Most companies aren’t actually cutting costs
PwC’s 29th Global CEO Survey ran across 4,454 chief executives in 95 countries, with fieldwork between 30 September and 10 November 2025 and publication in January 2026. It asked what AI had actually done to revenue and cost over the previous twelve months.
The answer people quote is that 56% saw neither higher revenues nor lower costs. The answer almost nobody quotes sits one line further into the report: 26% of CEOs say costs decreased because of AI, and 22% say costs increased.
Read that as a ratio. For every five companies that got the cost reduction they were aiming at, roughly four got the opposite. The single largest group in PwC’s matrix — 42% — reports no change on either axis at all. Only 12% moved both.
So the strategy most firms are running is failing on its own terms for roughly half of them. Most of the AI conversation assumes the saving is the easy part. It is not.
The saving that does land is a commodity
Assume you are in the 26%. You genuinely took hours out.
You bought your tools in the same quarter as everyone else in your market, from the same vendors, at roughly the same price. Whatever discount you extracted, your competitors extracted a version of it at the same time. A cost advantage that arrives simultaneously across an entire market is not an advantage. It is a new baseline, and baselines get priced in.
That last part is the bit agencies underrate. Clients can see this too. In twelve years inside agencies I never watched a visible efficiency gain survive two renewal cycles without showing up in the negotiation. You can hold the saving for a while. You are renting it, not owning it.
Which leaves one question that actually matters: what is the freed capacity pointed at?
Why the capacity never survives contact with the quarter
Two findings from the same PwC survey explain the failure better than any framework.
The first is about attention. Chief executives report devoting 47% of a typical schedule to work with a horizon under a year, and 16% to anything beyond five years — a three-to-one split in favour of the immediate.
The second is about machinery. PwC asked about six established innovation practices. Taking each on its own, roughly a quarter of CEOs said their company strongly does it: 26% tolerate high risk in innovation projects, 24% have routine processes for stopping underperforming R&D work, 23% run a defined innovation centre, incubator or corporate venturing division. Stack the practices up and the picture gets starker — fewer than one in ten companies (8%) have five or more of the six in place.
Put those together and the mechanism is obvious. Attention runs short-term, and in nine companies out of ten there is no functioning apparatus for turning recovered time into anything new. So when AI takes four hours a week out of a delivery team, nothing is waiting to catch them. The hours go where hours always go: into the work already in the queue.
That is not a motivation problem. It is missing plumbing. BCG’s 2026 work on cost advantage reaches the same place from the finance side — without a clear plan and tracking infrastructure, efficiency gains evaporate before they ever reach the P&L. Their recommendation is to sequence it deliberately: use early, provable savings to fund the deeper reinvention rather than banking them.
How to turn AI savings into agency innovation capacity
The mechanism has three parts, and each has a moment when it must be decided. Miss the moment and the part does not work.
- Before the saving exists — agree what you are measuring. Pick one named workflow and record hours in and out, before and after. Not “the team feels less swamped.” If you cannot state the number, there is nothing to allocate and everything downstream is theatre. (The workflow redesign that produces a real saving in the first place is its own discipline — I’ve written about that separately.)
- At the moment it lands — split it before anyone sees it. This is the part that does the work and the part that gets skipped. Fix the share the day the saving is confirmed: a fifth, a third, whatever the cash position bears. Capacity that enters the general pool does not come back out. Splitting it later is not a plan, it is a wish.
- After it is ring-fenced — give it an owner, a brief and an end date. Not “innovation,” which commits nobody. A named person, one defined question, a date by which you will know. Productise the retainer you keep rebuilding from scratch. Build the diagnostic you currently give away in pitches. Test the service line three clients have already asked for. And decide in the same conversation what evidence, on what date, means you stop — because a quarter of companies have that discipline and the rest end up funding zombies.
One more thing, and it is the one that fails quietly. There will be a crunch. Name now, in writing, the two or three situations that genuinely justify pulling capacity back into delivery. Everything else is off-limits. Decide it under pressure and you will decide it in favour of the invoice, every time.
Connected service — Business Strategy
Where recovered capacity goes is a strategy decision, not a tooling decision, and it is worth making deliberately while the saving is still yours to allocate. Business Strategy is a scoped engagement: I establish what your AI efficiency is genuinely worth at process level, where that capacity should be pointed given your market position, and what mechanism holds it in place once delivery pressure returns. Pricing is transparent — €2,000–3,500 depending on scope. See how Business Strategy works → or book a discovery call →.
Written by Daniel Nagy, founder of Uniquefield — fractional operations & strategy for agencies and founder-led teams, Budapest.
Sources: PwC — 29th Global CEO Survey, “Leading through uncertainty in the age of AI”, January 2026 · PwC press release, January 2026 · BCG — How Leaders Build an AI-First Cost Advantage, 2026