The best AI use case may be the person everyone is waiting for

You’re reading The AI Directive — practical AI intelligence for
business leaders. Subscribe
here if someone forwarded this.

It might be the finance director who approves every unusual payment.
The technical specialist who is pulled into every difficult customer
question. The commercial manager who checks every non-standard deal. Or
the experienced operations person who seems to know why the process has
failed when nobody else does.

They are valuable because their judgement is scarce.

They are also the person everybody is waiting for.

When leaders look for an AI use case, the usual starting point is
routine administration. Find a repetitive task, automate it and count
the hours saved. That can be perfectly sensible.

But it may not be the most valuable place to start.

Saving ten people twenty minutes each is useful. Freeing the one
person whose queue is delaying sales, customers, projects or cash may
change the speed of the whole business.

Imagine a project cannot move until a senior engineer approves a
change. The engineer may only need ten minutes to make the judgement.
Yet the request arrives with half the information missing, so they spend
another thirty minutes finding the drawings, checking the history and
asking what has already been agreed.

The scarce resource is not only engineering judgement. It is
engineering time consumed by everything wrapped around that
judgement.

The same pattern appears elsewhere:

In each case, the organisation does not merely lose the expert's
time. Work behind the decision slows down too.

That is the bottleneck worth examining.

The strongest AI role here is often quite modest.

It can collect the relevant records, put events in order, check
whether the expected information is present, highlight what is unusual
and state the decision that is required.

It can prepare the case so the expert begins with the question, not
the search.

That does not mean every case should be decided by AI. The person is
scarce for a reason. They understand the commercial history, the
customer, the risk, the awkward exception and the consequences of
getting it wrong.

The point is to stop using that scarce judgement for work that does
not require it.

For a difficult customer case, AI might assemble the order history,
correspondence, previous promises and policy position. The manager still
decides the remedy.

For a non-standard deal, AI might show the proposed terms, margin,
precedent and unresolved risks. The commercial owner still decides
whether to proceed.

For a technical exception, AI might gather the test results, known
failure history and missing evidence. The specialist still decides what
is safe.

The division is simple: AI prepares; the expert judges.

This is not an argument against automating routine admin.

It is an argument for being more selective about which admin matters
first.

If an expert spends half the week copying information between
systems, writing the same explanation or chasing incomplete requests,
that routine work is part of the bottleneck. Removing it may be far more
valuable than automating a larger task elsewhere.

The question is not: “Where do we have the most administration?”

It is: “Which administration is consuming the resource that the rest
of the business cannot easily replace?”

That change in question leads to different priorities.

Senior leaders usually know where these constraints are, even if they
have never measured them.

They show up in familiar phrases:

Those are not complaints about a person. They are clues about the
design of the work.

Start with five questions:

That final question matters most. A good use case does not only save
time inside one task. It releases value elsewhere.

Do not judge the experiment by how impressive the AI looks.

Measure the queue.

Did the expert receive complete cases more often? Did they spend less
time reconstructing the situation? Were decisions made sooner? Did fewer
requests come back for missing information? Did customers, projects or
commercial decisions move faster afterwards?

The benefit may appear several steps away from the AI itself.

If a technical review becomes twenty minutes faster but a project
starts two days earlier, the project delay is the more important result.
If a commercial manager clears exceptions sooner and orders are released
the same day, measure the released orders. If customer escalations
arrive properly prepared and senior managers resolve more of them
without another round of questions, measure the resolution time.

The point is not to make the expert busier.

It is to stop the rest of the organisation waiting unnecessarily for
their expertise.

Find the person in your organisation who always has a queue.

Take ten recent items that reached them. Separate the time they spent
applying judgement from the time they spent finding information,
checking completeness, restating the problem or sending the work
back.

Then ask what would move faster if AI removed that surrounding
work.

That may be a better first use case than the largest pile of admin on
your list.

This is the second of three articles looking at practical patterns
for finding AI value in real work.

The first looked at the context tax: the time skilled people lose
rebuilding the situation before they can act. This one looks for the
scarce expertise that has become a bottleneck for everybody else.

Next: AI can expose a broken process. It cannot fix your
organisation.

See you Tuesday.

— Toby