AI · · 3 min read
Which tasks should you automate first? A simple priority guide
Not everything should be automated, and the order decides whether AI pays off. Here is a simple way to rank your tasks — and find the one to start with.
By Mediseo
The question is rarely whether AI can automate a task — it is which task you should take on first. Choose right and you see value in weeks. Choose wrong and you burn time and trust on something that could never pay off.
Two simple axes
You do not need a spreadsheet to prioritise. Judge each task along two axes:
- How often does it happen? Daily beats monthly. Volume is what makes automation worth the bother.
- How much judgement does it need? Does the task follow a pattern, or does it require discretion and accountability?
The best candidates are high volume and low judgement. The worst are low volume and high judgement — there the solution costs more than the time it saves, and the risk is greatest.
Start bottom-right, not top-left
| Begin here (high ROI) | Wait or skip (low ROI) |
|---|---|
| Drafts for emails that resemble each other | Complaints and difficult customer relations |
| Meeting notes and summaries | Legal and financial judgements |
| Sorting and categorising | Creative work where the distinctiveness is the point |
| First drafts of quotes and reports | Rare one-off tasks |
| Finding your way around internal documents | Anything that carries accountability for the outcome |
If you recognise the left column in your own day, that is where you begin.
The four that pay off earliest
In most small businesses, the same four tasks give the most back first:
Email. Everyone has too much of it. AI can sort incoming mail by urgency and draft replies to enquiries that resemble each other. You read, adjust and send.
Meeting notes. AI transcribes the meeting and produces a structured summary: what was decided, who is doing what, and by when. Remember to tell participants the meeting is being recorded.
Quotes and reports. If they follow a template — and most do — AI can fill in a first draft from notes and earlier documents. A human checks the figures and terms before anything goes out.
Document handling. "Find last year's insurance agreement" is a question an internal assistant can answer in seconds, provided the documents are accessible to it.
Test before you bet
Even a good candidate should prove itself before you trust it blindly. The safest way:
- Build thin. Get the solution to handle the most common 80 per cent, not every exception at once.
- Keep a human in the loop. AI proposes, a human approves — until the solution has shown it is reliable.
- Measure one thing. How much time is saved per week? That number decides whether you expand.
This order lets you see value quickly while keeping risk low. Each expansion rests on something that already works.
When the answer is "none of them yet"
Sometimes the honest conclusion is that no task is ready for automation yet. That is perfectly fine.
- If the volume is low — three enquiries a day — the solution costs more than it saves.
- If the data lives only in one employee's head, the knowledge has to be written down first.
- If a simple rule or template solves it, that is cheaper than AI, and it never gets things wrong.
Waiting for the right task is not falling behind. It is declining to pay for a solution that could not pay off.
The simple test
When you are stuck, ask one question: do we do this task often, does it follow a pattern, and is it fine if it has to be checked by a human before it is finished? Three yeses mean you have found a good place to begin.
If you would like to work out what the repetitive tasks actually cost you in a year, you can use the time-wasters calculator — or book a quick call about which task is worth starting with in your business.