The optimisation button, and who gets paid when everyone presses it
How AI ad automation lifts CPMs across a category and where the commodity gap pays out
In 2005, Staples began selling the Easy Button. An advertising conceit (press it and office supplies appear) became a product: a red plastic dome that said ‘that was easy’ when pressed, and millions of people bought one.
Twenty years on, the advertising industry has built the real thing. Meta calls it Advantage+, Google calls it Performance Max, and by 2024 Meta was reporting a $20 billion annual run-rate through its version. You press ‘the button’, relax the levers, and the machine buys the media.
The button works. Here’s why.
The models train on more purchase behaviour than any advertiser will observe in a corporate lifetime, and they improve each quarter; give it a few weeks of learning and it becomes frighteningly effective. The trouble starts when the whole category presses it at once. The machines learn from the same signals, so they find the same shoppers, the ones already close to buying, and five brands’ algorithms end up bidding against each other for the same people. Auction prices rise until the CPM has swallowed the efficiency gain, and the CPM is the platform’s revenue line. Meta’s own full-year 2025 results show the mechanism at work: average price per ad up 9%, on top of 10% the year before, with advertising revenue up 22% in a year when the targeting on offer to advertisers had never been better.
Meanwhile the dashboards across the category keep showing sales, because the machines excel at taking credit for shoppers who were coming anyway. The category sells no more than before; the same buyers get reallocated at a higher media cost, and the platform banks the difference. No single brand can stop pressing, since sitting out hands share to the neighbours, so the category pays more each year to hold the same positions.
The button has become electricity. You cannot run the shop without it, and it differentiates very little.
In a nutshell
When everyone presses Advantage+ or Performance Max , the efficiency gains get competed away and auction prices rise.
The commodity gap is the margin between automation on default settings and a better-taught machine.
Brands close the gap with better inputs; agencies close it by augmenting platforms and turning those inputs into recurring data and tech fees.
Platforms and adtech profit from the same gap, monetising better data, better proof and better infrastructure.
Agentic automation frees agency people from the drudgery so they can build those better inputs, feeding the augmented system.
The distance between that commodity baseline and what a better-taught machine earns is the commodity gap. And it’s zero-sum: platforms capture the value, and the rest of the ecosystem plays a supporting act. This piece argues the reverse. The gap is a place where brands, agencies, platforms and adtech vendors can all make money at once, and mostly by helping each other. The same button is spreading to the open web. The Trade Desk sells Kokai as autopilot, DV360 pushes its automated tiers, and Yahoo’s DSP markets the same hands-off promise. The money leaks differently there, through platform fees and curation take rates rather than one auction the seller also banks (that side is covered in depth in a future issue).
Here are the layers in turn.
Brands: feed the machine what your rivals don't, and win the same auction
For a brand, the gap is a category-level opening, because your competitors’ default settings are the baseline you get measured against. While the category runs the button on default settings, you can augment the same machine and earn an advantage inside an identical auction. First-party data flows back through server-side signals instead of the same thin pixel the brand next door uses. Conversion values get rebuilt as margin-weighted lifetime value instead of revenue, so the algorithm stops treating a discount-hunter and a decade-long customer as the same win. Clean-room curation, the Amazon Marketing Cloud and Ads Data Hub generation of tools, produces audiences and incrementality reads the platform interface will not volunteer.
Google and Meta have already built the escape route out of the commoditised-auction trap. Advertisers don’t need to wait for permission or new tech - the tools already exist inside the platforms themselves. What they may lack, however, are the inputs.
Agencies: sell the data and the model, not the hours
For agencies, the gap addresses the most talked-about commercial problem in adland: fee income tied to headcount, benchmarked by procurement, and compressed by the same AI that is commoditising the buying. The gap is where replacement revenue lives, in shapes procurement cannot compare to an hourly rate card.
Here’s how
Custom cluster models, audience and contextual segmentation built on the client’s own data instead of off-the-shelf taxonomies, become licensed assets with a tech fee attached. Marketing mix modelling has become the growth discipline of the decade, helped by the platforms themselves: Meta open-sourced Robyn and Google shipped Meridian, free toolkits for building a model of which channels drive sales.
Agencies can still build MMM from scratch, but Robyn and Meridian cut the engineering lift, standardise the core methodology, and return outputs faster, so the entry point has never been easier. The free rails still need an operating layer, and agencies can own it, refreshing MMM reads at trading cadence and wiring them into bidding and budget allocation so the model steers spend in near real time, moving from slow, backward-looking data to fast-forward moving data feeding bidding logic and custom optimisers.
Clean-room brokerage adds a new intermediary role: the agency structures data collaborations between a brand, a retailer and a platform, and charges for the deal architecture, and a snip on incremental outcomes driven by the clean room, instead of the hours.
The awkward question is who builds all this, and agentic automation supplies the answer. The people capable of designing a clean room or an identity spine are the agency’s best, and today they sit booked solid assembling reports, reconciling numbers and feeding decks, because that is what the retainer pays for.
Agent workflows take the drudgery, and the freed senior time goes towards outcome-driven value for clients and the agency: sitting with clients, diagnosing the commercial problem, and building the clean rooms, identity architecture and MMM optimisers that answer it.
The candid economics make the case on their own. No client pays an FTE fee for a clean-room architect, and no utilisation-managed floor has spare weeks to build one, which is why most agencies never deliver these outputs at scale despite every capabilities deck promising them.
The agents create the capacity, the capacity builds the products, and the products bill as fees, with inventive commercial models. The same capacity finally lets agencies deliver the combined martech-and-adtech service clients keep asking for and cannot currently buy from any single supplier: the CDP, the media activation and the measurement, wired together by people who understand all three.
Each of these is tech-and-data-fee revenue: recurring, margin-rich, owned. An agency earning a quarter of its income this way has rebalanced its P&L away from the thing AI deflates and toward the thing AI inflates. The rest will keep selling hours at whatever price the button leaves them.
Platforms: they want you to teach the machine, and prove it
The platforms profit from the gap too, which is why they keep building doors into it. A model fed margin-accurate values and rich server-side signals performs better, and a better-performing model earns larger budgets; Meta’s own CAPI numbers amount to the company saying so in public. Advertisers who invest in inputs churn less and complain less.
They also hand the platforms something their earnings calls need: case studies with a mechanism, where the story reads ‘this brand taught the system its economics and beat its category,’ instead of the perennially suspect ‘this brand spent more and sales went up.’
A defensive logic runs underneath. Platform-graded homework is a trust problem with regulatory tension, and the platforms know it. Meta and Google both released open-source measurement frameworks that let advertisers mark them independently. An ecosystem where the buy side can verify value keeps spending.
So if everything becomes a black-box ‘press button, get results’ world with no explainable cause-and-effect, that eventually hurts the platform itself because buyers lose trust and pull spend.
Adtech: the gap is the whole business model
For the independent adtech layer, the gap is the business model, whether or not the deck says so. Contextual players like Seedtag build campaign-specific models trained on an advertiser’s own brief and attention players like Lumen augment platforms with far better predictors for media quality.
Video intelligence firms like Pixability sell the placement-level YouTube detail that Google’s reporting rounds off. The clean-room and collaboration layer (LiveRamp, InfoSum and their kin) is the plumbing every brand and agency play above runs through. These companies exist because the walled gardens accept your inputs but will not build them for you.
The gap also upgrades their proof. An adtech case study today is a feature list with a logo attached. A vendor whose client deployed pretested creative into the contexts it scores best in, through a custom model, and verified the lift in a clean room, sells itself to the next twenty prospects.
Closing the gap
Each layer’s product is another layer’s input, and each layer earns more when the layer beside it invests. In an industry built on adversarial economics, that alignment is rare. Put simply: if you’re not adding a layer, you’re someone else’s input and in this market, inputs get commoditised fastest.
Later issues take each layer apart with numbers and named examples: the sameness maths, the four inputs and their price tags, the clean-room deals, the agency fee models. The claim this series will keep testing is a short one. Your advantage lives in what you know and the machine doesn’t, and the whole ecosystem now gets paid when you teach it.
Staples wound the original Easy Button down, then brought it back last year for the campaign’s twentieth anniversary. The reboot carried a different pitch. The 2005 button promised office supplies on demand. The 2025 one, personified as “E.B.”, sells print, tech support, shipping and the rest of the company’s services. Staples aimed it at higher-margin work. The advertising industry’s button offers no such reinvention, and the brands funding it can ask themselves each quarter which side of it they sit on.
Subscribe to follow the series as it takes each layer apart with numbers and named examples. If you run a brand, an agency or an adtech business, tell me which side of the button you’re on - I read every reply. And forward this to the colleague still wholly pressing the button.
This is Erfan Djazmi, fortnightly notes on the commodity gap in marketing, from someone who spent five years at IPG decommoditising agency and client investment.