The Twelve Percent
Researched by an agentic pipeline · reviewed and gated by the author
AI is simply the largest strategic bet we have ever placed through that asymmetry.
Somewhere in the last twelve months, a board approved an AI investment in under twenty minutes.
Nobody wanted to be the person who said no.
The deck was good. There was a slide on market disruption, another on competitors, another on the extraordinary pace of AI adoption.
Then came the business case.
One number was in bold:
“Full-time equivalents avoided.”
Nobody in the room was being cynical. Everyone believed they had just funded the future.
What they had funded was a cost-reduction programme in an innovation costume.
That distinction matters.
Because the question is no longer whether your company is investing in AI.
It is whether you are investing in advantage — or in cheaper sameness.
The twelve percent
PwC’s 29th Global CEO Survey asked 4,454 chief executives across 95 countries what AI had actually delivered over the previous twelve months.
Only 12% reported both lower costs and higher revenue.
Fifty-six percent reported neither.
Deloitte found the same imbalance from another direction. Among more than 3,200 leaders, 66% reported productivity and efficiency benefits from AI and 40% reported cost reduction.
Only 20% reported improvements in products and innovation.
Only 20% reported increased revenue.
Yet 74% said they expected AI to drive revenue growth in the future.
Hold those last two numbers together.
Seventy-four percent hope. Twenty percent have.
The obvious conclusion is that AI is underdelivering.
I think that conclusion is wrong.
AI is delivering, with reasonable precision, the outcome organisations have made easiest to pursue.
Efficiency.
This is not a failure of ambition
It is tempting to blame executives.
To say companies are being short-termist.
That boards understand cost reduction but lack the imagination to pursue transformation.
I don’t think that explanation survives contact with how large organisations actually make decisions.
BCG found that 60% of companies monitor no financial KPIs relating to value creation from AI. Roughly a third track nothing at all. Another quarter track operational measures such as hours saved, tickets processed or cycle time — but nothing that reaches the revenue side of the P&L.
McKinsey, meanwhile, reports that efficiency is an objective for around 80% of organisations using AI, while only 39% report any enterprise-level EBIT impact.
This creates a very specific asymmetry.
- A headcount reduction is attributable.
- It is auditable.
- It can be assigned to an executive.
- It appears inside the current fiscal year.
- A finance director can defend it in a budget review.
Now compare that with:
Our underwriters make materially better decisions.
Or:
Our product team can now explore three times as many concepts.
Or:
Customers are receiving a level of service that was economically impossible before.
Those things may be more valuable.
But they are harder to isolate, harder to attribute, slower to mature and harder to defend under challenge.
Given those instruments, cost-first AI is not irrational.
It is rational.
What your management system can see eventually becomes what your strategy becomes.
That is the real problem.
AI didn’t create it.
Large organisations have always been better at counting what they spend than valuing what they build.
AI is simply the largest strategic bet we have ever placed through that asymmetry.
There is a strong case for cost
This matters because the argument cannot be that efficiency is somehow beneath strategy.
It isn’t.
Savings are verifiable.
Claims about innovation are often not.
Boards have been burned by transformation programmes whose returns existed primarily in PowerPoint.
And cost leadership can create enormous strategic advantage.
Walmart built an extraordinary operating model around structural cost advantage.
Ryanair did too.
Amazon spent years converting operational efficiency into lower prices, greater scale and still more efficiency.
Cost reduction can absolutely be strategic.
But notice what these examples have in common.
The saving was connected to a theory of advantage.
- Walmart’s costs enabled lower prices.
- Lower prices drove scale.
- Scale created still better economics.
That is a system.
Now compare it with reducing 8% of a support organisation because an AI assistant lets the remaining employees process tickets faster.
That may produce a worthwhile margin improvement.
But if every competitor buys broadly the same technology from broadly the same vendors, the advantage disappears.
The distinction is simple:
Cost reduction is a strategy when it is attached to a theory of advantage. Without one, it is subtraction with a project plan.
Savings can finance transformation.
They cannot substitute for it.
Klarna learned this in public
Klarna became one of Europe’s most prominent AI efficiency stories after deploying an AI customer-service assistant and reducing its dependence on human support staff.
The system worked.
It handled a huge volume of customer conversations.
Response times fell dramatically.
Repeat inquiries declined.
The technology wasn’t the problem.
The evaluation criterion was.
By 2025, CEO Sebastian Siemiatkowski was publicly acknowledging that cost had become too dominant a factor in how the company organised customer service, and that quality had suffered.
Klarna did not abandon AI.
It changed what it was optimising for.
That distinction deserves more attention.
Because many companies are asking:
How much can this technology remove?
The strategically interesting question is:
What can this technology allow us to become?
Those questions produce very different companies.
The business case is choosing the strategy
This is why exhorting executives to “think bigger about AI” will accomplish very little.
The problem is structural.
If value creation remains difficult to quantify while cost reduction remains immediately auditable, efficiency will win the budget meeting.
Every time.
So the answer is not simply better technology.
And it isn’t greater executive courage.
It is a different investment instrument.
For every significant AI initiative, leaders should require four things to be explicit before funding is approved.
The efficiency case
What capacity does this release?
What gets cheaper, faster or simpler?
There is nothing wrong with this question.
We should ask it.
But it cannot be the final question.
The performance case
What becomes materially better?
Not what the tool does.
What does the business do better?
Hours saved is not enough.
- Did first-contact resolution improve?
- Did claim accuracy increase?
- Did conversion rise?
- Did product development accelerate?
- Did customer effort fall?
Measure the outcome experienced by the business or customer, not the activity performed by the tool.
The advantage case
If this works exactly as planned, why does it matter competitively?
What becomes possible for us that wasn’t possible before?
- Does it strengthen scale?
- Increase switching costs?
- Improve quality?
- Compress learning cycles?
- Enable a new product?
- Change the economics of serving a segment?
And one uncomfortable question:
What part of this advantage will still exist when our competitors have access to the same model?
If the answer is “none,” you may have an efficiency programme.
You probably do not have a strategic advantage.
The reinvestment case
This is the most neglected part.
What happens to the capacity we release?
If ten thousand hours disappear from a workflow, where do those ten thousand hours go?
If $10 million comes out of a cost base, what happens to the $10 million?
If you don’t decide before the saving arrives, the answer is usually predictable.
It disappears into the P&L.
That can be appropriate.
But it should be a decision, not a default.
Pre-commit some portion of AI-created capacity to the capabilities, products, experiences or business models that create the next advantage.
Otherwise yesterday’s efficiency becomes tomorrow’s baseline — and nothing fundamental changes.
Change the questions and you change the programme
- A cost-first AI programme asks: Which roles can we reduce?
- A value-first programme asks: Which outcomes can we materially improve?
- A cost-first programme asks: How many hours did we save?
- A value-first programme asks: How much better did the decision, product or customer experience become?
- A cost-first programme asks: Which tasks can AI perform?
- A value-first programme asks: Which workflows should no longer exist in their current form?
- A cost-first programme asks: Which tool should we buy?
- A value-first programme asks: What system should we redesign?
And perhaps most importantly:
- A cost-first programme asks: How fast is the payback?
- A strategic programme also asks: What advantage remains 36 months from now?
This is not semantics.
Questions determine metrics.
Metrics determine investment.
Investment determines organisational behaviour.
And organisational behaviour eventually becomes strategy.
Four questions for the next board meeting
So before approving the next major AI initiative, I would ask four questions.
- If AI could not reduce headcount at all, would this business case still be exciting?
- What proportion of our AI KPIs measure subtraction — and who owns the metrics measuring what becomes better?
- Which core workflow have we actually redesigned, rather than simply made faster?
- Where has a quality standard gone up because of AI, and can we prove it?
If the first question kills the business case, that doesn’t necessarily mean the programme is bad.
But call it what it is.
You don’t have an AI strategy.
You have a restructuring programme enabled by AI.
Those are not the same thing.
The advantage problem
The next decade will not reward the companies that automated their org charts fastest.
Every serious competitor will have access to capable models.
Most will buy broadly similar tools from broadly similar vendors.
Most will discover broadly similar efficiencies.
Efficiency will increasingly become table stakes.
The interesting question is what companies do after efficiency.
- Which organisations use the released capacity to improve judgment?
- Which create better products?
- Which redesign customer experiences?
- Which raise standards that were previously uneconomic?
- Which discover entirely new ways of operating?
- And which simply become slightly cheaper versions of what they already were?
The technology will not answer that question.
Your management system will.
“Efficiency is a tactic. Advantage is a strategy. Confuse the two, and AI becomes an extraordinarily expensive way to stay ordinary.”
Evidence referenced
PwC, 29th Global CEO Survey; Deloitte, State of AI in the Enterprise; BCG, AI Radar; McKinsey, The State of AI; public reporting and commentary regarding Klarna’s AI customer-service strategy.