You’re reading The AI Directive — practical AI intelligence for
business leaders. Subscribe
here if someone forwarded this.
THE SIGNAL
THE AI WORKED. THE
BUSINESS STILL DIDN'T.
An AI pilot takes a customer request, pulls together the relevant
history and prepares a clear answer in minutes.
Then the work stops.
Sales believes Operations owns the next decision. Operations thinks
Technical must approve it. Technical is working from a different product
record, and nobody is certain who can accept the commercial risk.
The AI did what it was asked to do. It made one part of the work
faster.
It also exposed the fact that the wider process was never clear.
This is an awkward but useful moment. It is easy to conclude that the
AI has failed, or that it needs a better prompt, more data or a cleverer
agent. Sometimes the problem is simpler: the organisation has not agreed
how the work should flow from beginning to end.
AI can reveal that problem very quickly. It cannot make the missing
decisions on the organisation's behalf.
A FASTER TASK
DOES NOT CREATE A BETTER PROCESS
Leaders often introduce AI at the point where the pain is most
visible.
A team is overwhelmed, so AI is used to prepare the work. A queue is
slow, so AI is used to classify it. A report takes too long, so AI is
used to produce it faster.
That can work. The previous two articles looked at exactly those
kinds of opportunities: removing the context tax and freeing scarce
experts from the work surrounding their judgement.
But a business is not one task.
If the output still enters an unclear approval route, relies on
disputed data or reaches a team with no agreed responsibility, the
overall process may remain just as slow. The bottleneck simply
moves.
You can make the first three steps twice as fast and still gain
nothing if step four has no owner.
AI WILL
FIND THE GAPS YOU HAVE LEARNED TO WORK AROUND
Most organisations carry a surprising amount of ambiguity without
calling it ambiguity.
Experienced people know which spreadsheet is really current. They
know who to call when the process says one thing but the business does
another. They understand that “approved” means something different in
Sales, Finance and Operations. They remember which customer promise
matters even though it was never written into the system.
The work continues because people compensate for the gaps.
AI is less forgiving. When it tries to connect the work, it will
encounter:
two systems with different answers;
a process map that does not match what people actually do;
decisions with no clear owner;
the same data described differently by different teams; and
exceptions that depend on somebody's memory rather than an agreed
rule.
Those are not primarily AI problems. They are process, data and
ownership problems that AI has made harder to ignore.
DO NOT ASK AI TO
INVENT THE MISSING BUSINESS
When the organisation has not defined the process, AI will still try
to be helpful.
It may choose the most plausible data source. It may infer who owns
the decision. It may smooth over a contradiction and produce an answer
that looks complete.
That is exactly what you do not want.
A confident answer cannot replace an agreed decision right. A
polished workflow cannot decide which team carries the risk. A generated
process map cannot settle a commercial argument about how the business
should operate.
If ownership is unclear, leaders have to clarify it. If two systems
disagree, somebody has to decide which source is authoritative and fix
the underlying data. If teams use different definitions, they have to
agree the meaning.
AI can help with that work. It cannot make the organisational choices
legitimate simply by making them quickly.
USE AI AS A MIRROR
This does not make AI irrelevant. Quite the opposite.
AI can compare the documented process with the way work actually
moves. It can pull together examples from emails, tickets, meeting notes
and systems to show where the same confusion keeps returning. It can
identify where information is repeatedly re-entered, where decisions
wait and where teams use conflicting definitions.
It can help draft an end-to-end process, assemble a data glossary,
document decision rights and show where responsibility is missing or
duplicated.
That is valuable work—but it is diagnostic and preparatory.
The organisation still has to decide:
who owns the outcome from beginning to end;
which decisions sit with which roles;
which data is authoritative;
what happens when the normal path fails; and
which trade-offs the business is willing to accept.
That is the human and AI partnership at its most useful. AI makes the
inconsistency visible and helps organise the evidence. Leaders make the
business coherent.
MEASURE THE WHOLE FLOW
An AI use case should not be judged only by the task it improves.
If a report is produced in five minutes instead of an hour, what
happened next? If cases are prepared faster, did the queue actually
shorten? If customer requests are classified instantly, were they
resolved sooner?
Measure from the start of the work to the business outcome, not
merely the speed of the AI-enabled step.
That will tell you whether AI removed a constraint, moved it
somewhere else or exposed a deeper weakness that now needs leadership
attention.
All three results are useful. The mistake is calling the third one an
AI failure.
THE REAL OPPORTUNITY
The best outcome from an AI pilot may not be the automation you
expected.
It may be discovering that nobody owns the complete customer journey.
That the product data cannot support the decisions being asked of it.
That approval rights grew by accident. Or that the official process
survives only because experienced people quietly repair it every
day.
AI has not created those problems.
It has shown you where the organisation is getting in its own
way.
Use that evidence. Improve the process, clarify the ownership and fix
the data. Then AI has a much better chance of improving the whole flow
rather than making one isolated task faster.
ONE THING
Take one AI use case you are considering and trace the work from the
original request to the final business outcome.
Ask three questions:
Where does it still wait after the AI has done its part?
Which data or decision is disputed?
Who owns the outcome from beginning to end?
If those answers are unclear, do not ask AI to guess. Treat the gaps
as the real improvement work.
FROM THE EDITOR
This is the final article in a three-part series about finding
practical AI value in real work.
The first looked at the context people rebuild before they can act.
The second looked at scarce expertise trapped in preparation and
administration. This one is the limit: AI can improve those parts of the
work, but it cannot compensate for an organisation that has not agreed
its process, data and decision rights.
The best AI use cases are not always the largest tasks. They are
often the points where useful work is waiting—and where making that wait
visible tells leaders something important about the business itself.
In the next few articles, I want to keep that same practical test:
where AI genuinely helps, where it exposes deeper work in processes and
data, and where the honest answer is that doing the job yourself is
still faster.
See you Tuesday.
— Toby