Imagine two £20 million businesses competing in the same market.
They sell broadly similar products to similar customers. They employ similar numbers of people. Their margins are comparable. Neither has some extraordinary technological advantage over the other.
And both have access to exactly the same AI.
The first business does what many companies are beginning to do.
Employees use ChatGPT. Microsoft Copilot starts appearing inside the software they already use. Marketing experiments with generating content. Developers use AI to help write code. Someone investigates a customer-service chatbot.
There are workshops.
There are pilots.
There is probably an AI working group.
Before long, the company can reasonably say that it is “using AI”.
The second business starts with a different question.
Not:
Where can we use AI?
But:
Where could AI create the most value?
It looks for expensive decisions that could be improved. Bottlenecks created by scarce expertise. Revenue opportunities constrained by the economics of serving customers individually. Repetitive cognitive work consuming expensive people’s time. Valuable analysis that doesn’t happen because the answer isn’t worth the cost of producing it.
Then it prioritises the opportunities with the greatest economic potential and starts implementing them.
For the first few months, perhaps there isn’t much difference between the two companies.
But give them three years.
Do they still look like the same business?
Everyone can buy the technology#
One of the unusual things about the current AI wave is how widely available the underlying capability is becoming.
OpenAI, Anthropic and Google are competing to provide increasingly capable models through consumer products, enterprise offerings and APIs.
Microsoft is embedding AI throughout its software.
Google is doing the same.
Thousands of software companies are building on top of the same underlying models.
That creates an interesting competitive problem.
If your company can access powerful AI, your competitors probably can too.
The model itself therefore doesn’t automatically create an advantage.
A company announcing that it uses GPT-4o tells us very little about whether it is becoming a better business.
Eventually, saying your company uses AI may sound a little like saying your company uses the internet.
Of course it does.
The interesting question is what you’ve built with it.
This is why the increasing accessibility of AI may produce a slightly counterintuitive effect.
The more widely available powerful AI becomes, the less access to AI differentiates one business from another.
The competitive advantage has to come from somewhere else.
Adoption isn’t value#
We’re going to hear a lot about AI adoption.
How many employees are using AI?
How many have access to Copilot?
How many AI projects are underway?
How many processes contain an AI component?
How much usage has increased?
These are reasonable things to measure.
But none of them is the outcome a business ultimately cares about.
Imagine one company gives an AI assistant to every employee.
Usage is extraordinary.
People summarise documents faster. Emails take less time to write. Meetings get transcribed. Presentations are produced more quickly.
There is probably value in that.
Now imagine another company uses AI in only three areas.
In the first, it materially increases the conversion rate of a high-value sales process.
In the second, it allows a customer-service operation to resolve substantially more enquiries without reducing service quality.
In the third, it improves an expensive operational decision enough to increase gross margin.
Which business has the better AI strategy?
We don’t have enough information to know.
And that’s the point.
AI usage alone doesn’t tell us.
The important AI metric isn’t how much AI a business uses. It’s how much value the business creates because of it.
That sounds obvious.
I suspect it will prove surprisingly easy to forget.
Start with the economics#
Technology naturally encourages us to start with capability.
Here’s what the model can do.
Where could we use it?
Which teams might benefit?
What could we automate?
But businesses don’t create value by collecting technological capabilities.
They create value by improving economics.
Revenue increases.
Costs fall.
Margins improve.
Customers receive something better.
Risks reduce.
Decisions improve.
Constraints disappear.
People become capable of producing more valuable work.
Entirely new products or services become economically possible.
If I were looking for AI opportunities inside a business, that’s where I’d start.
Not with a list of AI use cases.
With a map of where value is created, where it is lost and what currently constrains more of it from being produced.
Then ask where AI changes the equation.
Perhaps a business has a sales process where every additional percentage point of conversion is worth hundreds of thousands of pounds.
Perhaps highly paid specialists spend a meaningful part of their week extracting information from documents.
Perhaps thousands of customer conversations contain commercially useful information that nobody can economically analyse.
Perhaps a company could offer customers something individually tailored, but the human effort required makes the service uneconomic.
Perhaps an important operational decision is repeatedly made using incomplete information because obtaining a better answer costs too much.
These are business problems before they are AI opportunities.
AI becomes interesting when it changes their economics.
The same AI can produce different outcomes#
This helps explain why two businesses with access to identical technology can achieve completely different results.
Technology doesn’t arrive in a vacuum.
It arrives inside an organisation.
That organisation has its own data.
Its own processes.
Its own customers.
Its own expertise.
Its own technology.
Its own incentives.
Its own constraints.
Its own ability — or inability — to change how work gets done.
A generic AI model knows none of this automatically.
The value emerges when general capability meets specific context.
A company with unusually good proprietary information may be able to create something its competitors cannot easily reproduce.
A company that deeply understands a particular customer problem may recognise an opportunity others miss.
A company with poor processes may simply use AI to perform a bad process faster.
A company capable of redesigning the process itself may get a completely different result.
This is why AI implementation cannot ultimately be separated from business understanding.
The hardest question in business AI may not be what AI can do. It may be where a better capability would create the most economic value.
Answering that requires considerably more than access to a model.
Small advantages can compound#
The competitive impact doesn’t necessarily arrive as one enormous transformation.
That may actually be the less likely scenario.
Imagine instead that one of our two £20 million businesses becomes slightly better at applying AI every quarter.
Its salespeople research opportunities more thoroughly.
Its developers produce software faster.
Its customer-service team resolves enquiries more efficiently.
Its managers can interrogate information that previously required somebody to prepare a report.
Its marketing becomes more personalised.
Its internal knowledge becomes easier to access.
Its people spend less time moving information between systems.
Its products improve a little faster because experimentation becomes cheaper.
Perhaps none of those improvements is particularly threatening in isolation.
But competitive advantage often works through accumulation.
A slightly higher conversion rate.
A slightly lower cost to serve.
A slightly faster product-development cycle.
A slightly better customer experience.
A slightly more informed decision.
Then again next quarter.
And again the quarter after that.
Three years later, what looked like a collection of marginal improvements may have created a meaningful gap.
The AI gap may emerge gradually and then become visible suddenly.
That’s what I think businesses should be more concerned about than dramatic predictions of entire industries disappearing overnight.
Your competitor doesn’t need to invent a revolutionary new form of artificial intelligence.
They may simply need to get better at applying the intelligence that already exists.
The advantage may move outside the model#
There is another consequence if frontier AI models continue improving while remaining broadly accessible.
The things surrounding the model become more important.
Proprietary data matters.
Distribution matters.
Customer relationships matter.
Domain expertise matters.
Workflow design matters.
Integration matters.
Leadership matters.
The ability to implement change matters.
Knowing which problem is worth solving matters.
In other words, the more AI capability becomes something businesses can simply buy, the more competitive differentiation may move into the organisation applying it.
This is good news for established businesses in one respect.
They already possess assets an AI model doesn’t.
They have years of customer interactions.
Industry knowledge.
Operational experience.
Processes.
Relationships.
Data.
Brand.
People who understand the peculiarities of the market.
The opportunity isn’t necessarily to throw those advantages away and become an “AI company”.
It may be to ask what happens when those existing advantages are combined with increasingly capable intelligence.
That is a much more interesting strategic question.
There will be plenty of wasted money#
None of this means every business should race to deploy AI everywhere.
Quite the opposite.
Periods of technological excitement create enormous amounts of bad investment.
Companies will buy AI products because competitors have them.
They’ll automate things that shouldn’t be automated.
They’ll run pilots without knowing what success looks like.
They’ll deploy technology into processes that weren’t particularly good in the first place.
Some projects will technically work and still produce almost no economic value.
And plenty of companies will confuse activity with progress.
That is precisely why value needs to become the organising principle.
The objective isn’t to maximise the number of AI projects.
It isn’t to maximise AI adoption.
It isn’t even necessarily to maximise automation.
The objective is to identify where the technology can produce meaningful business impact and then become unusually good at turning those opportunities into reality.
The winners may not be the companies that use the most AI.
They may be the ones that become best at deciding where it matters.
The difference is what you do with it#
Return to our two businesses.
Three years have passed.
Both have had access to broadly the same underlying AI.
The first has spent those years using it.
Employees are more productive in places. Plenty of software now contains AI features. There have been successful projects and failed ones.
The company has undoubtedly benefited.
The second business has spent the same period systematically learning where AI changes its economics.
It has looked for the places where better intelligence improves valuable decisions, removes expensive constraints, increases revenue, lowers costs and creates better experiences for customers.
Some experiments failed.
Some opportunities weren’t worth pursuing.
But the organisation became progressively better at identifying the ones that were.
That’s the difference I would be watching.
Because if increasingly capable AI becomes broadly available, simply possessing the technology cannot be the long-term advantage.
The advantage has to come from what the business does with it.
And that changes the question every leadership team should be asking.
Not:
How much AI are we using?
But:
Where can this technology create the most value in our business?
Your competitors don’t need better AI than you. They just need to become better at turning the same AI into value.