A few years ago, one of the hardest questions for a business interested in artificial intelligence might have been where it could use it.
I’m beginning to think the opposite problem could become more important.
What happens when you can see AI opportunities almost everywhere?
Look across a reasonably complex business and the list can grow quickly.
Sales teams could use AI for research, prospecting, qualification, proposals and account intelligence.
Customer-service teams could use it to retrieve information, draft responses, classify requests and identify recurring problems.
Finance teams could use it to analyse information, investigate anomalies, produce reports and support forecasting.
Operations teams could use it around scheduling, quality, maintenance, documentation and process improvement.
Employees across the organisation could use AI to research, write, analyse, summarise, plan and solve problems.
Then there are opportunities that cross departmental boundaries: organisational knowledge, management information, customer intelligence, workflow automation and decision support.
And the technology keeps improving.
That creates an unusual management problem.
The number of things a business could do with AI may be increasing much faster than the number of things it can implement well.
Businesses don’t have unlimited capital. They don’t have unlimited technical resources. More importantly, they don’t have unlimited leadership attention, employee capacity or tolerance for organisational change.
So perhaps the next phase of business AI isn’t primarily about discovering more use cases.
It’s about becoming much better at choosing between them.
The more AI can do, the more important it becomes to decide what not to do.
The shortage of AI ideas may be temporary#
Much of the early conversation about enterprise AI has naturally focused on identifying applications for the technology.
That made sense.
Generative AI was unfamiliar. Leadership teams needed to understand what these systems could do and where they might be useful.
But capability has been moving quickly.
OpenAI’s GPT-4 demonstrated a significant expansion beyond the earlier generation of language models. By 2024, GPT-4o had made multimodal interaction faster and more accessible, while Anthropic’s Claude 3 family and Google’s Gemini models were pushing capability across areas including reasoning, coding, multimodality and increasingly large amounts of context.
Then, in late 2024, OpenAI introduced its o1 models, designed to spend more time reasoning through difficult problems before responding.
The important business consequence isn’t that every organisation should immediately adopt every new model.
It’s almost the opposite.
Each expansion in capability potentially expands the set of economically plausible things businesses can attempt.
Something dismissed six months ago may deserve another look.
Something requiring a specialist may become accessible to a broader group of employees.
Something that needed bespoke software may increasingly be possible using general-purpose models combined with existing systems.
And existing applications may become significantly more useful as the underlying models improve.
The possibility space is expanding.
Implementation capacity isn’t expanding at the same speed.
That gap matters.
Possibility isn’t priority#
Suppose a leadership team identifies 50 plausible AI opportunities across its business.
Perhaps ten relate to sales.
Eight to customer service.
Seven to finance.
Ten to operations.
Five to organisational knowledge.
Another ten cut across productivity, management and internal processes.
What should it do?
“All of them” is unlikely to be a serious answer.
Every opportunity consumes something.
Money.
Technical capacity.
Management attention.
Employee time.
Data work.
Integration effort.
Training.
Governance.
Process change.
And perhaps most importantly, organisational attention.
There is an opportunity cost to pursuing an AI initiative because the organisation could have been doing something else instead.
That makes the distinction between possible and worthwhile extremely important.
An AI system might be technically capable of performing a task.
That tells us very little about whether implementing it is a good business decision.
The economic value may be small.
The implementation cost may be high.
The existing process may already work perfectly well.
The necessary data may be poor.
Employees may have little reason to adopt it.
The consequences of failure may be unacceptable.
Or another opportunity elsewhere in the organisation may simply be much more valuable.
Technical feasibility answers:
Can we do this?
Business prioritisation asks:
Should we?
Those are very different questions.
AI has an attention cost#
Capital is an obvious constraint because businesses are accustomed to allocating it.
Leadership attention is easier to underestimate.
Imagine an organisation launches six AI initiatives simultaneously.
Each has a project owner.
Each requires decisions.
Each needs meetings.
Each creates questions about technology, security, processes, data and employees.
Each needs some form of implementation.
Each eventually needs training, adoption and measurement.
None appears overwhelming individually.
Collectively, they create another layer of organisational complexity.
This matters because senior management attention is already scarce.
Every hour spent resolving a marginal AI initiative is an hour not spent on customers, people, operations, strategy or a potentially more important AI opportunity.
There is therefore a hidden cost to an AI portfolio that becomes too broad.
Fragmented experimentation can create the appearance of progress while dispersing the organisation’s ability to create meaningful results.
Ten departments running ten experiments may sound more advanced than a business pursuing two carefully selected opportunities.
It doesn’t necessarily mean it is.
The better question is what changed because of them.
The most exciting opportunity may not be the best one#
AI makes prioritisation particularly difficult because the technology can be extraordinarily impressive.
A demonstration can create an immediate emotional reaction.
You see something happen that previously seemed impossible and instinctively start imagining where it could be deployed.
That’s useful. Demonstrations expand our understanding of what is possible.
But technological novelty can distort prioritisation.
Consider two hypothetical opportunities.
The first uses an advanced AI capability to automate a complicated internal activity. It’s impressive, highly visible and makes for a brilliant demonstration. But the activity occurs infrequently and doesn’t cost the organisation very much.
The second uses comparatively ordinary AI to improve the quality of information available to a sales team before important customer conversations.
It isn’t nearly as exciting.
But perhaps the organisation has £30 million of revenue moving through those conversations every year.
Which is the bigger opportunity?
We don’t know.
We’d need considerably more information.
But that’s precisely the point.
You cannot rank AI opportunities by how impressive the AI looks.
You have to understand the economics surrounding them.
That’s why the most useful starting point remains the business itself and its economics.
Where is revenue created?
Where is margin lost?
Where are people constrained?
Where does poor information create problems?
Where are customers frustrated?
Where does expertise become a bottleneck?
Which decisions have disproportionate consequences?
Where does the organisation want to grow?
Then AI capability can be considered in that context.
Value isn’t the only variable#
Even economic value isn’t enough.
Imagine two opportunities each capable of creating £500,000 of annual value.
Again, purely hypothetically.
Opportunity A can be implemented using technology the company already owns, requires relatively clean data, affects a contained workflow and could potentially demonstrate value within a few months.
Opportunity B requires significant integration, uncertain data quality, changes across multiple departments and a major shift in employee behaviour.
They’re not equivalent investments.
Potential value is only one dimension.
Businesses also need to consider questions such as feasibility, cost, implementation complexity, time to value, organisational readiness, risk and confidence in the underlying assumptions.
This means a smaller opportunity can sometimes rationally deserve priority over a larger one.
A project with lower theoretical value but a high probability of successful implementation may be a better first investment than an enormous opportunity surrounded by uncertainty.
Conversely, some difficult opportunities may be sufficiently valuable that they’re worth the complexity.
There isn’t a universal formula.
The important thing is recognising that AI opportunity selection is a business decision under constraints.
That requires judgement.
Quick wins can become a trap#
One response to uncertainty is to focus on “quick wins.”
There’s nothing inherently wrong with that.
An organisation beginning to experiment with AI can learn a great deal from relatively simple implementations.
Early successes can build confidence, develop internal capability and reveal unexpected opportunities.
But quickness isn’t the same thing as value.
If businesses aren’t careful, “find some quick wins” can become another technology-first strategy.
We start searching for things that are easy to implement rather than things worth improving.
The result can be a collection of small productivity initiatives that demonstrate activity without materially affecting the performance of the organisation.
Sometimes the best first project will indeed be quick.
Sometimes it won’t.
The better objective is probably:
Find something valuable enough to matter and achievable enough to teach you something.
That balances ambition with reality.
Saying no is part of an AI strategy#
One of the strangest consequences of rapidly improving AI is that good strategy may increasingly involve rejecting perfectly reasonable ideas.
An employee finds an interesting tool.
No.
A department wants an AI assistant.
Not yet.
A vendor demonstrates an impressive capability.
Interesting, but no.
A competitor launches something.
Still no.
Not because the organisation is resistant to AI.
Because another opportunity matters more.
This is difficult.
People naturally associate innovation with doing things.
Launching pilots.
Buying tools.
Testing products.
Creating projects.
But prioritisation means accepting that resources allocated to one opportunity cannot simultaneously be allocated to another.
A serious AI strategy therefore needs a way of distinguishing between at least three categories:
Things worth doing now.
Things worth understanding but not yet pursuing.
Things that are possible but probably aren’t worth doing.
That last category could become surprisingly large.
And that’s healthy.
The quality of an AI strategy may eventually be judged as much by the opportunities it rejects as the ones it pursues.
Different businesses should make different choices#
This is also why I’m sceptical of universal lists of “the ten AI use cases every company should implement.”
There will undoubtedly be common patterns.
Most businesses communicate.
Most sell something.
Most have customers.
Most employ people.
Most create and retrieve information.
So some AI capabilities will have broad relevance.
But economic priority is organisation-specific.
A manufacturer constrained by engineering expertise faces a different opportunity landscape from a professional-services business constrained by senior employee capacity.
A distributor with thin margins and enormous transaction volumes faces different economics from a specialist business with high margins and a small number of valuable customers.
A rapidly growing company may value capacity more highly than immediate cost reduction.
A business under margin pressure may make the opposite choice.
Even two direct competitors may rationally prioritise different AI investments because their systems, people, customers, processes and economics differ.
This is important because AI strategy cannot simply begin with generic lists of use cases.
The technology might be common.
The context in which it creates value isn’t.
AI opportunity is becoming a portfolio problem#
Perhaps the most useful mental model is to stop thinking about AI as a sequence of isolated projects.
Think of it as a portfolio of potential investments.
Some opportunities may offer high potential value and relatively low complexity.
Some may offer modest value but provide useful learning.
Some may be strategically important but technically immature.
Some may depend on another capability being built first.
Some may be worth monitoring because the technology isn’t ready.
Some may have enormous theoretical potential but unacceptable risk.
And some should simply be discarded.
That portfolio will also change.
The economics of the business change.
Technology changes.
Model capability changes.
Implementation costs change.
Employee capability changes.
What isn’t attractive today may become compelling later.
That means prioritisation cannot be a one-off exercise.
Businesses may need to repeatedly reassess the relationship between what AI can do and what the organisation needs.
This is starting to look less like technology procurement and more like management.
Better AI creates a harder decision#
There is a natural assumption that improving AI will make AI strategy easier.
In one sense it will.
More capable models will solve more technical problems.
But they may simultaneously make the management problem harder.
If AI can do five useful things, choosing where to start isn’t especially complicated.
If it can do 500, prioritisation becomes essential.
Every improvement in frontier capability can create another set of potential applications.
Every new application competes for finite organisational resources.
This is also why access to better AI isn’t necessarily the source of competitive advantage.
So one of the most valuable capabilities a business can develop may have very little to do with building AI itself.
It may simply be the ability to answer, repeatedly and with discipline:
Where does AI matter most to us right now?
That requires understanding the technology.
But it also requires understanding the business: its economics, constraints, priorities, customers, people and ambitions.
The companies that create the most value from AI may therefore not be those that pursue the most opportunities.
They may be the ones that become unusually good at choosing.
Because as the number of things AI can do continues to grow, the scarce resource won’t necessarily be possibility.
It will be the organisation’s ability to decide which possibilities deserve to become reality.