How can we tell whether a process is ready to be automated?

Walk through it step by step and ask why each one exists. If the answer is some version of "because it always has", that step is suspect. Count the handoffs: more than two or three usually means the process is more tangled than it needs to be. Look at where errors cluster, because those are usually design faults rather than execution faults. And watch how the team copes with exceptions. If exceptions take more time than normal work, redesign before you automate.

What does the fix-first approach look like in practice?

Four steps in order. Map what actually happens, not what the process document claims, by talking to the people doing the work. Remove the steps that exist for outdated reasons or no reason at all. Simplify what remains, combining steps and shortening approval chains where you can. Then automate. The property management example moved from 22 steps and 12 working days to 10 steps and 5 days before any AI got involved, then 2 days and 20 minutes of human time per application after.

What goes wrong when businesses skip straight to automating an existing process?

You buy a very efficient version of something that should not exist in its current shape. The post mentions a business that spent £20,000 automating a complex approval workflow, only to realise six months later that half the steps came from an outdated policy, three of the approvals were rubber stamps, and two spreadsheets tracked the same data. The National Audit Office has found this is the most common cause of failure in government IT projects, and the same pattern shows up in private sector AI work.

Is process simplification on its own worth doing, even if AI never arrives?

Often yes. Process redesign before automation typically delivers 30 to 50 percent of the gains people expected from AI on its own, and CIMA research has shown it tends to double the return on investment when AI is added later. Sometimes the most useful advice we give a client is do not use AI yet. Fix the process first. It is not glamorous and it does not sell software, but it saves real money and a lot of frustration.

AI won't fix a broken process

I need to say something that a lot of AI consultancies will not tell you. AI is not magic. If your process is broken, automating it will just give you broken results faster. And you will have paid good money for the privilege.

This is the single most common reason AI projects fail. Not the technology. Not the data. Not the team. It is trying to automate a process that should not exist in its current form.

The "paving the cow paths" problem

There is an old saying in urban planning: the worst thing you can do is pave the cow paths. Cows wander. They take inefficient routes based on habit, not logic. If you pave those routes, you get permanent, expensive roads that go nowhere useful.

The same thing happens with AI. If your invoice approval process involves seven steps, three people, and two spreadsheets because "that's how we've always done it," then automating that process will give you a very efficient version of something that should be three steps, one person, and no spreadsheets.

I have seen this play out repeatedly. A business spends £20,000 automating a complex approval workflow, only to realise six months later that the workflow itself was the problem. Half the steps existed because of a policy that was no longer relevant. Three of the approvals were rubber stamps. The two spreadsheets were tracking the same data.

The National Audit Office's report on digital transformation found that the most common cause of failure in government IT projects was automating existing processes without first questioning whether those processes made sense. The same applies to private sector AI projects.

How to tell if your process is broken

Before you automate anything, ask these questions.

Why does each step exist?

Go through your process step by step and ask why. Not "what happens here?" but "why does this step exist?" If the answer is "because it's always been that way" or "I'm not sure," that is a red flag.

Healthy processes have steps that exist for clear, current reasons: legal compliance, quality assurance, genuine decision points. Broken processes have steps that exist because of historical accidents, departed employees' preferences, or problems that were solved years ago.

How many handoffs are there?

Every time work passes from one person to another, there is a delay, a risk of miscommunication, and a chance of something falling through the cracks. If your process involves more than two or three handoffs, it is probably more complicated than it needs to be.

Where do errors happen?

Errors cluster around specific points in a process. If the same step keeps causing problems, the step itself may be poorly designed. Automating it will not fix the design. It will just make the errors happen more consistently.

What happens when something goes wrong?

In a well-designed process, exceptions are handled smoothly. In a broken one, any deviation from the norm causes chaos. If your team spends more time dealing with exceptions than processing normal work, the process needs redesigning before it needs automating.

The fix-first approach

Here is what we recommend to every client before we talk about AI.

Step 1: Map the process honestly

Get the people who actually do the work to describe what really happens. Not what the process document says (if one even exists), but what actually happens day to day. These are often very different things.

We recently worked with a logistics company where the documented process for handling customer complaints had 5 steps. The actual process, as described by the team, had 14 steps including 3 workarounds that had been added over the years to deal with recurring problems. Nobody had updated the documentation. The British Standards Institution's guidance on process management recommends regular process audits for exactly this reason.

Step 2: Remove what is unnecessary

Look at every step and ask: what would happen if we stopped doing this? Some steps will be essential. Others will turn out to be leftovers from a different era. Remove them.

This sounds obvious, but it requires courage. People are attached to processes. "We've always done a second check on every order" might sound prudent, but if the second check catches an error once in every 500 orders, the cost of checking 500 orders to find one mistake is almost certainly higher than the cost of fixing the occasional error after the fact.

Step 3: Simplify what remains

Once you have removed the unnecessary steps, simplify what is left. Can two steps be combined? Can a three-person approval chain become a one-person approval with spot checks? Can information be entered once instead of twice?

Step 4: Then automate

Now you have a clean, simple process that makes sense. This is what you automate. The AI project will be cheaper because there is less to build. It will be faster because there are fewer edge cases. And it will deliver better results because the underlying logic is sound.

A real example

A property management company came to us wanting to automate their tenant onboarding process. When we mapped it, we found 22 separate steps involving 4 different people and 3 different systems. The process took an average of 12 working days from application to move-in.

Before touching any technology, we worked with the team to simplify the process. We removed 8 steps that were either redundant or based on outdated policies. We combined 4 others. We reduced the approval chain from 3 people to 1 (with the other 2 doing spot checks on a sample basis instead of reviewing every application).

The simplified process had 10 steps, involved 2 people, and took 5 working days. That was already a massive improvement, and we had not used any AI yet.

Then we automated it. The AI handled document collection, reference checking, and data entry. The simplified, automated process now takes 2 working days and requires about 20 minutes of human time per application.

If we had automated the original 22-step process, it would have taken longer to build, cost more, and delivered a worse result. You cannot automate your way out of a bad process.

That same "simplify before you automate" pattern shows up in every estate agency and lettings firm we work with. If you run one, the sector-specific guides for Manchester, Leeds, Newcastle, Edinburgh and Glasgow walk through the common trouble spots.

Signs your AI project is heading for trouble

Watch for the scope creeping steadily larger as people add "while we're at it" requests. That usually signals the underlying process is not well understood. If edge cases outnumber normal ones, the design is the problem, not the volume. And if nobody can explain why a particular step exists, it probably does not need doing. A related tell: steps that exist to fix a problem caused by another step. You do not need to automate that workaround. You need to fix the root cause.

The McKinsey Global Institute's research on automation consistently finds that the highest-ROI automation projects are those that simplify processes before automating them.

The uncomfortable truth

Sometimes the best AI advice is: do not use AI yet. Fix your processes first. It is not glamorous advice, and it does not sell software licences. But it is honest, and it saves businesses a lot of money and frustration.

The good news is that process simplification on its own often delivers 30 to 50% of the gains that people expected from AI. And when you do add AI on top of clean processes, the results are dramatically better.

The Chartered Institute of Management Accountants has published research showing that process redesign before automation typically doubles the return on investment compared to automating existing processes as they are.

Not sure where to start?

If you are thinking about AI but not sure whether your processes are ready, our free AI opportunity report will give you an honest assessment. We will tell you where AI makes sense right now, where you need to do some groundwork first, and where you might not need AI at all. No sales pitch, just practical advice.

Get your free AI opportunity report here