One of the most common questions being asked in organisations today is:

“Where can we use AI?”

Teams are being encouraged to identify AI use cases. Innovation workshops are being organised. Lists of opportunities are being created. Departments are being asked how generative AI or AI agents could improve productivity.

It is an understandable question.

But it may be the wrong place to start.

Instead of asking “Where can we use AI?”, perhaps organisations should begin with a much simpler question:

“Where is work not working properly?”

Find the broken processes first. Then determine whether AI is part of the solution.

AI should not be the starting point

Starting with AI creates a subtle problem. Once we begin by looking for AI use cases, we naturally start searching for activities that fit the technology.

We see an email and ask whether AI could write it. We see a report and ask whether AI could generate it. We see an administrative task and ask whether an AI agent could perform it.

But we often skip the more important question:

Should this activity exist in its current form at all?

Perhaps the report nobody enjoys producing is not actually used to make a meaningful decision. Perhaps the email exists because two systems do not communicate. Perhaps the administrative task is compensating for a badly designed workflow. Perhaps the approval process has six steps when it should have two.

Applying AI to these activities may save time.

But it may also preserve the waste.

We have simply made an unnecessary activity more efficient.

Start with the pain

A better approach to AI transformation is to begin with operational pain rather than technological possibility.

Where are customers waiting?

Where are employees frustrated?

Where is information entered more than once?

Where do people maintain spreadsheets outside the official system?

Where are approvals taking days when the actual decision takes minutes?

Where are quality problems recurring?

Where are highly skilled employees spending significant amounts of time on low-value administration?

Where are managers making decisions without the information they need?

These are potential transformation opportunities.

Some may ultimately require AI.

Others may require automation, system integration, process redesign, clearer ownership—or simply the removal of an unnecessary step.

That distinction is important.

Imagine a broken approval process

Consider a process that requires six approvals.

It takes ten days to complete and everyone agrees it is inefficient.

An AI-first approach might ask how AI could accelerate the approvals. Perhaps AI could summarise the request, assess the documentation, recommend approval and automatically chase people who have not responded.

That could certainly make the process faster.

But a Process Excellence approach asks a different question:

Why are there six approvals?

Perhaps four were added gradually following isolated incidents. Perhaps several approvers are reviewing exactly the same information. Perhaps nobody can explain why two of the approvals still exist.

After redesign, the organisation discovers that only two approvals are genuinely required.

The best AI solution to the other four approvals is not to automate them.

It is to eliminate them.

That is the difference between AI-driven efficiency and genuine process improvement.

Broken processes are often hiding in plain sight

Every organisation contains processes that employees know do not work particularly well.

They are usually easy to recognise.

People say things such as:

“That’s just the way we’ve always done it.”

“I keep my own spreadsheet because I don’t trust the system.”

“You need to email John because otherwise it won’t move.”

“We enter it in both systems.”

“Nobody really looks at that report.”

“It takes five minutes to do the work and three days to get it approved.”

These statements are gold mines for transformation.

They reveal workarounds, duplication, waiting, poor integration, unclear ownership and unnecessary complexity.

Instead of organising an AI brainstorming workshop and asking employees for AI ideas, ask them:

“What part of your job frustrates you most?”

“What activity do you do that adds the least value?”

“Where do you spend time waiting, searching, copying, checking or chasing?”

“What process would you redesign if you could?”

The answers will probably produce a much stronger AI opportunity portfolio.

Fix, eliminate, automate or augment?

Once a broken process has been identified, the next question should not automatically be how to apply AI.

Instead, each activity should be challenged.

Can we eliminate it?

If the activity adds no value and manages no meaningful risk, remove it.

Can we simplify it?

Perhaps five steps can become two.

Can we standardise it?

If unnecessary variation is creating problems, establish a clearer way of working.

Can conventional automation handle it?

A simple rule-based workflow may be cheaper, faster and more reliable than AI.

Does it require human judgement?

Some decisions should remain human, particularly where context, accountability, ethics or significant risk are involved.

Could AI augment the person?

AI may analyse information, identify patterns, prepare a recommendation or reduce administrative workload while leaving the final decision with the employee.

Could AI perform the activity autonomously?

For suitable low-risk, well-defined activities, an AI agent may increasingly be able to complete the work.

This creates a much healthier transformation sequence:

Eliminate → Simplify → Standardise → Automate → Augment → Autonomise

AI becomes one option within process redesign rather than the predetermined answer.

Process mining can help us find the opportunities

This is also where process mining becomes particularly interesting.

Traditional improvement depends heavily on workshops, observation and employee experience. These remain important, but process mining can add another layer by using system event data to reveal how processes actually behave.

It can identify where work waits, where activities repeat, where cases move backwards, where processes deviate from the standard route and where particular variants take significantly longer than others.

Imagine combining that capability with AI.

Process mining identifies the problem.

Process improvement helps understand why it exists.

AI helps analyse alternatives and, where appropriate, becomes part of the redesigned workflow.

That is a much more powerful proposition than simply searching for AI use cases.

AI should follow process redesign

This also changes the sequence of digital transformation.

Too often the sequence looks like this:

Technology → Use Case → Process → People

We buy or discover a capability, find somewhere to use it, modify the process and then train people to work with it.

A stronger sequence is:

Problem → Process → People → Change → Data → Technology

Start with the business problem.

Understand and redesign the process.

Develop the people who operate it.

Manage the organisational change.

Make sure the necessary data exists.

Then determine what technology best supports the new way of working.

Sometimes that technology will be AI.

Sometimes it will not.

Both outcomes are perfectly acceptable.

Measure the outcome, not the amount of AI

There is another danger in asking departments to identify AI use cases: it can unintentionally make AI adoption itself the objective.

Success starts being measured by the number of pilots, licences, agents or use cases deployed.

But the business does not benefit from having more AI.

It benefits from better outcomes.

Did cycle time fall?

Did quality improve?

Did the customer experience improve?

Did employees spend less time on low-value administration?

Did decision-making improve?

Did costs fall?

Did risk decrease?

If an organisation removes an unnecessary three-hour process without using any AI, that may be a better transformation than implementing a sophisticated AI solution that saves 20 minutes.

The objective should always be performance improvement, not technology adoption.

The best AI opportunities may emerge naturally

Interestingly, when organisations follow this approach, the strongest AI opportunities tend to become much clearer.

Once the process is understood, we can see where people are spending time reading large volumes of information, comparing documents, searching for knowledge, identifying patterns, summarising cases, preparing decisions or repeatedly answering similar questions.

Those are often excellent opportunities for AI.

But now the AI is solving a clearly defined process problem.

We know what the activity is.

We know why it exists.

We know who owns it.

We understand the inputs and outputs.

We know what good performance looks like.

And we can measure whether AI actually makes it better.

That dramatically increases the likelihood of delivering real value.

AI is a tool, not the transformation

AI is undoubtedly one of the most important technologies organisations will deploy over the coming years. It will change processes, roles and operating models.

But AI itself is not the transformation.

The transformation is the redesign of how work gets done.

AI is one of the tools that can enable that redesign.

That distinction matters because organisations that start with technology often end up trying to find problems for their solutions.

Organisations that start with processes do the opposite.

They find the problems worth solving and then select the best solution.

Conclusion

Perhaps we should stop asking employees:

“Where can we use AI?”

And start asking:

“Where is work unnecessarily difficult, slow, repetitive or frustrating?”

Find the waiting.

Find the duplication.

Find the workarounds.

Find the unnecessary approvals.

Find the repeated problems.

Find the highly skilled people spending hours on low-value activities.

Then improve the process.

Eliminate what does not need to exist. Simplify what remains. Standardise where appropriate. Automate predictable activities. Use AI where intelligence genuinely adds value. Keep human judgement where it matters.

The organisations that get the most value from AI will probably not be those with the longest list of AI use cases.

They will be the ones with the clearest understanding of their processes and the discipline to solve the right problems.

Stop looking for AI use cases. Start looking for broken processes.

The AI opportunities will follow.