budget pacing, routine optimization, knowledge retrieval — is the natural candidate for internal AI. Not because agencies necessarily perform these activities badly, but because the economic rationale for outsourcing them is changing.

Decision rights should remain explicit. The organization needs clear boundaries: which decisions machines may make, where humans approve, and who owns the business outcome the system produces.

The agency earns its premium where it brings something genuinely difficult for the client to reproduce: a breakthrough creative idea, deep category expertise, specialized technical capability, cross-market pattern recognition, rapid access to rare talent, independent strategic judgment, or demonstrably better commercial outcomes. That is a healthier partnership than paying for a large bundle and assuming every component retains the same value indefinitely.

The market is becoming a barbell. At one end: technology-enabled execution — more automated, more self-service, more agentic, more likely to move inside the company, and progressively less expensive per unit of output. At the other: high-value human and institutional capability — strategy, creativity, technical architecture, transformation, judgment, relationships and accountability. The traditional middle, large teams performing repeatable execution under labour-heavy scopes, faces greater pressure.

08Outcome pricing makes sense in theory. Contracts are moving more slowly.

It is tempting to conclude that every agency should move to outcome-based compensation. The evidence calls for more care.

WFA found that 58% of advertisers planned to increase output- or outcome-based remuneration, but also that 84% identified inadequate data and measurement as a barrier to evolving compensation.² Actual contracting is moving more slowly than the rhetoric.

The ANA's 2025 Trends in Agency Compensation study, covering 99 member organizations, found that performance incentive programs had fallen to 15%, the lowest level since 1994, partly because of the complexity involved in structuring effective agreements and uncertainty about whether they materially improve agency performance. ANA also noted that respondents had not yet reported AI significantly affecting their agency compensation agreements.¹⁰

Outcome pricing works best where the metric genuinely matters, attribution is credible, the agency can meaningfully influence the result, and both sides agree on the baseline. The likely destination is not one universal model but a mix of fixed fees, output pricing, platform fees, specialist retainers and outcome-linked components where measurement supports them.

The principle matters more than the mechanism: compensation should increasingly reflect value created rather than labour consumed.

09Run the full equation before reorganizing the department

The business case that appears frequently in this conversation: the agency costs $1.6 million on $20 million of spend, so bring it inside and save $1.6 million.

That is an incomplete equation.

The correct version subtracts internal talent, AI and marketing technology, data and measurement infrastructure, integration, governance, change management and the specialist partners still required. There can still be a meaningful prize. But the objective should not be maximum in-housing. It should be maximum marketing leverage.

That is what marketing leaders should be trying to accomplish with AI.

10Six questions worth asking before renewing an agency scope

• What are we actually buying? Separate strategy, judgment, execution, technology, production, data, access and talent. An agency relationship is a portfolio of capabilities, not one thing.
• Which work has become repeatable enough for AI? Look for high volume, standardized inputs, measurable outputs and systems accessible through APIs.
• Which capabilities compound in value when we own them? Customer intelligence, measurement, experimentation history and institutional learning frequently fall here.
• What does the agency possess that we genuinely cannot reproduce economically? Be precise. "Expertise" is not a specific answer.
• Does the compensation model reward the behaviour we actually want? If agency revenue automatically increases when media spending increases, understand the economic logic that still connects those two things.
• If AI makes the work dramatically more productive, who receives the productivity dividend? The agency, the advertiser, or both? That conversation is only beginning.

11How we think about this at StatsLateral

We do not begin an AI marketing engagement with a list of tasks a model can perform. We start with the growth objective, then work backward.

What must become true for the business to achieve that objective? Which capabilities create genuine competitive advantage? Which knowledge and data should compound inside the organization? Where is external specialization more valuable? Which workflows can now be redesigned around AI? Which decisions should remain human? What should be built, bought or partnered for? And how will leadership know whether the new operating model is producing a better business result?

Our published work reflects that sequence. In AdTech, it meant turning fragmented cross-channel marketing complexity into a production-ready product in 14 weeks and validating a $10 million-plus revenue opportunity. In fintech, it meant stepping above campaign optimization and helping leadership reallocate €20 million toward the next source of growth. In a PE-backed business, it meant treating the GTM capability itself as a product and converting fragmented commercial knowledge into a working operating system.

Technology changes and will continue to evolve. The management problem is where we focus: decide what the company needs to become unusually good at, then design the system, talent and partners around that choice.

12This is not the end of the advertising agency

Treating this as brands versus agencies misses the larger opportunity.

Brands can bring inside capabilities they increasingly need to consider core. Agencies can stop defending labour that software is steadily commoditizing and charge more for the things technology makes scarcer: judgment, ideas, specialized expertise, independent perspective, accountability and measurable results.

The best agency may therefore become more valuable, not less. But for a different reason.