What AI Should Do in Your Business (And What It Shouldn't)

There is a line in almost every business where automation stops being useful and starts being a liability. Not because the technology fails, but because the decision on the other side of that line requires something a machine cannot provide.

Knowing where that line sits is the most important thing to understand before you bring AI into your business.

The tasks AI is genuinely good at

Start with the work nobody enjoys. The repetitive, low-stakes tasks that have to happen but contribute nothing to what makes your business worth choosing.

Chasing unpaid invoices. Logging enquiry details into a CRM. Sending appointment reminders. Categorising expenses. Pulling together a weekly summary from scattered data. Routing incoming emails to the right inbox.

None of these require judgement. They require accuracy, consistency, and the ability to run at two in the morning without anyone watching. That is what automation does well.

A solicitor we spoke to recently was spending roughly four hours every Friday afternoon moving information from one system to another. The work had to be done. It was accurate. But it was four hours of a qualified solicitor’s time spent on data entry. That is not what they went to law school for, and it is not what their clients are paying for.

Automating that task does not replace the solicitor. It gives them four hours back. Four hours they can spend with clients, on case work, or simply finishing the week at a reasonable hour.

That is what good automation looks like. It removes the friction so that the person doing the meaningful work can actually do it.

The tasks AI should not be doing

This is where some businesses make the wrong turn, usually without realising it.

A financial planner who uses AI to summarise meeting notes, generate draft reports, or flag gaps in a client’s portfolio review is using it sensibly. An AI system that decides whether a client should move their pension into higher-risk funds is a different matter entirely.

The difference is not just regulatory, though regulation is part of it. The issue is the nature of the decision itself.

Financial planning is not a calculation. The maths is the easy part. The hard part is understanding that the client who says “I want higher returns” also has a daughter starting university next year, a marriage that is under pressure, and a genuine risk tolerance that is quite different from what they say when asked directly. A good financial planner picks up on all of that. They read the room. They adjust. They ask the question that changes the conversation.

No AI model, regardless of how well it has been trained, can sit across a table from someone and notice that they are more anxious than they are letting on.

Why healthcare is the same argument

The same principle applies in clinical settings. There are real, practical uses for AI in healthcare: flagging anomalies in scan results that a radiologist then reviews, cross-referencing drug interactions at the prescribing stage, summarising a patient’s history so a GP has what they need in the first thirty seconds of a consultation.

These uses reduce the chance of something being missed. They support clinical decision-making rather than replacing it.

The conversation with a patient about their diagnosis is different. The judgement call about treatment options, the decision to refer to a specialist, the assessment of whether a patient is ready to go home: these stay with the clinician. Not because AI cannot process the relevant information, but because the person on the other side of the desk deserves a human being who is accountable, who can explain their reasoning in plain language, and who understands that they are not looking at a dataset. They are talking to a person who is frightened.

Accountability matters here too. If an AI-assisted recommendation causes harm, who is responsible? In clinical practice, as in financial services and legal work, the answer is that a named professional remains accountable for every decision made about the people in their care. You cannot outsource accountability. Automation does not change that.

Education: the paperwork can change, the decisions cannot

It comes up in schools as well. AI can help a teacher save time on lesson planning, identify early where a student is falling behind based on assessment patterns, and produce a first draft of a progress report that the teacher then reviews, rewrites, and signs off.

What it cannot do is replace the judgement of a teacher who knows that a particular student’s recent drop in performance has nothing to do with their ability, and that the right response is not an intervention programme. It is a quiet conversation.

A SENCO working with a student with complex additional needs is making decisions that shape that child’s educational path, their confidence, and in some cases their life beyond school. The paperwork that surrounds that work is a legitimate place for automation. The decisions themselves are not.

This is not an argument against using AI in education. It is an argument for being clear about what it is for. Teachers should be spending less time on administrative work and more time with students. AI can help with the first part. It cannot help with the second, because the second part is the job.

What this means for professional services businesses

If your business handles sensitive client information, whether that is financial data, medical records, legal case files, personnel information, or anything where the stakes are high for the person whose data it is, there are two questions worth sitting with.

First: which parts of your work are genuinely repetitive and low-stakes? That is your automation opportunity. It is almost certainly larger than you think. Most professional services businesses are carrying a significant admin burden that could be handled by well-configured software.

Second: which parts of your work require professional judgement, personal knowledge of a client, or accountability that sits with a named individual? Those stay with your people. Full stop.

The businesses that get this right end up doing a better version of what they already do. Less time on routine tasks. More time on the work that justifies their fees. Clients who still feel like they are dealing with a human being, because they are.

A word on where your data goes

If your business handles the kind of information described above, there is a further question worth asking: when you use an AI tool, where does the data go?

Most AI services work by sending your inputs to a cloud server run by a third party. For general business admin, that is usually acceptable. For a legal firm summarising client case files, or a financial advisory business using AI to review client data, it raises questions that are harder to dismiss.

There are practical alternatives. Local AI models can run on hardware inside your own office, on a network you control. Nothing is transmitted to an external server. No data policy to read and hope for the best. No question about where a client’s financial history has ended up.

This is not a niche concern for large enterprises. For any professional services business with data protection obligations, it is a real option that is worth understanding. The hardware cost is a one-off. The models run on-site. And the answer to “where does our client data go?” becomes straightforward.

The short version

AI is a tool. Like any tool, its value depends entirely on what you use it for.

Use it for the tasks that nobody needs a human for: the chasing, the logging, the reminders, the formatting, the first-pass summaries that still get reviewed before they go anywhere. Clear this work out of the way and your people can focus on what actually requires them.

Do not use it to make decisions that belong to people. Where a client’s financial future is at stake. Where a patient’s care is involved. Where a child’s education is being shaped. Where professional accountability sits with a named individual who can be called to explain themselves.

The businesses that get this right are not the most technically sophisticated ones. They are the ones that have thought clearly about what their business actually is, what their clients are paying for, and what it would mean to lose the thing that makes them worth choosing.

Your clients came to you because of you. Good automation protects that. It does not replace it.

There is always a hand on the tiller. The technology changes what you can do. It does not change who is responsible for where you end up.