The conversation about AI in small businesses tends to get framed as all-or-nothing: either you step into the age of artificial intelligence or you fall behind.
In practice it looks nothing like that. Automating is a list of specific tasks, and each one has its own maths. Some pay for themselves in weeks, others cost more than they save, and a few you shouldn’t touch even though technically you could. This post sorts that list into three groups.
The filter
Before the list, the rule we use to classify. A task is a good candidate when it meets four conditions at once: it repeats, because if it happens once a month automating it costs more than it saves; it has clear rules, because if two people on your team would do it differently without agreeing first, what’s missing is a defined process rather than software; it doesn’t need judgement but consistency, memory, or speed; and it has a measurable cost, meaning you can estimate the hours it eats or the sales lost when it goes wrong.
All four together. A repetitive task without clear rules isn’t ready yet, and a task with clear rules that happens twice a year isn’t either.
What pays for itself
In almost any service business, between 40% and 60% of incoming messages are the same five questions: price, hours, location, availability, payment methods. It’s real work, but it doesn’t need a person, and it’s what eats most of the day for whoever handles enquiries. If you’re going to automate one thing, start here.
Then there’s booking. The typical back-and-forth takes five messages — proposal, counter-proposal, time, confirmation, reminder — and an agent connected to your calendar closes it in one or two. The side benefit usually beats the obvious one: automated reminders cut no-shows, and that’s money you’re currently losing outright.
In sectors where the sales team burns time on contacts who were never going to buy — real estate being the clearest case — qualifying before assigning changes the output of the whole team. Budget, area, decision timeline: an agent asks those just as well, and it asks them at eleven at night. And there’s post-service follow-up, the task everyone knows they should do and nobody does because it’s the first thing to go when the week gets busy. It’s also among the most profitable, since selling to someone who already bought costs far less than finding someone new.
What can wait
There are perfectly automatable things that just aren’t urgent, and confusing them with the group above is a common way to misspend a budget.
Automated reporting has decent returns but is rarely what hurts most, so it belongs after you’ve fixed first contact with customers. Reconciling data between systems is worth a lot once volume justifies it, though in a small business it’s often still cheaper to do by hand once a week. And content generation helps you produce faster but needs constant human review, so it frees up less time than promised.
What you shouldn’t automate
Decisions requiring professional judgement stay with a person: a diagnosis, a technical recommendation, a negotiation. An agent can gather information and set things up, but not make the call. Neither should the relationship with your best customers be automated, because someone who’s been with you for years expects recognition, and automating that contact saves minutes in exchange for something that took years to build.
Urgent issues and complaints go to a human immediately, since someone upset needs to talk to whoever can fix it rather than receive a correct and empty reply. And irreversible decisions — charging, cancelling, deleting records — need human confirmation even when they’re technically automatable, because the cost of one mistake outweighs any saving.
Above all of the above: don’t automate a process that doesn’t work. This is the most expensive mistake and the most common one. Automation amplifies what you already have, so if the process is confused, what you get is faster confusion at greater scale. Sorting that out first isn’t an optional preliminary, it’s part of the job.
Is it your turn yet?
Eight questions. Count how many apply to your business:
- You get more than 50 customer messages a month
- You answer the same five questions constantly
- Messages arrive outside your working hours
- Booking an appointment takes more than three messages back and forth
- Someone on your team spends over an hour a day on WhatsApp
- You’ve lost customers to slow replies, and you know it
- You have no way of knowing how many messages went unanswered
- Your follow-up process exists in theory but not in practice
Two or fewer and it isn’t yet: your constraint is demand rather than your ability to handle it, and that’s where the money should go. Three to five means you have a leak, worth measuring for two weeks before investing in fixing it. Six or more means you’re losing money measurably every week, and probably have been for longer than you think.
Two numbers and a subtraction
To decide whether a specific task is worth it, first work out what the current process costs: the weekly hours it consumes, times a loaded hourly cost — around COP $12,000 for an administrative role in Colombia — annualised, plus lost sales if you can estimate them. Then work out what automating it would cost, which is implementation plus monthly operation.
If the difference doesn’t clearly favour automating, don’t automate yet. And if you can’t calculate the first number, that’s the prerequisite work: measure before deciding.
At Katalis we run free 45-minute diagnostics: we review how your operation works, identify the three tasks with the highest hidden cost, and tell you which ones can be automated now and which shouldn’t be automated at all. If we conclude you don’t need it yet, we’ll say so. Get in touch.