Some of the most economically important things that happen inside a business can take only a few seconds.
A salesperson decides which opportunity to call next.
A customer-service agent decides whether a complaint needs escalating.
A manager decides how much stock to order.
Someone decides what price to quote.
A marketing team decides where to allocate its budget.
An employee decides whether something looks unusual enough to investigate.
Individually, these decisions can appear insignificant.
They may not have a budget.
They don’t appear as a line item in the accounts.
Nobody necessarily measures what they cost.
But businesses make thousands — sometimes millions — of decisions like these.
And the consequences accumulate.
A salesperson repeatedly prioritising slightly better opportunities might produce more revenue.
A customer-service team identifying unhappy customers slightly earlier might reduce churn.
Better purchasing decisions might improve working capital or reduce waste.
Better pricing decisions might improve margin.
Better fraud decisions might prevent losses.
The decision itself can be almost free.
The economic consequence of making it better can be enormous.
That makes decisions an interesting place to look for AI value.
Processes are easier to see than decisions#
When businesses look for efficiency opportunities, they naturally gravitate towards processes.
Processes are visible.
We can draw them.
Measure how long they take.
Count the people involved.
Calculate their cost.
That makes them particularly attractive when looking for AI opportunities.
If 20 employees each spend five hours a week processing documents and AI can reduce that by half, we can build a relatively straightforward business case.
Time saved multiplied by labour cost gives us a number.
Decision quality is harder.
Imagine one of those employees spends 30 seconds during the process deciding whether a particular case requires further investigation.
The labour cost of those 30 seconds is negligible.
A traditional efficiency exercise might barely notice it.
But what if the decision determines whether the business identifies an expensive problem before it occurs?
Suddenly, measuring the cost of making the decision tells us almost nothing about its economic importance.
This creates a blind spot.
We tend to notice expensive activities more easily than inexpensive activities with expensive consequences.
AI may make that distinction increasingly important.
Businesses are decision systems#
Look beneath the departments, processes and job titles of a company and you find decisions everywhere.
Who should we sell to?
What should we charge?
What should we buy?
How much should we buy?
Which customer needs attention?
Which opportunity deserves investment?
Which problem should be investigated?
Which supplier should we choose?
What should we build next?
When should we intervene?
Where should we allocate another pound of capital?
Businesses convert information into decisions and decisions into outcomes.
Some decisions happen annually in boardrooms.
Others happen thousands of times every day without anybody describing them as decisions at all.
A salesperson scanning a list of prospects and choosing the next person to contact is making one.
So is an employee deciding which email deserves immediate attention.
So is a support agent deciding which solution is most likely to resolve a customer’s problem.
So is a manager looking at several pieces of information and deciding that something doesn’t feel right.
This matters because AI is becoming increasingly capable of working with the raw material from which many decisions are made:
information.
It can retrieve it.
Summarise it.
Compare it.
Classify it.
Identify patterns within it.
Apply instructions to it.
And increasingly, help people reason across much larger quantities of it than they could practically examine themselves.
The opportunity isn’t necessarily removing the person making the decision.
It may simply be changing what they know when they make it.
A small improvement can have a large value#
Consider a hypothetical business that makes 100,000 economically meaningful decisions each year.
Suppose the average difference between a good outcome and a poor one is £100.
That means those decisions collectively sit across an economic surface worth as much as £10 million.
Now imagine AI doesn’t automate a single one of them.
It simply provides additional information that allows employees to improve the outcome in a small proportion of cases.
Even a modest improvement could become economically significant when repeated across a sufficiently large number of decisions.
The numbers here are illustrative, but the mechanism is what matters:
Frequency × consequence × improvement.
A decision doesn’t have to be individually enormous.
It can become economically important because it happens repeatedly.
Equally, a decision doesn’t have to happen frequently if its consequences are sufficiently large.
A pricing decision affecting a major contract may happen once.
A product-development decision might influence years of investment.
A purchasing decision could commit significant capital.
A decision about whether to intervene with a valuable customer could affect a relationship worth hundreds of thousands of pounds.
This gives us two different forms of decision leverage:
small decisions made at enormous scale, and infrequent decisions with enormous consequences.
Both deserve attention.
The cost of making a decision and the value of making a better decision can be radically different.
Better information can be more valuable than automation#
Much of the AI conversation is understandably focused on what machines can do autonomously.
But autonomy isn’t required for decision value.
Imagine a salesperson deciding which of 50 opportunities deserves attention today.
AI could theoretically make that decision for them.
But it doesn’t have to.
Perhaps the system examines previous interactions, account characteristics, recent activity and other relevant information, then presents the salesperson with five opportunities and explains why each may deserve attention.
The human still decides.
What changed is the information available at the moment of decision.
The same architecture could apply elsewhere.
An AI system could flag customers showing unusual patterns without deciding what intervention should follow.
It could identify inconsistencies in a contract without determining whether the company should sign it.
It could surface anomalies in operational data without deciding whether a process should stop.
It could present relevant historical information before a manager makes an important judgement.
It could rank possibilities without choosing between them.
This distinction matters because businesses shouldn’t measure AI progress by how many decisions they can remove humans from.
Sometimes automation will be appropriate.
Sometimes it won’t.
The more useful question is:
Did the technology improve the economics of the decision?
That improvement might come from greater accuracy.
Or speed.
Or consistency.
Or access to more information.
Or recognising something a person might otherwise have missed.
AI can create decision value without becoming the decision-maker.
Look for decisions with leverage#
If I were examining a business for AI opportunities, I wouldn’t only map its processes.
I’d start mapping its decisions.
Not every decision.
The ones with leverage.
I’d look for decisions that happen frequently.
Decisions where mistakes are expensive.
Decisions influencing significant revenue or cost.
Decisions constrained by how much information a person can realistically examine.
Decisions where experienced people consistently outperform inexperienced ones.
Decisions made under time pressure.
Decisions where relevant information exists but is difficult to retrieve.
Decisions where inconsistency creates problems.
Decisions that determine what happens next in an economically important process.
Then I’d ask why the decision is difficult.
Perhaps the information is fragmented.
Perhaps there is simply too much of it.
Perhaps interpreting it requires expertise.
Perhaps people don’t have enough time.
Perhaps the feedback loop is poor, so nobody learns which decisions were actually good.
Perhaps the decision is already easy and AI would add little.
That’s useful to know too.
The objective isn’t to insert AI into every judgement.
It’s to find where better intelligence changes an outcome worth changing.
Expertise may be hiding inside decisions#
There is another reason decisions are interesting.
They often contain expertise that organisations don’t explicitly recognise.
Ask an experienced salesperson why they pursued one opportunity rather than another and they may struggle to give you a precise formula.
They noticed something.
A combination of signals accumulated through years of experience.
An experienced operations manager might recognise that a situation requires intervention before a newer employee does.
A customer-service employee may know from a handful of clues that a seemingly ordinary complaint could become serious.
A buyer may notice characteristics that suggest a supplier is likely to cause problems.
Businesses contain thousands of these small acts of judgement.
Some will remain deeply human.
But AI creates the possibility of supporting more people with more of the information and context that previously sat disproportionately with experienced individuals.
That doesn’t mean capturing someone’s expertise is easy.
Often it isn’t.
And a model trained on general information doesn’t automatically understand the peculiarities of a particular business.
But it does suggest an interesting question:
Where does the quality of an outcome currently depend on who happens to be making the decision?
If the gap between the best decision-makers and everyone else is economically meaningful, helping more people make decisions closer to the quality of the best could create substantial value.
That is a very different opportunity from replacing those people.
Decision value is harder to measure#
There is an obvious problem with all this.
Measuring decision improvement is considerably harder than measuring time saved.
If AI reduces a task from an hour to 30 minutes, the productivity gain is visible.
If AI helps someone make a better decision, proving causality can be difficult.
Would they have made the same decision anyway?
What would have happened without the intervention?
Was the eventual outcome caused by the decision or by something else?
How do we define a “good” decision when outcomes sometimes depend on luck?
These aren’t trivial questions.
And they create a danger.
Businesses may naturally prioritise AI projects where the value is easiest to measure rather than where the value is greatest.
That’s understandable.
But measurement difficulty doesn’t make an economic effect disappear.
It means the organisation needs to think more carefully about how it evaluates it.
Sometimes that might involve controlled experiments.
Sometimes historical comparisons.
Sometimes measuring intermediate outcomes.
Sometimes explicitly tracking recommendations and subsequent results.
And sometimes the uncertainty will simply be too great to justify investment.
The important thing is recognising that easy-to-measure value and large value aren’t necessarily the same thing.
AI should improve the system, not just the moment#
There’s an even more interesting possibility.
Once decisions become visible, businesses can start learning from them.
Imagine recording:
What information was available?
What did the AI recommend?
What did the person decide?
What happened afterwards?
Now the organisation has something it often lacks:
a feedback loop.
Over time, patterns may emerge.
Certain signals might prove more predictive than expected.
Some recommendations may consistently perform poorly.
Experienced employees may override the system in particular circumstances and repeatedly be right.
Perhaps the AI needs improving.
Perhaps the business process does.
Perhaps the organisation discovers that a decision it considered important barely affects the outcome at all.
This is where AI can become more than a tool attached to an individual task.
It can become part of a system through which the business gets better at understanding how its decisions create results.
That could ultimately be more valuable than saving a few minutes making them.
The invisible opportunity#
Businesses will find plenty of obvious AI opportunities.
Documents to summarise.
Content to generate.
Processes to automate.
Questions to answer.
Those opportunities are real, and many will be valuable.
But the most visible opportunities won’t necessarily be the largest.
Somewhere inside almost every business are thousands of moments where people convert incomplete information into judgement.
Most are too small to appear on a board agenda.
Many have never had a business case attached to them.
Some happen so routinely that nobody notices they’re decisions at all.
Yet collectively, they determine which customers buy, which customers leave, where money gets spent, which problems receive attention and how resources are allocated.
AI doesn’t need to take those decisions away from people to matter.
If it can make economically important decisions slightly better, slightly faster or slightly more consistent, the effect can accumulate.
So when looking for AI value, don’t just ask which processes cost the most.
Look inside them.
Some of the most valuable AI opportunities in your business may be hiding inside decisions so small that nobody has ever thought to put a price on them.