When Should You Hand a Process Over to AI? Four Questions.

When someone asks us “what could AI actually do for us?”, the first answer is almost always the same: “Let's look at what your team does the same way, every single week — without really thinking about it.” This article is for those who don't want to dive into an AI project blindly, but first want to understand which processes are genuinely worth it.

The four questions to ask about every process

There are four criteria we examine when evaluating a business process. You don't need a yes to every one — but the more yeses you have, the stronger the business case.

1. Does it repeat regularly?

AI delivers its best return where the same work needs doing daily, weekly, or monthly. Not because AI can't handle one-off tasks — but because for one-off tasks, the time spent on setup and configuration is disproportionate to the benefit.

If a process takes five minutes a week but only runs once a month, it's not the best target. If that same process runs daily and adds up to several hours a week in total, that's a different matter entirely.

2. Does it follow consistent logic?

AI handles well what operates according to defined logic: “if this condition is met, do this.” The less a given step depends on subjective human judgement, the more readily it can be automated.

A typical example: extracting incoming order confirmations from emails at a trading company and organising them into a uniform spreadsheet. No judgement calls, no exceptions — just a repeating, structured logic. That is precisely what process automation with AI is built for.

The opposite end of the spectrum: handling a customer complaint, where every case is different, the stakes of the decision are high, and the client relationship demands genuine human empathy. Fully automating that isn't wise — but automating the administrative side of it (data entry, categorisation, prioritisation) is.

3. What is the cost of an error?

Every automation project raises the same question: what happens when the AI gets it wrong? For some processes, the risk is low — if the system miscategorises an email, a person reviews it and corrects it. For others, a mistake carries serious consequences: a financial decision, a regulatory compliance issue, a client-facing communication.

In most businesses, we find that the smartest approach is to combine AI with a human review step rather than remove oversight entirely. The AI handles the mechanical work; a person approves the output. Error risk stays controlled, while the workload drops to a fraction of what it was.

When assessing error risk, it's also worth asking how the current, human-operated process performs. If a manually run process regularly produces typos or omissions, that's a risk too — we've simply grown used to it.

4. Is the data accessible?

This is where most projects get stuck — and unfortunately, it usually only becomes apparent halfway through.

AI needs data. If a process starts with a paper form, a password-protected internal file share that IT takes three weeks to grant access to, or a legacy system with no export functionality, the automation will be considerably harder than it first appeared.

Not impossible — but mapping out data accessibility at the start of a project saves a great deal of unpleasant surprises later.

Where most projects stall

In our experience, the most common obstacle isn't technical. It's that nobody has ever documented the process. Everyone knows how to do it — but in their head, and everyone does it slightly differently.

Before anything can be automated, you need to write down how the process actually works today. That exercise is valuable in its own right: it frequently reveals steps that are entirely unnecessary and shouldn't have been there in the first place.

The other typical obstacle: there is no clear internal owner for the project. IT doesn't want to get involved because “it's not their area.” Operations doesn't feel empowered to make a decision. Leadership never gets around to making it a priority. In that situation, even the best intentions come to nothing.

That's why it matters to have an internal sponsor from the outset — someone for whom this is genuinely their responsibility, and who keeps things moving.

What about smaller businesses?

We often hear: “This sounds useful, but we're too small for it.” In most cases, that's not quite right. In a ten- or twenty-person company, the situation where the accounts assistant spends several hours a week manually typing up incoming invoice data is just as real — it's simply smaller in scale.

Process automation with AI depends on the process, not on the size of the company.

A practical summary

If you want to identify which process to start with, begin with these four questions:

• Repetition: Does it run at least once a week, and does the cumulative time add up to something noticeable?
• Consistent logic: Does it have a clear, step-by-step logic that can be written down?
• Error risk: If the AI makes a mistake, can it be caught and corrected before it causes harm?
• Data accessibility: Are the inputs available digitally, in a machine-readable format?

Where you reach four yeses, that's your best starting point. Where you have two or three, it's probably achievable but will require preparation. Where you have one or none, the place to start isn't automation — it's getting the process itself in order.

None of this is rocket science. But in most businesses, nobody has ever sat down and asked these four questions together, systematically. It's worth doing.