Agency Reset Addendum Q1 2026: The Control Shift

Note: Throughout this section and the remainder of the report, I refer to “Tiers”. This model classifies marketing work by how much strategic or creative origination is required, not by agency size, seniority, channel or importance. Across the Spring CC and IHALC reports, the same basic structure holds:

Tier 1, Origination: strategic and creative work from the brief, e.g. brand development, campaign or channel strategy, original concepts, high-level problem-solving.

Tier 2, Development: turning an approved idea into channel-ready work, e.g. design, copy, content development, localisation, rollout, campaign adaptation and project coordination.

Tier 3, Production: execution against defined specifications, e.g. resizing, versioning, artwork, templating, copy swaps, asset adaptation, trafficking and high-volume production.

BenchPress 2025 showed a stressed model: hourly rates fell, redundancy levels rose, and margins were under pressure. BenchPress 2026 then showed a partial rebound. Sixty percent of £1m+ agencies grew fee income in 2025, average blended hourly rate rose 6% to £122, and average gross profit rose from 39% to 43%. The share hitting the 50% gross profit benchmark rose from 24% to 36%.

That looks healthier. But the underlying pattern matters more than the top-line recovery. BenchPress 2026 says agencies generated more fee income with the same overall number of staff, while 68% report delivering more quickly because of AI. It also says explicitly that AI is playing a big role in that shift. Yet only 16% say AI has made them more profitable, while 32% say clients now expect to pay less, and 38% say clients have taken work in-house.

Some agencies may currently be enjoying an AI efficiency dividend before the full Pricing Shockwave hits. BenchPress 2026 points to the pattern: agencies are delivering faster, generating more fee income with the same number of staff, and recovering gross profits from 39% to 43%. However, buyer pressure almost always lags supplier efficiency. Contract cycles, procurement reviews and client-side AI adoption mean the commercial consequences arrive later than the agency’s efficiency gains. The danger is that agencies bank the margin benefit without changing the pricing model, only to find at renewal that clients now understand the work takes fewer people, fewer hours and less cost to deliver. At that point, the efficiency dividend gets negotiated away unless the agency has moved pricing toward value, outcomes, orchestration, IP, measurement and accountability.

The UK employment picture adds force to the argument.

The IPA Agency Census 2025 reports that IPA member agencies employed 24,963 people on 1 September 2025, down 6.8% from 26,787 a year earlier. The split matters more than the headline. Employment in creative and other non-media agencies fell 14.3%, while media agency employment rose 2.4%.

Employees aged 25 and under fell 19.2%, reported vacancies fell 40.8%, and only 43.4% of responding agencies still employed graduate trainees, apprentices, or school-leaver apprentices.

Taken together, this points to a labour model being selectively hollowed out, especially on the creative and junior side. The big risk here is that the “apprentice” system that produced future senior judgement is potentially starting to break. AI can compress or remove the very work through which juniors historically learned, e.g. research, first drafts, versioning, QA, reporting, campaign analysis, platform activation, and client exposure.

The Pyramid Collapse is therefore not only a cost-structure issue, it’s a succession issue. If agencies hollow out junior and early-career roles without deliberately redesigning learning, supervision, and progression, they may preserve profitability in the short term while creating a future shortage of people capable of creative judgement, strategic interpretation, client leadership, and responsible AI oversight. The industry cannot yet know the full consequences, but agency leaders, trade bodies, educators and policymakers need to think seriously about how early-career talent is trained when the traditional “apprenticeship” ladder is being pulled up gradually.

The VerityRI Relationship Report remains one of the clearest correctives to simplistic automation logic. It shows that the work itself ranks only 30^th out of 91 attributes in driving high client ratings. Clients who score agencies 9 out of 10 are far more likely to praise the experience than the output. At the same time, 1 in 3 clients now rate their agencies lower for confidence to meet future needs than for satisfaction today.

Human value is not disappearing, but it’s narrowing toward strategic vision, business understanding, confidence, trust, and relationship quality. In other words, the human layer becomes more valuable, but only where it provides direction, interpretation, and accountability.

“However, relationship capital becomes defensible only when paired with stronger control points, i.e. delivery quality, governance, measurement, and accountability” Simon Billington, EVP, Team Lewis.

This is directionally aligned with the original Agency Reset report, but incomplete because relationship capital on its own is no longer enough given the weight of the Q1 evidence. Human trust now has to be paired with stronger control points, better measurement, and a clearer commercial model.

The in-housing story has sharpened materially.

The IHALC 2025 benchmarking survey shows that in-house teams are not merely absorbing production. Only 22% currently identify as lead agencies, yet 44% want to become one. In-house teams are also pushing into more Tier 1 creative origination work and broader ownership of brand systems and workflow control.

That trend sits alongside the broader ANA finding that 82% of member marketers now have an in-house agency of some kind. The wider direction of travel is clear: more brands are building internal capacity, internal knowledge, and internal control.

“AI compresses knowledge acquisition, but it doesn’t automatically create better judgement. Agencies that win will operationalise cross-category learning into repeatable decision advantage” Simon Billington, EVP, Team Lewis.

The risk to agencies is therefore not just the loss of low-value production. The deeper risk is that clients internalise brand knowledge, workflow governance, asset infrastructure, and selected strategy-adjacent work, leaving the external agency with an ever-thinner, easier-to-swap layer.

Although the Network Agencies and big Indies are not a definitive guide to market trends, they give a glimpse into what many clients may start to expect from their agencies over time. Here are the Q1 signals worth highlighting.

Note: Sources have been checked as far as possible at the time of publication. Any material errors will be corrected in future versions on request.

Publicis: the leaders are turning AI, data, and measurement into operating advantage

Publicis reported 4.5% organic net revenue growth in Q1 2026, re-affirmed full-year guidance, and continued to signal a willingness to deploy capital into capability-building acquisitions.

Publicis appears to be widening the gap with other agencies, large and small, by turning AI, data, and measurement into a stronger operating system. That’s what the winning side of Agency Reset looks like in practice, in my opinion.

Another example is its acquisition of 160over90, which strengthens its attempt to position itself as an end-to-end sports ecosystem where cultural relevance drives measurable growth.

Omnicom/IPG

Omnicom completed the IPG acquisition in November 2025, creating a group with pro forma revenue above $25bn; the deal was also expected to deliver around $750m in annual cost synergies. The signal is clear, the largest players are rebuilding around scale, operating leverage, and platform-like control. For smaller independent agencies, the warning is sharper: fragmented sub-scale labour-based delivery is becoming less defensible, while value is moving toward data, workflow infrastructure, measurement, orchestration and client-access control.

WPP: the old holding-company architecture is being compressed

WPP’s restructuring is, in my opinion, a live admission that the old multi-brand holding-company model is too duplicated, too expensive, and too slow for AI-era economics. Again, opinion not fact.

This isn’t just a cost story, it’s a control story. Complexity that once looked like breadth now looks like drag. Value is moving toward integrated operating layers, shared systems, and clearer accountability.

WPP + Adobe: orchestration is becoming a commercial product

The expanded WPP-Adobe partnership makes the point explicit: orchestration is becoming a commercial product, not just an internal workflow improvement. The lesson here is that orchestration is becoming sellable. The control point is moving from making assets to coordinating/orchestrating the system that plans, creates, governs, quality assures, and activates them.

Stagwell

Stagwell’s push around The Machine and Search+ shows that larger groups are trying to own the orchestration and AI-discoverability layers, not just use AI to make existing delivery cheaper.

The key insight for me is that AI-search visibility and workflow operating systems are becoming products in their own right, and therefore potential control points.

Mega and other AI-native challengers: software is attacking the agency model directly

Mega’s funding and positioning show how agency-like outcomes can be packaged into software-style delivery. It’s a warning shot that shows how quickly AI-native challengers can turn execution-heavy services into AI-agent-driven systems.

Meta Advantage+: the platforms are absorbing the optimisation layer

Meta’s momentum with Advantage+ is another reminder that the platforms are not standing still while agencies debate commercial models.

If media platforms increasingly automate targeting, optimisation, and creative iteration inside their own stack, agencies lose part of the execution layer and “media platform expert knowledge” angle, accelerating the need to focus on strategic, creative and cross-platform services.

Deloitte Australia, AI without assurance creates liability

Deloitte agreed to partially refund the Australian government after a report was found to contain fabricated references, non-existent academic papers, and an invented quote from a federal court judgement (this was reported in Oct 2025 across various sources). Deloitte did not directly attribute the errors to AI, and the department said the corrected report’s core findings were unchanged. Warning for agencies, this is the “faster wrong answers, confidently delivered” problem made public. As AI accelerates delivery, QA, governance, human review, evidence standards, and accountability become control points, not back-office hygiene.

The original framework remains directionally correct, but needs a structural upgrade.

The old Cockpit and Engine model captured an important truth: future value would move upward toward senior judgement and downward toward systemised execution at scale. What it under-described was the increasingly important middle layer that coordinates the whole system.

The new canonical model: Cockpit -> Cabin -> Engine

If you read Sangeet Paul Choudary’s ecosystem framework, and his book Reshuffle, and apply it to agencies, the implication is clear in my view: agencies only exist because they remove constraints in the client’s growth system.

To put this in context, here’s a brief summary of the Theory of Constraints by Eliyahu Goldratt, introduced in his 1984 business novel The Goal. In it, he states that in any system, performance is determined not by overall capacity, but by the single binding constraint that limits throughput, and that sustainable competitive advantage comes from identifying, exploiting, and then elevating that constraint before finding the next one.

When AI, software, platforms, and in-house capabilities reduce or remove old client constraints, agency value doesn’t survive by default. It has to migrate to address the new constraints.

In the traditional model, and in many ways why the agency model worked so well, the common client constraints included:

• Execution capacity: not enough internal resource to produce and distribute content, campaigns, and assets at the required volume, speed, and accuracy
• Strategic capability: lack of in-house capacity to translate business objectives into coherent marketing strategy, channel prioritisation, and audience insight, particularly at pace and across multiple markets
• Creative capability: insufficient internal talent, creative infrastructure, and production expertise to conceive, develop, and execute distinctive brand and campaign creative to a professional standard
• Specialist skills and labour: lack of in-house expertise across disciplines such as paid media, SEO, creative, data, social, influencer, and technology
• Campaign coordination: inability to manage complex, multi-workstream campaigns coherently across internal teams and external partners
• Ad platform activation: limited internal knowledge and bandwidth to set up, optimise, and govern activity across Google, Meta, and emerging ad platforms
• Multi-channel access and integration: no single partner capable of connecting strategy to execution coherently across all relevant channels

In the emerging AI-enabled marketing ecosystem, the new client constraints are shifting toward:

• Signal coherence: no single coherent view of performance across channels, platforms, and AI-generated touchpoints, making strategic decision-making increasingly unreliable without expert synthesis. Note: Signal fragmentation is the symptom, e.g. too many dashboards, too many AI-generated outputs, no clarity on where to invest attention. Coherence is the actual constraint agencies can build a durable position around.
• AI governance and decision rights: unclear accountability for AI agent actions, outputs, and escalation paths across client, agency, and platform boundaries
• Adoption and change management: as brands deploy AI into marketing workflows, the problem is increasingly whether employees understand “the art of the possible”, trust the system, understand the rules, use it safely, and experience the change positively. As with all technology enablement, change management is always a major challenge.
• Measurement integrity: attribution and proof of value becoming structurally harder as campaign variety, speed, and AI-generated content volume outpace existing measurement frameworks
• Orchestration coherence: inability to align tools, workflows, teams, and platforms into a coherent operating system, particularly as AI tooling proliferates faster than integration capability
• Strategic coherence across human and AI decision-making: as AI generates more options, more data, and more automated actions simultaneously, the binding constraint shifts from having a strategy to maintaining a coherent one. Clients need senior judgement to set clear priorities, make confident trade-offs, and hold a consistent growth narrative as the pace and complexity of decisions accelerates beyond internal capacity
• Creative distinctiveness at scale: AI removes the production constraint almost entirely but immediately creates a new one. When every competitor can generate content at volume and speed, the binding constraint becomes creative differentiation, human taste, cultural intelligence, and brand judgement to ensure that what gets produced at scale actually builds brand equity and performance rather than diluting it
• Experimentation velocity: the speed and scale of AI-enabled testing now outpace most clients’ ability to design, interpret, and act on experiments without introducing noise and false learning
• Discoverability in AI-shaped environments: brand and content visibility in AI-mediated search, recommendation, and answer engines requires fundamentally different strategic and technical capability than legacy SEO and paid media. The pendulum here goes well beyond findability. As AI agents become active participants in discovery and purchase, not just as assistants but as autonomous buyers, brand engagement applies equally to Agent customers and Human customers. That creates new risks: fraudulent or unauthorised purchasing by agents, liability for agent-initiated transactions, and the reality that human adoption of agent-led financial independence will be uneven and slower than the technology allows.
• Trust and brand integrity in AI outputs: ensuring AI-generated content, personalisation, and agent interactions remain on-brand, compliant, and commercially safe at scale

In summary, old constraints built the existing agency model. New constraints will either rebuild it or replace it. Which side of that line you land on is a choice, and 2026 is the year that choice gets made.

The original report used the language of control points. This addendum makes that concept explicit.

A control point is not just a capability. It’s a position in the system that shapes decisions, improves with use, influences resource allocation, and is difficult to replace without losing coherence or performance. Control points are where bargaining power accumulates, and where agency fees can be defended or lost.

Control points that are weakening:

Several traditional agency control points are becoming less defensible:

• Volume production and content execution
• Routine campaign optimisation and BAU management
• Standard performance reporting and dashboards
• Generic channel execution across paid and organic media
• Access to generalist creative and specialist labour
• Opaque time-based pricing as a proxy for value

These are not disappearing overnight, and many will remain monetisable for a period. But they are increasingly benchmarkable, susceptible to technology substitution, or movable in-house without meaningful performance loss.

Control points that are strengthening:

The more durable control points now sit higher in the system, closer to decision-making, governance, orchestration, and proof of value:

• Measurement architecture: designing the systems that connect marketing activity to real business outcomes, not just surface-level outputs
• Signal design and data quality: ensuring the inputs that feed AI systems, attribution models, and optimisation engines are structured, clean, and strategically sound
• Experimentation design and learning loops: building the frameworks that turn AI-enabled testing volume into genuine organisational learning rather than noise
• Orchestration coherence: governing fragmented tools, platforms, teams, and workflows as a single coherent system rather than a collection of parallel activities
• AI governance and exception management: setting the rules, boundaries, and human oversight mechanisms that keep AI-enabled workflows on-brand, compliant, and commercially safe
• Discoverability strategy: shaping brand and content visibility across search, social, retail media, and AI-mediated recommendation environments as the rules of each change simultaneously. Note: AI-mediated discoverability also changes the role of trusted publishers and specialist media. As more buyers use AI systems, answer engines, and recommendation environments to form views before speaking to suppliers, authority signals become critical. Visibility will depend not only on technical optimisation, but on whether a brand is present, cited, discussed, and trusted inside ecosystems.
• Senior growth decision support: providing the interpretive layer that translates data, signals, and market complexity into clear strategic priorities and confident commercial decisions.

Therefore, if the client can buy execution more cheaply, they will likely value, and pay external partners (agencies) for one or more of these three things:

• Better decisions
• Better orchestration
• Or better proof

The agencies that hold durable control points will be positioned to deliver all three. Those that do not will find themselves competing on lowest-cost-provider economics, where AI wins by default.

Why measurement is becoming one of the critical control points to hold

Multiple sources point in the same direction. As media buying is progressively absorbed into smart campaign systems, agency value migrates toward signal setup, data quality, attribution design, and measurement architecture. If marketing ROI is what the C-suite demands, and in 2026 it consistently is more than in previous years, then reporting must connect activity to real business impact, not just campaign metrics. The agency that owns measurement owns the proof of its own value. The agency that doesn’t will always be vulnerable to the question:

What exactly are we paying these agencies for?

This reinforces the central thesis: the agency of the future can’t simply be the producer of work, it must become the designer of measurement, learning, and decision systems.

A word of caution about measurement systems. Clients need to understand not just what is being measured, but where their data lives and who controls it. As agencies build measurement architecture, data ownership becomes a structural risk for clients, particularly around portability and vendor lock-in. If all the measurement infrastructure is embedded in an agency system, what happens when the relationship changes? Clients should be very concerned about this. Agencies that can answer it clearly, and offer portability as a feature, hold a stronger position as a “trusted advisor”. Those that can’t are creating a hidden liability which never sits well with clients once exposed.

Agency leaders need to see where their own model is exposed and what a plausible alternative looks like. The following are hypotheses about how agency models could start to change. They’re not definitive, but they look directionally correct, based on my Q1 research and agency meetings.

Live market signals worth noticing

A number of current examples show where agencies, brands, and specialist providers are moving:

• Brainlabs: Great article on LinkedIn about Building an AI-Native Company from their CEO, Dan Gilbert
• Directive / Stratos: an agency building a proprietary B2B revenue-intelligence layer connecting CRM, paid media, SEO, finance, and ops data
• NoGood / Goodie: a signal that AI visibility and answer-engine visibility are becoming their own specialist layer
• Rogers Communications / Invoca: a case showing how better call-outcome data can improve keyword and campaign optimisation
• DISH / Invoca: a case showing how media optimisation is moving toward value-weighted decisioning, not just lead volume
• Unilever: a global brand reorganising creative and production systems around AI-enabled asset creation, digital product twins, and internal studio capability

Taken together, these examples support a sharper point: the market is moving beyond “agencies using AI” toward providers and brands building proprietary layers around revenue intelligence, discoverability, signal quality, and internal production systems.

Business model archetypes

Note: This is just my hypothesis, based on what I can see, who I talk to, and what may happen in the next three months. If you have a different view, please get in contact, I’m always open to constructive challenge.

What Agency CEOs should take from this

The common pattern is consistent. Execution-heavy revenue is not necessarily dead today, but it’s less defensible over a one-to-three-year horizon. The agencies with the best runway are not those doing the old work slightly better. They’re those migrating upward into control, proof, data, measurement, and orchestration; in other words, into the Cockpit and Cabin layers.

Agency leaders need a design logic, not just a market diagnosis. Here’s my list of design principles around which you can start to reimagine your agency of the future.

Start with control points

Don’t ask where you can add AI. Ask where value will be created and captured in the new client ecosystem, and whether your role in that system gets stronger or weaker with your existing model.

Democratise Tools or Risk Preserving the Bottleneck

Flatter delivery structures only work if access to technology is genuinely democratised across the agency, not licensed to a select few, not siloed by seniority or function. When only limited roles or teams have access to AI tooling, you can’t run a flat structure. The bottleneck doesn’t disappear, it just moves.

Use the Theory of Constraints

Every agency offer should be built around a real bottleneck in the client’s growth system. If you’re not relieving a meaningful constraint shift, you’re selling activity.

Apply Jobs to Be Done discipline

Redesign the agency proposition around the progress the client is trying to make, in a world where the old constraints have moved and AI and agents have changed the landscape, not the capabilities the agency prefers to sell.

Stress-test with business model theory (Christensen)

A new value proposition may not fit the old business model. If it clashes with your resources, processes, or profit formula, the existing agency organisation will reject it, even if leadership claims to support it.

Unbundle Tasks Before You Rebuild Roles

AI should never simply be bolted onto existing job descriptions. You have to break existing roles into tasks, move repeatable work into AI-enabled systems, then rebuild human roles around judgement, orchestration, quality control, client context and accountability. This is Sangeet Choudary’s point: AI doesn’t just change the job, it changes the system around it. The winners redesign workflows, decision rights and commercial models for the new world, the laggards just make yesterday’s operating model faster.

Never confuse tools with solutions

A tool speeds up a task. A solution removes a client constraint and manages risk. The strategic danger is not a weak AI stack, it’s mistaking enablement and efficiency for client-value.

Distinguish system design from task automation

Automating a step inside a broken workflow is not transformation. AI creates more value when workflows, roles, incentives, and decisions are redesigned around it.

Optimise for pricing power, not just efficiency

Efficiency without value capture is a gift to your client/procurement. Agencies should treat saved time as fuel for better outputs, better proof, more testing, and stronger client outcomes.

Protect judgement and automate repetition

Repetitive, rules-based, data-rich work should increasingly move to the Engine. Interpretation, taste, challenge, and accountability should move upward to the Cabin and Cockpit.

Design the Cabin deliberately

The middle layer is no longer administrative glue. It is where orchestration, governance, QA, learning loops, and exception handling sit. In many cases it will become the most valuable layer in the model. This is where human and AI collaboration is most active, it’s where junior-to-mid staff are working alongside AI, not just as a coordination layer, but as a production-and-governance hybrid layer. Warning, unless senior agency management truly understands the role of the Cabin, it will get treated as “administrative overhead” rather than the high-value orchestration layer that holds it all together.

Put measurement before scale

If you cannot show what good looks like, you cannot scale intelligently. Measurement is not reporting; it’s the basis for commercial defence and resource allocation.

Recognise the risk of cognitive surrender

As AI outputs become easier to trust, agencies must build stronger review logic, human override mechanisms, and quality governance. Faster wrong answers, delivered confidently, are not an advantage.

Treat leadership attention as a scarce resource

Senior leadership time is already consumed by administration, inbox load, and “work about work.” That strengthens the case for Cockpit-level offers built around decision support, synthesis, and signal clarity rather than more noise.

Assume temporary gains will be competed away unless protected

Productivity gains can improve gross profit before pricing adjusts. Agencies should assume those gains are temporary unless they are protected through differentiated positioning, stronger methodologies, better proof, commercial redesign, or proprietary assets.

Avoid the obvious mistakes

• Treating AI as a tooling/tech project
• Defending pricing with effort narratives that are getting weaker
• Carrying service lines that depend on labour-heavy economics with no credible migration path
• Assuming relationship goodwill will compensate for weak proof of value

Redesign your business and operating models

• Think about the operating model as Cockpit -> Cabin -> Engine
• Shift pricing away from hours where possible and toward outputs, outcomes, and risk sharing
• Introduce new client-facing offers around new client constraints, not legacy ones
• Think strategically about capability development for a world with fewer junior repetitions and more hybrid, orchestration-judgement type roles

Think like a systems engineer

• Build repeatable delivery engines where the work is standardisable
• Build orchestration layers where systems, handoffs, and measurement matter
• Build decision-support assets and capability where clients need synthesis and commercial clarity

Run the numbers, always

• Protect existing revenue streams while you work out how they get migrated onto new models
• Run investment scenarios and cash-flow sensitivities to build a runway as you transform the agency
• Protect margins by deciding where efficiency gains are retained, reinvested, or shared with clients

Prioritise ruthlessly

The old model may still generate profitable revenue for a while, perhaps even a few years. The question isn’t whether the agency can still monetise it today, it’s whether it should keep building around it for the future. Agency leaders making hard prioritisation calls owe their teams a clear articulation of the vision behind those choices, not just the what but the why. The people closest to the work often have the strongest signal on where the real opportunities are. A culture that only takes direction downwards will miss some of the best ideas for what to build next.