5 min read
This is an illustrative story. Marco is a composite based on common situations solo trades owners describe, not a real person, and the numbers are realistic examples rather than reported results.
It was 9:40 on a Sunday night, and Marco was sitting in his van outside his own house, scrolling through his invoicing app. Fourteen unpaid invoices. Some were three days old, some were five weeks old, and one was from a landlord who always paid eventually but never on time. Marco is a solo handyman. He fixes doors, hangs shelves, patches drywall and assembles furniture nobody else wants to touch. He is good at the work. He is terrible at chasing money.
“I’d rather re-tile a bathroom than write ‘just following up’ one more time,” he says. This is how he stopped writing it at all.
The real problem was not the money
Marco’s clients were not bad payers. Most of them simply forgot. The problem was that reminding them depended entirely on Marco remembering, on a Sunday night, after a full week of physical work. Some weeks he sent nothing. Other weeks he sent a slightly annoyed message to everyone at once, including people who had paid that morning.
When he finally added it up, he had around $3,800 sitting in invoices older than 14 days, and he was spending close to two hours a week on reminder messages that he hated writing.
That detail matters. A November 2026 analysis by the automation consultancy Flow Digital found that 47% of the AI projects it implemented started as an ordinary operational problem a client described, not as an “AI idea.” Marco’s story fits perfectly. He did not want AI. He wanted to stop chasing invoices.
Starting small on purpose
Marco’s first instinct was to look for an all-in-one “AI office manager.” A friend talked him out of it. Instead, he set one rule: one workflow, one AI step, one week of testing.
He already used Zapier (an app that connects other apps and moves information between them) to send new web enquiries into a spreadsheet. The same Flow Digital analysis found that 61% of the AI pieces it built ran as steps inside Zapier workflows, which reassured him that he was not doing something exotic. If you are choosing between automation tools yourself, our breakdown of what one AI step costs in Zapier, Make and n8n is a good place to start.
The workflow he built in one evening
Here is the whole thing, in plain words:
- Trigger: every weekday at 8 a.m., the automation checks his invoicing app for invoices that are 7, 14 or 30 days overdue.
- AI step: for each one, an AI model writes a short reminder using the client’s first name, the job description, the amount and the number of days overdue. The tone changes with the stage: friendly at 7 days, clear at 14, firm but polite at 30.
- Review step: the drafts land in a single daily email to Marco, not straight to clients. He reads them over coffee and approves each one with a tap.
- Send: approved reminders go out from his own email address with the payment link attached.
The prompt was the part he spent the most time on. He gave the AI three of his own past messages as examples, a list of phrases he never uses (“per my last email” was banned immediately) and one firm instruction: never threaten, never mention late fees unless the invoice is over 30 days and the original quote included them.
What went wrong in week one
Not everything worked. The first drafts were too long and oddly formal. One reminder thanked a client “for choosing our services,” which made Marco laugh out loud, because he is one man with a van. He trimmed the prompt, added “maximum three sentences” and “write like a tradesperson texting a neighbor,” and the second batch was much better.
The bigger catch came on day three. A client had paid in cash, and Marco had not marked it in the app. The draft reminder would have gone out if he had not been reviewing every message. That single moment convinced him to keep the human check permanently. It costs him about five minutes a day, and it protects relationships that took years to build.
The results after six weeks
Here is what changed for Marco, using illustrative but realistic numbers:
- Time on reminders: from about two hours a week to roughly 25 minutes, almost all of it quick approvals.
- Invoices over 14 days old: from around $3,800 to under $900.
- Average days to get paid: down from about 19 to about 9, mostly because reminders now go out on time every time.
- Cost: a modest monthly automation plan he already partly used, plus a few cents of AI usage per reminder.
He also noticed something he did not expect. Two clients replied to their reminders asking him to book more work. A polite, well-timed message turned out to be a quiet marketing channel.
What made this work, and what you can copy
Big companies are learning the same lessons the hard way. Genpact’s November 2026 survey of large enterprises found that, on average, only 49% of AI initiatives funded in the previous two years delivered the value expected. Marco’s version succeeded for simple reasons you can borrow:
- He started from a pain, not a product. “I hate chasing invoices” is a better starting point than “I should use AI.”
- He measured before and after. Two numbers (hours spent, money overdue) were enough to know it worked.
- He kept a human in the loop. The review step caught the one mistake that would have hurt.
- He wrote the prompt in his own voice. Real examples beat generic instructions every time.
This is the same pattern we saw with a mobile mechanic who stopped losing calls and a house cleaner who cut quote time: one boring task, automated carefully, beats a dozen half-finished experiments.
Your turn
Open your invoicing app right now and count the invoices older than 14 days. If the number makes you wince, you have found your first AI workflow. What is the one admin task you would happily never do by hand again?
