There is a question being asked in businesses everywhere:
What are our AI use cases?
It’s a reasonable question.
A new technology becomes available, so organisations naturally start looking for places to use it.
Could AI write our marketing content?
Could we build a customer-service chatbot?
Could it summarise meetings?
Could it help employees search documents?
Could salespeople use it to write emails?
Could developers use it to generate code?
Before long, somebody has a spreadsheet containing fifty potential AI use cases.
They get scored.
Prioritised.
Turned into pilots.
Some work. Some don’t. A few save people time.
The company is now “doing AI”.
But there’s a problem with this approach.
There is no particular reason why the AI applications that are easiest to imagine are also the places where AI can create the most value.
In fact, starting with AI may be causing businesses to look at the opportunity backwards.
Instead of asking where we can use AI, perhaps we should start by asking where our business creates and loses economic value.
Then investigate whether AI can change the equation.
Use cases start with the technology#
Imagine gathering the leadership team of a business into a room and asking everyone to identify potential uses for AI.
What happens?
People naturally begin with things they already know AI can do.
Someone has used ChatGPT to draft an email, so content generation appears on the list.
Someone has seen a customer-service chatbot, so customer support gets added.
Meeting transcription.
Document summarisation.
Internal search.
Marketing copy.
Sales emails.
Perhaps some coding assistance.
There’s nothing wrong with any of these ideas.
The problem is how they arrived on the list.
They were identified because there is an obvious connection between an existing AI capability and an existing business activity.
We effectively ask:
Here’s what AI can do. Where could we put it?
That’s technology looking for a problem.
And it creates a subtle bias towards opportunities that are easy to see rather than opportunities that are economically important.
The two aren’t necessarily the same.
A small use case can solve a big problem#
Consider two hypothetical AI projects.
The first helps 100 employees write emails and documents more quickly.
Suppose it saves each person ten minutes a day.
That’s a lot of time across a year.
It’s easy to understand, relatively easy to measure and probably worth investigating.
Now imagine a second opportunity.
A company receives thousands of sales enquiries each year, but its sales team can’t investigate every opportunity in depth.
Some potentially valuable prospects are consequently prioritised poorly.
Suppose better analysis could improve how those opportunities are identified and routed.
The AI might only influence one decision:
Which opportunities deserve attention first?
As an AI “use case”, that sounds less impressive than deploying a tool to an entire workforce.
But if improving that decision increases conversion on a large revenue base, its economic value could be substantially greater.
The size of the technology deployment tells us very little about the size of the business opportunity.
That distinction matters.
A small AI intervention applied to an economically important problem can be worth far more than a highly visible AI deployment applied to an inexpensive one.
Start with where the money moves#
If AI use cases aren’t the starting point, what is?
I’d start with the economics of the business.
Where is revenue created?
Where is it lost?
What determines margin?
Where does expensive human effort go?
Which constraints prevent the business from growing?
Where does poor information create bad outcomes?
Where are customers receiving a worse experience because providing a better one is too expensive?
Which errors create disproportionate downstream costs?
Where does speed materially affect an outcome?
What valuable activity isn’t performed because the economics don’t currently justify doing it?
These questions have nothing specifically to do with artificial intelligence.
That’s precisely why they’re useful.
They force us to understand the business before reaching for the technology.
Once we’ve identified where meaningful economic value exists, then AI becomes relevant.
Could better analysis improve this decision?
Could a model give an employee access to expertise they don’t currently have?
Could information that takes hours to assemble become available in seconds?
Could something currently performed for 5% of customers economically be performed for everyone?
Could an expensive bottleneck disappear?
Could we personalise an experience that currently has to be standardised?
Could we identify something valuable in information that humans don’t have the capacity to examine?
Now we’re no longer looking for somewhere to deploy AI.
We’re investigating whether AI can alter the economics of something that already matters.
The best opportunity may not involve automation#
Starting with business value also helps avoid another trap.
When companies look for AI use cases, the conversation often gravitates towards automation.
What work can AI do instead of a person?
That’s an important category of opportunity.
If a repetitive task consumes thousands of hours and AI can reliably perform part of it at substantially lower cost, the economics can be compelling.
But automation is only one way AI can create value.
Consider a salesperson preparing for an important meeting.
The objective doesn’t necessarily need to be eliminating the salesperson’s preparation time.
Perhaps AI allows them to examine more information about the customer, understand the account more deeply and enter the meeting better prepared.
The employee still does the work.
But the quality of the work changes.
Or consider someone making an operational decision.
Perhaps AI doesn’t make the decision at all.
It simply gives the person better information before they make it.
If that decision has significant economic consequences, improving it slightly may be enormously valuable.
The distinction is between asking:
Can AI remove this work?
and:
Can AI improve the economic outcome of this work?
Those questions can lead to very different opportunities.
Look for constraints, not just tasks#
There is another place I’d look.
Constraints.
Every business has things it would like to do more of but can’t.
Perhaps there aren’t enough experienced people.
Perhaps analysis takes too long.
Perhaps software-development capacity is limited.
Perhaps serving each customer individually would be too expensive.
Perhaps information exists but is spread across so many systems that nobody can practically use it.
Perhaps an expert becomes a bottleneck because hundreds of people depend on knowledge held by a handful of individuals.
Businesses adapt themselves around these constraints.
They create queues.
Prioritisation rules.
Standardised products.
Customer segments.
Approval processes.
Minimum deal sizes.
Service tiers.
All perfectly rational responses to scarcity.
But AI can potentially change some forms of scarcity.
That means an interesting AI opportunity may not appear on a process map as a task waiting to be automated.
It may appear as something the organisation has accepted as a fact about how the business has to operate.
This connects to a much bigger question.
What would we do differently if intelligence were cheaper?
Perhaps we’d research every prospect rather than the largest ones.
Analyse every customer conversation rather than a sample.
Build software for smaller internal problems.
Give more employees access to specialist knowledge.
Personalise experiences for individual customers rather than broad segments.
Perform analysis that previously wasn’t worth the cost.
Some of these possibilities may not be viable with today’s technology.
But they’re exactly the sort of opportunities that can be missed if the search begins with a catalogue of familiar AI use cases.
Not every valuable problem needs AI#
There is an important discipline required here.
Starting with valuable business problems doesn’t mean every valuable business problem should become an AI project.
Sometimes the answer is ordinary software.
Sometimes it’s better data.
Sometimes it’s redesigning a process.
Sometimes it’s training.
Sometimes the technology required already exists and the company simply hasn’t implemented it properly.
And sometimes the problem isn’t worth solving at all.
This matters because enthusiasm for a technology can distort decision-making.
If an organisation creates an “AI programme”, the programme naturally wants to find AI projects.
Success starts to mean deploying AI.
But the business doesn’t actually care whether a problem is solved using a large language model, traditional software, better process design or a spreadsheet.
It cares about the outcome.
That suggests a useful test.
If you would lose interest in solving the problem if AI wasn’t involved, it probably isn’t a sufficiently important business problem.
The technology should earn its place in the solution.
Prioritise value, not novelty#
Eventually, businesses will have more potential AI opportunities than they can realistically pursue.
That creates another challenge.
Prioritisation.
A useful starting point might be to consider two things separately.
First:
If we solved this problem, how much economic value could it create?
Then:
How capable are we of solving it with AI today?
Those are different questions.
A low-value problem with an easy technical solution may be a perfectly sensible quick win.
But it shouldn’t automatically outrank a difficult problem worth ten times as much.
Equally, an enormous economic opportunity isn’t useful if the technology is nowhere near capable enough to address it reliably.
The interesting opportunities sit somewhere between the two.
Enough economic value to matter.
Enough technical feasibility to act.
And because AI capability is changing quickly, the position of an opportunity can move.
Something valuable but technically unrealistic today may become worth revisiting in six months.
That gives businesses something more useful than a static list of AI ideas.
It gives them a portfolio of economic opportunities whose technical feasibility can be continually reassessed.
AI strategy should begin with the business#
There will be no shortage of AI use cases.
Every new model release will create more.
Software vendors will suggest them.
Consultancies will catalogue them.
Employees will discover them.
Competitors will talk about them.
The challenge for businesses won’t be generating enough ideas.
It will be distinguishing the economically important ones from everything else.
That requires reversing the usual sequence.
Don’t begin with AI and search for somewhere to deploy it.
Begin with the business.
Understand where value is created, where it disappears, which decisions have disproportionate consequences, where scarce expertise constrains performance and what valuable things aren’t currently economical to do.
Then ask whether AI changes any of those equations.
Because the objective isn’t to find the greatest number of AI use cases.
It’s to find the opportunities where better intelligence makes the greatest difference.
Stop asking where you can use AI. Start asking where AI could change the economics of your business.