- Good first tasks are frequent, follow clear rules, carry little risk, rely on records you already keep, and annoy the people who do them.
- Score each candidate task out of five. Start with the highest scores.
- Start inside the company, then suppliers, and customers last.
- Do not automate a process that is not clear yet. Make it clear first.
Once a company decides to hand some work to AI, the next question is always the same: where do we start?
The choice matters more than the tool. A good first task shows results within weeks and wins your team's support. A poor one, too risky, too rare or too vague, can take months and leave everyone doubting the idea.
Five questions for every task
Make a list of tasks your team does every week. Then ask five questions about each one.
- Is it frequent? A task done every day or every week repays the effort of setting it up. A task done twice a year usually does not.
- Does it follow clear rules? "When an invoice arrives, save it in the supplier's folder with the date and number in the name" is a rule. "Decide whether this customer deserves a discount" is not.
- Is the risk low if something goes wrong? A file saved in the wrong folder is easy to spot and fix. A wrong email to a customer is not.
- Does it rely on records you already keep? If the information is in your email, files or systems, the task can use it. If it lives only in someone's head, write it down first.
- Does it annoy the people who do it? The tasks people dislike most are often the ones that slip. Taking them away is also the quickest way to win your team's support.
Give one point for each "yes". Tasks that score four or five are good first candidates.
An example
| Task | Frequent | Clear rules | Low risk | In records | Annoying | Score |
|---|---|---|---|---|---|---|
| Filing invoices that arrive by email | ✓ | ✓ | ✓ | ✓ | ✓ | 5 |
| Chasing missing timesheets | ✓ | ✓ | ✓ | ✓ | ✓ | 5 |
| Preparing a new starter pack | ✓ | ✓ | ✓ | ✓ | 4 | |
| Requesting expired supplier certificates | ✓ | ✓ | ✓ | ✓ | ✓ | 5 |
| Drafting a quote for a regular customer | ✓ | ✓ | ✓ | 3 | ||
| Deciding which supplier to use | ✓ | 1 |
The first four are good places to start. The quote is worth doing later, with a person checking every draft. The supplier decision is not a task to automate at all: it needs judgement.
Follow an order
Even among high-scoring tasks, the order matters.
- 1Inside the company
Filing, reminders to colleagues, starter packs. Mistakes are seen only by your own team, and are easy to correct.
- 2Suppliers
Routine requests and reminders, such as an updated certificate or a delivery date.
- 3Customers
Last, and with a person checking anything a customer will read before it is sent.
Each step builds confidence for the next, both in the tool and in your team.
What to leave alone for now
- Tasks that need judgement. Pricing decisions, hiring, handling a complaint from an upset customer.
- Tasks that happen rarely. The set-up effort is better spent on work that repeats.
- Tasks where a mistake cannot be undone. Payments, contract signatures, anything sent to a regulator. Let AI prepare them; let a person decide.
In short
Choose first tasks that are frequent, rule-based, low-risk, rely on your records, and annoy the people who do them. Start inside the company, move to suppliers, and leave customers until last. Make a process clear before you automate it, and leave judgement to people.