A few minutes from our office sits the spot where James “Jem” White opened what’s widely considered the world’s first advertising agency. Two centuries later, the original business model is still recognisable, but exhausted, and if the last twenty years have strained it, AI will force a final reckoning.
What broke the model
The traditional agency model hasn’t been undone by AI alone. Its foundations have been eroded by a set of converging pressures:
- The collapse of media scarcity: digital abundance undermined the premium agencies once charged for access and placement.
- The democratisation and industrialisation of buying: programmatic turned craft into commodity and volume into an incentive structure.
- The shift from brand to performance: short-termism hollowed out strategic work and inflated operational overhead.
- And now AI: the final accelerant that exposes just how much of the work is repetitive, rules-based, and economically mis-priced.
To survive the next decade, agencies need to stop treating all work as equal. Instead, we need to bifurcate the business model into two distinct enterprises:
High-value human enterprise:
Strategy, analytical insight generation, creative judgement, editorial control, and taste. This is the real IP of agencies – where business impact and competitive advantage is created. This is where we must borrow from the consultants’ playbook and ‘Price to Impact,’ not to effort.
Near-zero-value operational tasks:
The messy middle of implementational planning and campaign logistics – pacing, trafficking, reporting, low-order optimisation loops, reconciliation. This should be delivered for as close to zero cost as possible, primarily through technology-led automation.
Clients should not be paying for the taxi meter of human hours. They should be paying for the intelligence that moves their business. Everything else the machine should (and soon will) do at negligible marginal cost. This bifurcation is the foundation for how agencies must redesign themselves in the AI era, and it underpins the three major shifts required.
- Put humans in control of the high-value enterprise, and hand everything else to the machine
The industry has spent decades pretending that every task a human touches has worth. They don’t. The operational backbone of media – trafficking, pacing, reporting, optimisation – is analogous to the processes that other industries automated years ago. Rules-based, repetitive, plentiful, and never meaningfully valued by clients. However, agencies didn’t automate because the economics of their business model rewarded inefficiency: more complexity = more hours = more fee. AI removes that illusion and machines will devour this layer, fast.
Where humans create disproportionate value – and always will – is in the places machines struggle to reach:
- Asking the right strategic questions
- Generating analytic insight
- Interpreting unlabelled signals
- Creating ideas, invention, and innovation
- Exercising taste, judgement, and editorial control
- Understanding culture, nuance, meaning and context
These skills are rare, expensive, and irreplaceable. Clients should pay more for the best thinking in this layer, while the middle miles of media collapse into automation. Planners may not trust AI to set strategy (nor should they), but they should absolutely trust it to handle operational drudgery. This is the first structural split: human-led value, machine-delivered execution.
- Engineer business performance with proprietary intelligence: the new revenue engine
Once the operational middle collapses to near-zero marginal cost, the economics must shift toward creating value in the intelligence layer – the platforms, models and engines that scale human judgement.
This does not mean dashboards or licensed tech, rather it means proprietary systems that encode frameworks, data models, heuristics and engineered tools that help clients make smarter decisions across the entirety of their marketing ecosystems – the things that cannot be rented from the vested, siloed and self-serving interests of big tech.
Two examples:
Next Best Decisioning (NBD)
An econometric predictive modelling system that runs thousands of simulations in real time to optimise portfolios (brands, markets), channels, and audiences toward real business outcomes, such as footfall, revenue, profit, rather than media proxies.
Contextually Aware Optimiser (CAO)
A machine-learning engine blending AI-led editorial semiotics with attention signals to prioritise high-quality media environments and dynamically serve culturally and contextually relevant creative. It is proven to drive 1.5x engagement and 2x brand recall over standard impressions.
Tools like NBD and CAO haven’t replaced human judgement but they have augmented and amplified it. Machines handle the precision engineering and humans steer the logic.
And so emerges the second bifurcation:
Human IP: frameworks, principles, models, analysis, strategic logic
Machine IP: engines that execute and scale that logic at speed
Together they form the new operating model of media: a decision-intelligence layer that guides investment, forecasts outcomes, allocates budgets and generates adaptive plans. If the old model sold hours, the new model sells intelligence and outcomes.
- Shift from outputs to outcomes, and redesign pricing around value
Once value splits between high-value human IP and low-cost machine execution, pricing must change accordingly. You cannot charge uniformly for work of fundamentally different value. Nor can you pretend impressions equal impact.
Advances in marketing science now let us optimise for the things that actually grow businesses:
- Behavioural impact
- Penetration growth
- Memory and salience lift
This is where Pricing to Impact should become non-negotiable. Clients want outcomes rather than hours, and they want the thinking that changes their trajectory, versus the factory work that machines will soon produce at negligible cost. Machines will do the 80% of work that currently drives cost, machine-augmented humans will drive the 20% of work that determines the outcome, and pricing must follow the outcome.
The future agency is not an agency, it’s a decision-intelligence company. It will be the intelligence layer sitting above all of them, ingesting signals, running scenarios, forecasting outcomes, allocating investment, generating live adaptive plans, and codifying the logic of growth at a scale no human team can match.
This company will be defined by proprietary IP and technology built around that IP, reduced headcount, enterprise licensing and recurring, high-margin revenue, and high scalability.
And the talent model must shift just as dramatically. If machines handle the middle, we need strategists who frame problems better, analysts who extract meaning rather than produce dashboards, creativity-minded planners with taste and cultural fluency, hybrid thinkers across strategy × data, creative × tech, culture × performance and people who know what to automate, what to challenge, and how to turn signals into judgement, and judgement into IP.
In other words, a smaller, sharper, more interdisciplinary team that creates and sells systems, not labour.
This is where the industry is heading, whether it realises it or not.
The agencies that get there first won’t just survive the AI era; they will define it.
