Once owners see what today's AI and automation tools can do, the problem changes. It is no longer "is there anything we could automate?" but "where do we start?" Every department has a list of tedious jobs, every supplier is offering something, and it is hard to know which choice will actually make a difference.
The first project matters more than it might seem. A good one saves time, builds confidence and teaches you how to do the next one. A poor one, usually something too big or too vague, can put a team off for years. Here is how we suggest choosing.
Look for work that is frequent, dull and checkable
The best first candidates tend to share three qualities.
- It happens often. Something done twenty times a week is a better start than something done twice a year, even if the yearly task is more painful. Frequent work means time saved quickly and plenty of chances to check the results.
- It follows a pattern. Reading an order and entering it, matching invoices to delivery notes, turning site notes into a report, answering the same handful of customer questions. If you could explain it to a new starter in an afternoon, it is probably a good fit.
- A person can easily check the result. Early on, someone should review what the automation produces. Choose work where a quick look tells you whether it is right.
Avoid these for a first project
Some things are tempting but are better left for later.
Anything where a mistake is costly and hard to spot. Pricing decisions, payments, legal commitments and safety matters can all benefit from automation eventually, but not as the first thing you try.
Processes nobody agrees on. If three people do the same job three different ways, automating it will just surface the disagreement. Settle how it should be done first.
Large, cross-department changes. "Automate the whole order-to-invoice process" is a programme, not a first project. Pick one step within it.
Ask the people who do the work
Owners often have a good sense of where time goes, but the people doing the work know exactly which parts are tedious, which are error-prone, and which they would happily never do again. A short conversation with them usually produces a better list than any workshop.
It also matters for adoption. People are far more likely to use a new tool that removes a job they dislike than one that was chosen for them. Involve them early and let them help test it.
A simple way to compare options
Write down three to five candidates. For each, answer four questions in a sentence or two:
- Roughly how much time does this take each week, and whose time is it?
- What goes wrong today, and what does that cost us?
- Where does the information come from, and where does it need to end up?
- If the automation made a mistake, how quickly would we notice?
The right first project is usually the one with a clear answer to all four, a reasonable amount of time saved, and a low cost of error. It is rarely the most ambitious item on the list.
Keep the first version small
Aim to have something working within weeks, not months. A first version might handle only the most common case and pass everything else to a person. That is fine. It is better to automate the straightforward, common cases reliably than to attempt every exception and finish nothing.
Agree in advance how you will know it is working: time saved, fewer errors, faster responses to customers. Then look at those after a month and decide whether to extend it, adjust it or stop.
Then do the next one
The real benefit of a good first project is what it teaches you: how your systems connect, how your team responds to change, and what "good enough" looks like in your business. The second project is usually quicker, and the third quicker still.
Where we can help
If you aren't sure which task to pick, our Executive AI Session works through real examples from your business and ends with three concrete places to start. If you already know the task, an Operational Improvement Project takes it from idea to something your team is using, working alongside the people who do the job today.
