AI makes organisations better. That may be exactly the problem.

Last week I argued that leaders need to know when not to trust AI. This week asks a harder question. What happens when everyone trusts the same answers?

I did not expect to end up here. I started the second half of this series intending to write about decision-making. The research took me somewhere more uncomfortable.

Let me be precise about what follows. The evidence does not yet show that entire organisations are converging. Nobody has run that study. What the evidence does show is the mechanism. AI can improve individual output while simultaneously reducing diversity across a group.

My question is what happens when that mechanism moves from stories and ideas into strategy.

Sameness costs you margin. Importing the wrong assumptions costs you the proposition.

The experiment that changed my view

Anil Doshi and Oliver Hauser ran a controlled experiment on creative work, published in Science Advances in 2024. Writers produced short stories. Some received story ideas from a large language model. Some did not.

The AI-assisted stories were judged more creative, better written and more enjoyable. The gain was largest among the writers who started out least creative. So far, this mirrors the Harvard Business School and Boston Consulting Group consulting experiment I covered earlier in the series. AI lifts the floor.

Then came the second finding. The AI-assisted stories were measurably more similar to each other than the stories written by humans alone.

Individual creativity rose. Collective diversity fell.

Barrett Anderson and colleagues found a comparable pattern in idea generation. Participants using a large language model produced more ideas, across more categories. Those ideas homogenised across users.

Neither study looked at organisations. Both looked at people producing creative work. But read them together and a strategic question appears. Individual outputs improve. The variation between them can shrink.

If that same pattern carries into organisational strategy, it creates a very different competitive problem from the one most leaders are currently planning for.

This already has a name

I want to be clear about what is mine and what is not.

Computer scientists have been studying related ideas for several years. Jon Kleinberg and Manish Raghavan formalised one in 2021 as algorithmic monoculture. Their model showed something counter-intuitive. The same algorithm could improve decisions for an individual user while producing worse outcomes across the wider system when many decision-makers relied on it.

Individual improvement. Collective vulnerability.

Strategy researchers have approached the question from another direction. Krakowski, Luger and Raisch examined how AI changes the sources of competitive advantage in the Strategic Management Journal in 2023. More on their findings shortly, because they complicate my argument in a useful way.

So the mechanism is not my discovery. The question I think is under-asked is the commercial one. If AI converges how organisations think, what happens to competitive positioning?

Convergence is not new. What may be new is where it happens.

Businesses have converged before, repeatedly, without any help from AI:

• Total Quality Management.
• Business Process Re-engineering.
• Balanced Scorecards.
• Lean.
• Six Sigma.
• Agile.

Most created real value. Many also produced organisations that looked increasingly like one another.

Sociologists named this in 1983. DiMaggio and Powell called it institutional isomorphism. Organisations in the same field come to resemble each other through shared advisers, shared training and shared professional norms. Same consultants. Same business schools. Same conference circuit. Same frameworks.

So convergence is old news. My claim is narrower, and it is an interpretation rather than a measured finding. AI potentially changes two things about this pattern.

Speed. Practices that once had to travel through consultants, conferences, publications and training can now be surfaced almost instantly, by anyone, at negligible cost.

Location. More importantly, convergence can now enter at the point of idea generation itself, before the organisation has evaluated anything.

That second shift is the one that should concern leaders. You cannot review a strategic option you were never shown.

Porter cuts both ways

Michael Porter argued in 1996 that operational effectiveness is not strategy. Best practice rarely creates lasting advantage because competitors can copy it. Real strategy requires making genuinely different choices.

It is tempting to present AI as the arrival of Porter's warning. That would overstate it. Porter's point was that best practice was always copyable. That is precisely why he separated it from strategy in the first place.

AI has not created the problem. It may be driving the cost of copying towards zero, and moving the copying earlier in the process. A difference of degree, not of kind.

But when a difference of degree becomes large enough, it changes what strategy has to do.

Where this gets serious

Now the part that matters most in the markets I work in. The available research already gives us a reason to be cautious.

Dhruv Agarwal, Mor Naaman and Aditya Vashistha presented work at CHI in 2025 showing that AI writing suggestions can pull output towards Western styles and diminish cultural nuance. Their study was about writing. Not banking strategy.

What follows is therefore my inference, not their finding. If AI-assisted thinking carries assumptions embedded in its training and post-training environment into strategic work, those assumptions may fit some markets far better than others.

In African payments, seemingly small assumptions matter enormously:

• Whether the primary rail is card, account-to-account or mobile money.
• Whether distribution happens through branches, merchants or agents.
• Whether formal credit history exists for the merchant in front of you.
• Whether cash is marginal or central to everyday commerce.
• Whether the merchant is registered at all.

Get those wrong and everything downstream is wrong. Pricing. Onboarding. Underwriting. Channel. Partner selection. A strategy built on the wrong assumptions can be beautifully reasoned and commercially useless.

So the convergence risk here is not simply that banks converge with each other. It is that they may converge on a template built for somebody else's market.

Sameness costs you margin. Importing the wrong assumptions costs you the proposition.

I am describing a risk, not a measured outcome. But if you are building an SME merchant strategy in Accra, Nairobi or Lusaka this quarter, and the first draft came out of a frontier model, it is worth asking whose market that draft was really written for.

The objections that could sink this

Three arguments could prove me wrong. All three are stronger than the usual counterpoint about AI generating more options.

One. The model landscape is diversifying, not consolidating. Monoculture assumes everyone queries the same model with the same context. Enterprise deployment is trending the other way: multiple frontier labs, open-weight models, fine-tuning, retrieval over proprietary data. If enterprises increasingly run models grounded in their own information, convergence may weaken rather than strengthen. I do not know whether that diversification will be enough to offset convergence. That is one of the things worth watching.

Two. Strategy is constrained by assets, not by ideas. This is the objection I take most seriously, and it comes straight out of my own working week. A tier-two Kenyan bank and a tier-one Nigerian bank can receive word-for-word identical AI advice and still cannot execute the same strategy. Different balance sheets. Different licences. Different regulators. Different distribution. Different core banking constraints. If constraint binds harder than ideation, convergence in documents never becomes convergence in market behaviour.

My honest answer is that this limits the thesis rather than defeating it. Constraint shapes what you can execute. It does not restore the strategic options the model never surfaced.

Three. Convergence is self-correcting. If everyone crowds the consensus position, the contrarian position becomes cheaper and more rewarding. Markets have historically punished crowding. I think this is right. It is also the argument for acting deliberately rather than waiting. If convergence creates a premium on difference, that premium goes to whoever notices first.

A rule for what is actually scarce

I have nominated several capabilities across this series as the thing that becomes scarce: judgement, execution, reputation, calibration and, this week, original thinking.

A fair reader should push back on that. A framework that finds a new scarce asset every fortnight risks explaining everything and predicting nothing. So let me apply a test, including to my own argument.

A capability is genuinely scarce when it passes two conditions. The production test: AI cannot reliably produce it. The durability test: it resists copying once competitors can observe it.

Run the five through that.

• Judgement passes both. It is tacit, contextual and hard to transfer even when visible.
• Execution passes both. Competitors can see your operating model and still fail to reproduce it.
• Reputation passes both. It cannot be generated. It cannot be borrowed.
• Calibration passes the production test but is weaker on durability. Knowing when not to trust AI is teachable. Teachable things spread.
• Original thinking passes the production test and fails the durability test.

That last one is worth stating plainly because it works against the article I am writing. An original strategy stops being original the moment the market can see it. Porter made that point decades before anyone had heard of a language model.

So original thinking is scarce. But it is perishable. Which leads somewhere I did not anticipate.

Which is why execution matters more, not less

Earlier in this series I argued that execution becomes the differentiator. This week's argument looks like it competes with that. It does not. It completes it.

If AI converges the plans, the plan stops being where advantage lives. Original thinking buys you a window. Execution is what you do with it before the window closes.

There is also a strong reason not to conclude that AI inevitably produces sameness. Krakowski, Luger and Raisch studied chess by comparing the same players across human, AI and hybrid tournaments. They found that AI adoption made traditional competitive capabilities less valuable while creating new and persistent differences at the human-machine intersection.

One detail deserves attention. The new capability that emerged was unrelated to, and in some cases negatively related to, the player's original ability.

Read that again if you lead an organisation. The people who were best before are not automatically the people who will be best after.

Advantage did not disappear when the machines arrived. It relocated. And it relocated somewhere the previous hierarchy did not predict.

That is consistent with everything this series has found. AI may homogenise some capabilities while creating new places to differentiate. The work is finding them before your competitors do.

The question leaders should be asking

Most strategy conversations still open with the same question: "What is the best strategy?"

There is a more useful one now: "What strategy would every competitor arrive at after asking AI the same question we just did?"

Map that answer honestly. Then decide, deliberately, whether you want to be standing on it.

Sometimes the consensus position is correct and you should take it, knowing your margin will compress. Sometimes it is correct for a market that is not yours. Either way, make it a decision rather than a default.

A prediction I can be wrong about

A thesis that cannot lose is not worth much. So here is one that can, specified in advance.

• Market: Kenya.
• Sample: The five largest banks by total assets.
• Source: Published annual reports and investor presentations.
• Measure: Similarity in the stated SME and merchant propositions, assessed on language and strategic priorities.
• Claim: By the end of 2028, those five propositions will be measurably more similar to each other than they were in 2024.

I am naming the market and the method now, rather than later, so that I cannot quietly select whichever country happens to support the argument. The 2024 baseline is already published. I intend to run the comparison and report the result either way.

The bottom line

Every technology shift redistributes advantage. This one may be redistributing it in an unfamiliar direction.

The evidence we have does not suggest AI makes individual performance worse. In several settings, it improves it. The question is what happens to differentiation when those improvements spread across an entire field. Better individual outputs. Potentially narrower gaps.

If that is right, the advantage does not disappear. It relocates, to whatever the model cannot hand everyone else at the same time.

So let me leave you with this. If your competitors asked AI exactly the same strategic question tomorrow morning, how confident are you that your strategy would still look different?

---

References

Doshi, A. R. & Hauser, O. P. (2024). "Generative AI enhances individual creativity but reduces the collective diversity of novel content." Science Advances, 10(28), eadn5290. doi.org/10.1126/sciadv.adn5290

Anderson, B. R., Shah, J. H. & Kreminski, M. (2024). "Homogenization Effects of Large Language Models on Human Creative Ideation." Proceedings of the 16th Conference on Creativity & Cognition, 413–425.

Agarwal, D., Naaman, M. & Vashistha, A. (2025). "AI Suggestions Homogenize Writing Toward Western Styles and Diminish Cultural Nuances." Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems.

Kleinberg, J. & Raghavan, M. (2021). "Algorithmic Monoculture and Social Welfare." Proceedings of the National Academy of Sciences, 118(22), e2018340118.

Krakowski, S., Luger, J. & Raisch, S. (2023). "Artificial Intelligence and the Changing Sources of Competitive Advantage." Strategic Management Journal, 44(6), 1425–1452. doi.org/10.1002/smj.3387

DiMaggio, P. J. & Powell, W. W. (1983). "The Iron Cage Revisited: Institutional Isomorphism and Collective Rationality in Organizational Fields." American Sociological Review, 48(2), 147–160.

Porter, M. E. (1996). "What Is Strategy?" Harvard Business Review, November–December 1996.