Most advice on this is written for companies with a hundred staff and a systems budget. Under that size the answer is different, because your constraint isn't money, it's that nobody has a spare afternoon.
So this is ordered by what pays back fastest for the least setup.
Start here: the ones that pay back in an afternoon
Each of these is mostly turning on something you already pay for.
- Invoices and chasing them. Out automatically, reminders at seven days and fourteen, a receipt when it's paid, and a flag when a card fails. Every accounting package does this and most businesses have it switched off.
- Booking and the reminder before it. Times from a real calendar, an instant confirmation, and a message the day before. The reminder is the one that pays for itself, because a no-show costs the whole slot.
- The Monday numbers. Whatever you log into three systems to check, sent to you instead. Sales, outstanding invoices, jobs booked, review count.
- The reply you send fifty times a month. Hours, service area, parking, what to have ready. A saved reply is automation and it takes ten minutes.
Worth a few days: the ones that need something built
- Enquiries, sorted before anyone opens them. By what they're about and how urgent. This is where AI earns its place, because a subject line doesn't tell you which of those it is.
- Paperwork that arrives on paper. Work orders, delivery notes, timesheets. Photographed, read, and into whatever system you already use, with somebody checking the totals.
- Stock and reordering. Reorder points, supplier orders drafted, and a list of what's running out before you find out from a customer.
- Moving data between two systems that don't talk. The commonest job there is, and the one most likely to be already solved by a connector you can buy.
Worth doing, worth doing last
These pay back well and they need real work, so they're a poor first project.
Hiring. Applications acknowledged the same hour, sorted against what the job needs, and interviews scheduled without six emails. Worth it if you hire often. If you hire twice a year, do it by hand.
Quoting. The arithmetic and the document, from a few inputs. Not the price itself, which is judgement and should stay yours. Quoting is where the largest time savings usually sit, and it's also the one where a mistake costs the most, so it goes last on purpose.
What should stay with a person
Worth being clear about, because the temptation runs the other way once the first two work.
- A customer who's unhappy. An automated reply to a complaint makes it worse in a way that's hard to undo.
- Deciding a price. The math can be automatic. What you charge is a judgement about the customer, the job and your week.
- Anything where the exception is the point. If a job needs a decision every time, that's not a process, and forcing it into one costs you the flexibility you're being paid for.
- The first conversation with a new customer. It's the one they judge you on.
How to pick the first one
Two questions, and they take about a minute each.
Which of these happens most often in your business? Frequency beats size. A three-minute job done twenty times a week is worth more than an hour-long job done monthly.
Which one would somebody be relieved to lose? That person will help you get it right, and they'll tell you when it's wrong. An automation nobody wanted gets worked around within a month.