Creative & Content

Ready for the influx: a procurement guide to AI-labelled creative opportunities

The AI creative supplier landscape is moving faster than most SOW templates. The governance and partner-selection goalposts have shifted in ways that didn’t exist in 2023. Here is an evidence-based toolkit for the conversations procurement should be having before the contracts arrive — grounded in WFA, ISBA and MIT data, and in the failure modes already showing up in the market.

Every marketing procurement lead reading this is about to be approached — if they haven’t been already — by a queue of vendors carrying the letters AI in their pitch deck. Some of those vendors will be substantial. Most will not.

This isn’t an argument against AI in marketing. The shift in what creative production can deliver in 2026 is real, the cost compression is real, and the pace of change is real. The argument is narrower: the diligence questions that worked on a 2022 production supplier do not yet cover what a 2026 AI creative supplier looks like.

The intent of this piece is to lift the lid on the new pitfalls, surface what the current evidence is telling us, and give procurement a checklist to navigate the shift.

The data on what happens when organisations get this wrong is no longer anecdotal. It is published, it is from the bodies this audience already reads, and it is sobering.

What the evidence is telling us

Start with the marketing procurement community itself. The WFA Sourcing Forum’s August 2025 benchmark survey of 54 marketing procurement leaders — collectively responsible for around $97bn of marketing spend — found that 81% described themselves as tactical or beginners on AI, with only 4% identifying as advanced.

Optimism is also softening: the proportion who are “very positive” about GenAI’s impact on their role dropped from 31% in December 2024 to 23% in August 2025. Almost three in four lack KPIs to monitor AI/GenAI impact, although roughly 70% now have frameworks in place to guide its use. The picture is of a function moving carefully, methodically, and — honestly — still figuring out how to measure what good looks like.

The contracting picture tells a sharper story. ISBA’s July 2025 Generative AI member survey found that the proportion of UK advertisers running at least one live GenAI use case had more than quadrupled in 15 months — from 9% in April 2024 to 41% in July 2025.

Adoption is moving fast. Contracting is not. Across the same period, the proportion of advertisers who had updated agency contracts to include AI terms moved from 8% to 10%, with a further 37% in progress. Concern is high — WFA’s 2024 research found 80% of brands worried about how their creative and media agencies were using GenAI — yet the documentation gap is widening rather than closing.

Use cases are running ahead of the paperwork they require, and the gap is now the central operational risk in this category.

The wider AI investment data sets the context for why this matters now. MIT’s 2025 GenAI Divide: State of AI in Business study, based on 150 leadership interviews, 350 employees and analysis of 300 public AI deployments, found that 95% of enterprise GenAI pilots deliver no measurable impact on profit and loss.

S&P Global’s 2025 research adds the trajectory: 42% of companies abandoned the majority of their AI initiatives in 2025, up from 17% the previous year. The abandonment rate more than doubled in twelve months — not because AI got worse, but because the gap between vendor claims and operational reality became impossible to ignore.

MIT’s researchers were direct about the cause, quoting one CIO summarising what most of the market now knows: “We’ve seen dozens of demos this year. Maybe one or two are genuinely useful. The rest are wrappers or science projects.”

This is the supplier landscape procurement is being asked to evaluate. It is not the dot-com bust, but it shares a structural feature with it: a wave of well-funded entrants, many of whom will not survive the shake-out that is now visibly underway. The job is to tell the difference — and to do it before, rather than after, the contract is signed.

And the conversation procurement isn’t always in

None of this is happening in a vacuum. By the time a vendor contract reaches a procurement desk for review, the strategic direction has often already been set in rooms procurement was not in.

Three forces are pushing on the C-suite at once. Strategic advisers are in CEO and CFO ears about AI transformation, framing the agenda before the diligence work begins. Marketers — reading the same headlines and trade publications as everyone else — are arriving at meetings already half-sold, often with vendor decks they collected themselves.

And IT is trying to keep track of an explosion of unsanctioned tools, untracked subscriptions paid on individual cards or department budgets, and platform sprawl, mostly in defensive mode rather than orchestrating mode. The result is pressure and chaos, in roughly equal measure.

This isn’t a failure of marketing. Marketers are operating in exactly the same noise procurement is, just from a different angle. The promise that AI tools will deliver everything for nothing is in their inbox every morning. It would be strange if some of it didn’t land.

The harder fact for procurement is that by the time the conversation reaches the contract stage, the political room to ask hard questions has often already shrunk — because the strategic direction has often already been endorsed, somewhere upstream.

Which is why procurement’s most valuable move on AI right now is not better diligence at the contract stage. It is earlier diligence, before the contract stage exists. The questions that follow are calibrated to that purpose: not a checklist for signing day, but a set of conversations procurement should be having with marketing, finance and IT in the months before any of them sign anything.

The internal trap: marketers don’t want to make assets in machines

There is a second pitfall, less discussed than vendor stability, that hits after the contract is signed.

A meaningful portion of the AI creative tooling on the market today — SaaS platforms with bright dashboards, prompt fields and template libraries — is sold on the implicit promise that marketing teams will operate them in-house. That is the unit economics of the model. License the platform, train the team, scale the output.

The promise depends on a behaviour change that mostly does not happen.

The evidence on in-house creative adoption is consistent across multiple bodies of research. The ANA’s most recent in-house agency study — the largest of its kind, based on 162 respondents — found that 88% of in-house teams reported their workload had increased in the past year, with 67% saying it had increased “a lot.” The same study found that 92% of brands with in-house agencies still rely on external agencies primarily for bandwidth reasons: the internal team is simply too busy.

WFA’s own 2025 research into how in-house agencies are actually using AI, conducted with 30 global in-house agency leaders, found that only 17% have fully integrated AI into their operations, while 61% remain stuck in early testing. The gap between the platform demo and the operational reality — teams already stretched, running at capacity, and unable to absorb a new craft on top of an existing brief load — is not an edge case. It is the norm.

The more honest pattern that emerges in conversations with marketing teams who have tried this: a small group of enthusiasts use the platform extensively for a few months, novelty fades, the dashboards quieten, and the licence renews automatically.

The work that the platform was supposed to absorb either reverts to the agency or quietly disappears off the brief sheet. Internal resources lose motivation to push creative boundaries on a tool they did not ask for and were never given proper time to master.

Shadow-AI use of consumer tools picks up the slack — MIT’s study found over 90% of workers report using personal AI tools at work, against only 40% of companies with an official enterprise subscription.

This is not a marketer-blame argument. It is a structural observation. The skills, time, and craft motivation needed to make AI creative tools deliver inside a marketing team are simply not in the marketing team’s job description, and asking them to acquire all three on top of their existing role is a category error.

The procurement consequence is direct: the negotiated saving on agency fees rarely lands cleanly. WFA’s 2025 Global Content Production survey found two-thirds of brands have changed their agency model in the last four years, with hybrid in-house/outsourced arrangements rising sharply — a sign that few of these transitions are settling cleanly into the originally-promised economics.

The work doesn’t actually move in-house. It quietly redistributes — back to agencies, into shadow tools, off the brief sheet altogether.

Six questions to ask before signing any AI creative contract

Procurement does not need a new playbook for AI. It needs a sharper version of the diligence it already does for any supplier, calibrated to the failure modes the current evidence describes — and started earlier than a contract review. Six questions are doing most of the work in the conversations going well.

1. What is the underlying technology stack — and is it yours, or someone else’s?

A meaningful share of AI creative platforms in the market today are, as MIT’s researchers put it bluntly, wrappers. A user interface sitting on top of a foundation model the vendor neither owns nor controls.

Ask directly: what proportion of the fee you are charging us is for your IP, and what proportion is access to a third-party model? If the foundation model changes its pricing, terms, or availability, what happens to our service?

2. What is the operational continuity story?

Production companies were durable. Their cost base was physical, their failure modes were slow and visible. SaaS-shaped vendors fail differently. They pivot, get acquired, lose model access, change pricing tiers overnight, or simply switch off.

With 42% of AI initiatives abandoned in 2025 and the rate accelerating, vendor stability has moved from a tail risk to a working assumption.

Ask for the financial runway, the customer concentration, the contractual exit terms, and the plan if the platform changes hands. A serious supplier will have answers ready.

3. Who actually does the work — and is that supportable?

This is the question that exposes the in-house assumption. If the commercial model assumes our marketers will be operating the tool, what is the realistic time commitment per asset, what training is included, and what evidence does the vendor have of similar teams sustaining usage past the first six months?

If the realistic answer is that it takes a half-day of marketer time to produce what an agency would produce in two hours, the saving is illusory. Insist on a deployed-customer reference where the in-house adoption pattern has held up over time, not a launch-week case study.

4. What is the framework, not the pilot?

MIT shows 95% of pilots deliver no P&L impact. The pattern is clear: scattered small pilots are how organisations guarantee they will not learn anything, because no individual experiment is large enough to merit the investment that would actually make it work.

Procurement should push back on the small-pilot reflex and ask: who is accountable for the realised outcome of this investment, and what is the framework that holds them to it? Marketing is the most opportunity-rich and risk-rich function for AI in most businesses.

It deserves a serious framework, not a budget line that any individual marketer can lose without anyone noticing.

5. What are the AI-specific clauses in the SOW?

Most creative production SOWs were drafted before generative AI was a commercial reality. They do not cover synthetic-media disclosure, training-data provenance, talent likeness rights, indemnity on generative output, or data residency on the inputs.

With only 10% of UK advertisers having amended their agency contracts for AI and a further 37% still in progress (ISBA, July 2025), the gap is the rule rather than the exception. The good news for UK advertisers is that the work has been started.

The 2025 ISBA/IPA Creative Services Framework Agreement — the industry-standard contract template, fully overhauled for the first time since 2015 — includes new generative AI clauses and is freely available to ISBA and IPA members. It is not the whole answer, but it is materially better than starting from a 2020 SOW and patching.

6. What does our organisation own, and what walks out of the room if the supplier does?

Two failure modes sit beneath this question, and procurement is the function uniquely placed to address both.

The first is the live one: when output isn’t right. With a pure-platform supplier, the operator (often the marketing team) is the accountable party — the supplier’s exposure typically ends at the prompt window, and remediation falls back on the buyer.

With a service partner, the supplier carries delivery risk, output indemnity, and the cost of rework. Neither model is wrong, but they allocate exposure to the buying organisation very differently. The question to put on the table early is: in the failure mode where output disappoints, who pays for the rework, and on what timeframe?

The second is structural: what happens when the supplier itself changes. With 42% of AI initiatives abandoned in 2025 and the rate accelerating, supplier discontinuity is now a working assumption rather than a tail risk.

The procurement clauses that handle this — output ownership, IP rights over fine-tuned models or trained brand assets, data portability, escrow arrangements for proprietary models, migration support obligations — are well established for traditional SaaS but conspicuously absent from most AI creative SOWs.

They should be standard. The principle is simple: when the supplier changes, the organisation should keep what it paid for.

The opportunity hiding inside the evidence

The numbers in this article are alarming on their face: 95% of pilots delivering no P&L impact, 42% of programmes abandoned in a single year, only 10% of UK advertisers with AI clauses in their agency contracts. They sound like reasons to slow down.

They are not. They are the strongest possible argument for procurement to lean in earlier and harder than it has on previous technology waves.

The 5% who succeed are not succeeding by accident. MIT’s research is consistent on what they do differently: they buy from specialist vendors rather than building in-house (a 67% success rate versus a third of that for internal builds), they tie pilots to revenue-focused objectives from day one, and they bring procurement, IT and security into the conversation before the pilot starts rather than nine months in.

Procurement is the function uniquely placed to drive that pattern. Marketing wants the capability. Finance wants the saving. Legal wants the protection. Procurement is the only office that sees all three sides of the trade and can make the diligence stick.

As Laura Forcetti, WFA’s Director of Global Marketing Sourcing, has put it, “deeper collaboration with marketing procurement can be a quiet competitive advantage for CMOs — and a powerful ally.” The AI moment is exactly the situation where that ally is most useful.

There is a version of the next twelve months where marketing procurement spends them mostly cleaning up after engagements that were mis-scoped from the start.

There is another version where it spends them as the strategic function that brought the right partners and the right risk allocation to the table — translating commercial, legal and continuity risk into terms the business could stand behind.

The evidence is in.

Ask the six questions — and ask them early, before the strategic direction is set without you. The 5% that succeed will be the organisations whose procurement teams asked them first.

Luke Beauchamp, COO Novai Creative & Technology Group