The five jobs worth handing to AI first
Most business owners approach AI backwards. They ask "what can this thing do?" (which produces a list of impressive-sounding possibilities, none of which are urgent) instead of "what is quietly costing me money every week?"
Start with the second question. Here are the five answers that come up most often, roughly in order of how quickly they pay for themselves.
1. The calls you don't answer
This is almost always the biggest one, and almost always the most invisible. Nobody leaves a voicemail saying "I was going to spend ¥80,000 with you but you didn't pick up." They just call the next business on the list.
Count it for one week. Every missed call, every voicemail you returned the next day, every "sorry, we were slammed." If you're a service business, a meaningful share of those were people ready to buy.
An AI receptionist answers on the first ring, at 9pm, on Sunday. It asks what they need, when they need it, and how to reach them, then sends you the details. It doesn't need to be clever. It needs to exist.
Worth doing when: you miss more than a handful of calls a week and your work comes from phone enquiries.
2. The same ten questions
Every business has them. What are your hours, do you take walk-ins, do you service my area, how much is X roughly, do you do Y.
If you're answering these by hand (in email, in DMs, on the phone) you're doing unpaid work. Worse, you're doing it slowly, which means people wait, and some of them stop waiting.
The fix isn't a chatbot that pretends to be a person. It's a small, well-fed assistant that knows your actual answers and gives them immediately, then hands anything unusual to you. The trick is writing down the ten answers properly once. That part takes an afternoon and nobody wants to do it, which is exactly why it's valuable.
Worth doing when: you find yourself typing the same reply for the third time this week.
3. Following up
Almost nobody follows up. Someone enquires, you send a quote, they go quiet, and that's the end of it. Not because they said no, but because life got in the way and neither of you circled back.
A follow-up sequence is unglamorous and it works. Two days later: "just checking you got this." A week later: "still happy to answer any questions." That's it. No pressure, no cleverness.
This one barely needs AI at all. It needs a system that remembers. AI helps by drafting messages that sound like you rather than like a template, and by noticing when a reply needs a human.
Worth doing when: you send quotes or proposals and don't have a reliable way of chasing them.
4. Turning mess into records
Photos of receipts. Handwritten job notes. A voice memo you left yourself in the van. Emails with the details buried three replies deep.
All of that is information you already have and can't easily use. AI is genuinely good at reading messy input and producing structured output: turning a photo of a receipt into a line in your accounts, or a rambling voice note into a job record with a date, a client and a price.
This won't win you customers. It will give you back an evening a week, which for a lot of owners is the more valuable thing.
Worth doing when: admin is eating your evenings or weekends.
5. Writing the things you avoid writing
Quotes. Service descriptions. The About page you've been meaning to redo since 2023. The email to the client you're dreading.
AI is a good first-draft machine and a poor final-draft machine. Used properly, it removes the blank page problem. The reason those tasks sit undone for months isn't that they're hard, it's that starting is hard.
Worth doing when: there's a piece of writing you've been putting off for more than a month.
What to ignore for now
Anything where being wrong is expensive and hard to spot. Pricing decisions. Anything legal or medical. Anything that goes out under your name without you reading it first. Anything that touches money moving.
Those aren't permanently off-limits, but they're not where you start.
The honest version
Pick one. Not five. The most common failure isn't picking the wrong thing. It's picking six things, half-finishing all of them, and concluding that AI doesn't work.
Take the one that costs you the most this month, fix it properly, and let it run for a few weeks before you touch anything else.