Which Tasks Should a Business Automate With AI First?

Here’s how to find that first task in your own business, in five steps.

Why “what should we automate?” is the wrong first question

It’s easy to start with the tool. You see a demo, buy a licence, then look for somewhere to use it.

That’s backwards. Automating a pointless task just makes a pointless task run faster. You’ve paid to keep waste alive.

The better first question is simpler: what is my team actually doing all week, and should they be doing it at all?

Answer that, and the real candidates for AI start to show. Clarity first. Technology second.

Step 1: Write down what really happened last week

Not the job description. The real week.

Ask each team lead to work with their team to list what they actually did last week, including the small tasks nobody tracks:

• Chasing a client for a signed approval
• Copying call notes into the CRM
• Rewriting the same proposal intro for the fifth time
• Answering “where’s this at?” messages

Small tasks often hide big chunks of time. So do handovers, rework and the exceptions nobody planned for. Capture those too.

Step 2: Screen each task, then check it properly

Three quick questions help you spot candidates:

1. Does it happen again and again? Frequent tasks give small time savings more chances to add up.
2. Is it mostly reading, writing or summarising? That’s where today’s AI tools are strongest.
3. Could you explain it to a new starter? If the steps can be written down, the work is easier to brief.

A yes means it’s worth a look, not that it’s ready. Before choosing a task, check three more things:

• The information is available, and safe to use with the AI tool and settings you have
• Mistakes would be easy to catch before they reach a client
• Checking the output takes less effort than doing the task yourself

That last check is easy to overlook. An AI draft that takes longer to fix than to write has saved no time.

Step 3: Remove, simplify, then automate or hand off

Before any technology goes near a task, work through it in this order:

• Remove it. Does anyone use the output? The meeting that could be a message. The report nobody opens. Stop doing it.
• Simplify it. If it’s needed, make it lighter. A 12-page report might become a one-page summary.
• Automate it. If it’s still needed and repeats, decide which kind of automation fits.
• Hand it off. If it’s needed but not the best use of this person’s time, move it. But fix it first: handing off a messy task just moves the cost to someone else.

Some tasks reveal a different problem. If your team spends hours chasing approvals, the fix may be clearer ownership, not technology.

Not every automation needs AI.

• Fixed rules? Standard automation may be enough. A booking link. A form. A simple CRM rule.
• Needs interpreting? That’s where AI helps. Turning messy call notes into a structured draft, for example.

Step 4: Decide what AI should do for each task that stays

AI can help with a task in three ways. For each task you keep, decide which one you need:

• Taking it off your plate. First drafts, data entry, meeting summaries. Your team reviews instead of starting from scratch.
• Making your team sharper. Suggesting options, spotting gaps in a proposal, pulling client history before a call. The person still decides.
• Doing what wasn’t practical before. Spotting patterns across hundreds of project records, for example. Powerful, but harder to set up and check. Save it for later, once you have the basics working.

Start with the admin your team dreads: first drafts, data entry, meeting notes. Quick to test. Easy to check. Real time back when it works.

Step 5: Run one pilot, measure it, and keep a person checking

Start with one task. Shortlist two more for later. One pilot is easier to run properly, especially when your team is already stretched.

Measure before and after. Before the pilot, record how often the task happens, how long it takes and what usually goes wrong. During the pilot, count the time spent checking and correcting the AI’s work too. Continue only if the overall result improves.

Brief the AI like a capable new hire. What the task is for. The rules to follow. What good looks like. The format you want back.

Make the human check specific. AI can produce convincing work that contains errors or misses important context. Name the person responsible and agree what they check. For a proposal, that might be scope, pricing, delivery dates and client details. Agree what happens when the output is incomplete or uncertain.

What this looks like in practice

Picture a founder-led digital agency of around 60 people. The same principles apply across many businesses. Here’s how five everyday tasks might come out:

The first move is simplifying, not buying a tool. One task needs no AI at all. And every AI draft has a named person checking it before it goes anywhere.

Frequently asked questions

What should a business automate with AI first?

Start with one frequent, low-risk task that involves reading or writing, such as drafting meeting summaries or CRM updates from call notes. Remove or simplify unnecessary work first, and check that reviewing the AI output takes less effort than doing the task yourself.

Should we cut tasks before automating them?

Yes. Remove what is unnecessary and simplify what remains. Automating a task nobody needs only makes waste run faster.

How do I know if a task is right for AI?

Use three questions to spot candidates: does it repeat, is it mostly reading or writing, and could you explain it to a new starter? Then check the information is available and safe to use, mistakes are easy to catch, and checking the output is quicker than doing the work.

What is the difference between AI and standard automation?

If a task follows fixed rules, such as sending a booking link, standard automation may be enough. AI is useful when the task involves interpreting information, such as turning varied call notes into a structured draft.

Do we still need people checking AI work?

Yes. AI can produce convincing output that contains errors or misses important context. Name the person responsible and agree exactly what they check against the source information and your standards.

How many AI use cases should we start with?

One. Shortlist two more, measure the first pilot against how the task was done before, and continue only if the overall result improves.

Selina Fischer, Founder | AI Adoption & Transformation Consultant