AI is changing digital advertising. The buying model is changing with it

Digital advertising spent two decades automating itself. Now AI is making the last manual pockets cheaper, faster, and less central to the job.

The shift is already in the numbers. US digital ad revenue hit $259 billion in 2024, up 15 percent year on year per IAB and PwC. Programmatic reached $134.8 billion, up 18 percent. Retail media hit $53.7 billion, up 23 percent.

These figures describe a large, growing market. They do not prove AI caused the growth. But they show the environment AI is walking into: more inventory, more platforms, more data, and more pressure to explain where the money actually went.

The scale is already here

IAB Europe's 2025 AI advertising survey: 85 percent of respondents already using AI-based marketing tools. Nearly three quarters said at least one campaign function was AI-powered. Targeting and content generation led the charge.

The same survey exposed a gap. Sixty-eight percent of companies had general internal AI guidelines. Only 43 percent had guidelines specifically for marketing and advertising. Privacy topped the concern list. Internal expertise and training were also major barriers.

That gap matters. The industry is not deciding whether AI enters advertising. It is deciding how much of the operating model can shift before the guardrails catch up.

That gap matters. The industry is not deciding whether AI enters advertising. It is deciding how much of the operating model can shift before the guardrails catch up.

Three phases of the change

Phase one: assistance

AI began as a faster junior. It drafted copy, summarised reports, resized assets, suggested audiences, answered campaign questions. A person still pulled the trigger. The gain was time.

This phase is now baseline. The real distinction is not whether a team uses an AI tool. It is whether the output enters the work without a named owner checking it.

Phase two: variation

Generative AI rewrites the economics of testing. IAB's 2025 video report: 86 percent of buyers using or planning generative AI for video creative. They expect AI-generated creative to hit 40 percent of all ads by 2026.

Buyers are building audience versions, visual-style swaps, context-specific cuts. Practical uses. They collapse the distance between hypothesis and test. But the real shift is what happens next.

Phase three: adaptation

The next phase is not more variants. It is a system that reallocates budget, rewrites creative, retargets audiences as evidence arrives.

An agent shifts spend between campaigns. A creative system spins a new version when a hook fails. A planning system merges product, audience, outcome data before a media plan is signed.

The decision stays commercial. The system just acts on it more often.

Creative becomes a testing system

AI makes production abundant. That creates a selection problem.

When a team can produce ten video versions instead of two, the work does not end at export. Each version needs a hypothesis: a different opening, proof point, product demo, offer, audience context.

The IAB report: buyers use generative AI for audience versions, visual-style swaps, contextual cuts. More useful than making one finished ad cheaper. It turns creative into a test set. The risk is volume without learning. A hundred variants do not make a campaign intelligent if nobody knows which variable moved or whether the shift mattered.

Inflexion's view: give each creative test a job. If the brief cannot say what the version tests, it is production activity, not learning.

Media buying becomes continuous

Most media operations still run on a human calendar. Plan, launch, review, adjust in the next meeting. AI collapses the time between those decisions.

Automated bidding is not new. What changes is signal volume: product availability, retailer search position, conversion quality, audience response, creative fatigue, viewability, supply path, business outcomes. The machine responds continuously. The planning question moves upstream. Instead of which bid to change, the team defines the outcome, the permitted trade-offs, the conditions requiring human review.

CTV makes this visible. IAB: buyers expected 47 percent of CTV inventory biddable, up from 34 percent. Yet over 80 percent of digital video buyers still wanted human assistance from sell-side partners when activating CTV programmatically.

Automation is not independence. Buyers want programmatic control and the judgement of a partner who knows the inventory.

Measurement gets harder before it gets better

AI can make a campaign look more efficient while making the explanation murkier.

More automated decisions mean more possible causes for an outcome. A conversion rate shifts because of audience, creative, bid, product page, retailer algorithm, inventory, seasonality, or a mix. Privacy adds another constraint. IAB and PwC's 2024 report flags data security, interoperability, integration as ongoing problems. It also notes 70 percent of agencies, brands, publishers had not yet fully scaled AI across media planning, activation, analysis.

The measurement response is not a larger dashboard. It is a clearer evidence chain:

1. What decision did the system make?
2. Which signal caused it to make that decision?
3. What changed in the market at the same time?
4. What outcome moved, and over what period?
5. Could the same result have happened without the intervention?

That last question is the hard one. Attribution describes what happened after exposure. Incrementality asks what the exposure changed. The gap between them widens every time automation adds another decision layer without an experiment design to isolate the effect.

Inflexion's view

We do not see AI replacing media expertise. We see it shifting where expertise applies.

Senior time should not go to copying reports between platforms, resizing creative files, moving budgets because a dashboard changed colour. Those tasks automate.

People should decide: what the campaign achieves, which evidence deserves confidence, what trade-offs the system may make, when the apparent result is too thin to act on.

That is why our media work starts with the operating model, not the tool. We map signals, separate paid and organic effects, define decision rules, then automate where it improves the work. The advantage is not a more impressive model. It is a clearer line between business objective, media action, and the evidence that follows.

What advertisers should do now

Start with work that is repetitive and measurable. List the campaign decisions your team makes every week. Separate decisions needing judgement from tasks needing only speed. Check the data feeding each automated recommendation. Set explicit limits for budget, brand safety, inventory, audience use. Build a creative testing plan before generating asset volume. Report the intervention and the outcome together.

Do not begin by buying another platform. Begin by writing down what the system may do and what it must never decide alone. AI will make digital advertising more adaptive. It will also make weak objectives, incomplete data, poor measurement move faster. The advertisers who benefit will be the ones improving the decision system, not just adding tools.