Digee / Wiki / Introducing AI
Guide
How to choose your first task for AI.
The most common mistake when introducing AI is starting too broadly. It works much better to pick one specific, repetitive task, build an agent for it, prove its value in live operation – and only then move on.
In short
- A good first task comes up often, has a clear input and output, and can be checked.
- Start where a mistake won't cause serious harm.
- Measure how the task is done today – otherwise you cannot prove the benefit.
On this page
5 criteria for a good first task
- It repeats. Ideally daily or weekly, tens to hundreds of cases.
- It has a clear input and output. You know what comes in (an email, invoice, request) and what should come out (a record, reply, summary).
- It can be checked. A person can tell in a moment whether the result is right.
- A mistake doesn't hurt. An error is easy to fix and has no serious consequences.
- It genuinely holds someone up. Ideally someone whose time is expensive or needed elsewhere.
Typical good first tasks
- Sorting incoming mail and enquiries by topic and urgency.
- Copying details from invoices, orders or forms into a system.
- Checking that applications and documents are complete and asking for what is missing.
- Drafting replies to recurring questions based on your own materials.
- Deadline reminders and chasing missing documents.
You will find specific tasks by industry in the agent catalogue.
What not to start with
- Tasks one person does once a month – not worth it.
- Decisions about money or people without a check – high risk, see human-in-the-loop.
- Processes that are currently changing – settle them first, then automate.
- ‘We want AI for everything’ – without a specific task, the benefit cannot be measured.
Checklist before you start
- We can describe the task in five steps.
- We know how many times a week it comes up and how long it takes.
- We know who will check the agent's work.
- We have the materials the agent should work from, and they are up to date.
- We know which systems the agent needs to access.
- We know how we will tell it has worked.
If you are unsure about any point, we will go through it with you in a free, no-obligation consultation.
Frequently asked questions
How long does it take to set up the first task?
It depends on the number of connected systems. A simple task working with email and one system can usually be tested within weeks, not months.
Do we need an AI strategy first?
No. A strategy is much easier to shape once you have one proven task behind you and know how AI works in your business.
What if the first task doesn't work out?
That is why you choose a low-risk task and measure the result. Even an unsuccessful test gives a clear answer about where AI makes no sense – and that is valuable information too.
Want to know where AI would save you time?
We go through a normal week with you and find the first task worth handing to an agent. No obligation.
Related articles
How much does an AI agent cost
What makes up the price of an AI agent: design, connecting your systems, running the model and support. How to estimate the return without guesswork.
→Explainer · 3 minHuman-in-the-loop: a person decides
Human-in-the-loop means AI prepares the work and a person approves it. How to set an agent’s limits so it saves time without making costly mistakes.
→Explainer · 4 minWhat is an AI agent
An AI agent does more than answer: it finishes a task. It reads an email, finds the details, prepares the output and hands it to a person to approve.
→