← All articles
Article · Potential · 6 min read

Scaling Without Hiring: Three Criteria to Choose Which Tasks to Automate First

By Anis Hammouche·November 15, 2026

When your activity grows, automate before you hire, provided you choose the tasks carefully. Start with the ones that come back often, take a lot of your time, and where an error is easy to spot and fix. These are the tasks that free up the capacity you need to sell, deliver and decide. A badly chosen automation costs as much as a badly placed hire.

Why automate before hiring

Hiring is not a decision you can easily undo. A new collaborator needs time to be trained, time to be supervised, and only becomes productive after several weeks. An automation can be tested in a few days, switched off without conflict, and corrected without an interview.

There is a second point. A repetitive task you hand to someone without simplifying it remains a repetitive task. You then pay a person to do work that a clear rule would do better. Automating forces you to describe the process, and that is often where you discover what serves no purpose.

A useful link on this point: the AI skills gap is not a hiring problem. Automating is first a question of method, not of headcount.

The three-criteria grid

For each recurring task, answer three questions. Note the answers on one line, in a few seconds.

CriterionQuestion to askWhat makes it a priority
FrequencyHow many times a week do you do it?The more often it comes back, the more the gain accumulates
TimeHow many minutes does each occurrence take, counting interruptions?The longer it is, the more visible the gain
Error riskIf the output is wrong, what happens, and do you spot it quickly?A low risk lets you automate without a heavy safety net

The three criteria combine. A frequent, long, low-risk task goes first. A frequent task with high risk can be automated, but with a human check step. A rare task, even a long one, often does not deserve automation for now: the cost of setting it up exceeds what it returns.

Where to start: example tasks

The most profitable tasks for a small business are often transfer tasks: copying information from one tool to another, sorting incoming requests, preparing a document from data you already have, writing a first draft of a standard reply.

As a worked example, and not a real case: a twelve-minute task repeated five times a week takes one hour a week, roughly four hours a month. If automation brings it down to two minutes and creates no errors to fix, the gain is about three hours a month. Compare that figure with the time needed to set the automation up.

The table below gives an indicative reading. Your context may change the order.

TaskFrequencyError riskIndicative priority
Copying data between two toolsHighLow, and easy to checkAutomate first
Sorting and tagging incoming requestsHighMedium, a wrong category is easy to correctAutomate, with a weekly review
Preparing minutes from notesMediumLow to medium, with a reviewAutomate, review mandatory
Answering a client question about a priceHighHigh, a wrong answer is a commitmentPrepare a validated template, not a free answer
Deciding a price or a contractLowVery highDo not automate

The line on client replies matters most. An AI that answers price or terms questions on its own creates commitments you will have to honor. A reply template that you validate once, and that the tool adapts to each case, is safer.

Measure before you automate

Many people are wrong about how much time they spend on a given task. Before choosing, keep a log for two weeks. It does not need to be perfect:

Prompt
Date | Task | Duration (min) | Interrupted? | Error found?
14/11 | Copy quote into CRM | 12 | yes | no
14/11 | Answer form requests | 25 | no | yes (wrong link)

After two weeks, the task list almost sorts itself. You can see the ones that come back, the ones that take the most time, and the ones that already break something. Those are your candidates.

Automate without breaking what works

Once the task is chosen, change only one thing at a time. If you automate three processes in the same week, you will not know which one causes the problem when an output is wrong. Follow these four rules:

  • One flow at a time. Test for a week before adding another.
  • One checkpoint. Any output that goes to a client or to a payment tool keeps a human check.
  • An error log. Note every wrong output and its cause. That log tells you where to strengthen the process.
  • A way back. Keep the manual method documented, so you can resume in under an hour if the tool fails.

When automation is no longer enough

Automation frees up time, but does not always replace the work. When a task stays repetitive after automation and still needs human judgment, that is a signal, not a failure. It is time to think about how work is shared, about training, or about a targeted hire for that specific point.

The goal is not to never hire. It is to hire for what truly needs a person, with a clear process to hand over.

Frequently asked questions

What is the first task to automate? The one that is frequent, long, and where an error shows up immediately. Most often it is copying data between two tools. Measure before deciding, because gut feeling about time spent is often wrong.

How do you know whether an automation is worth the set-up cost? Multiply the time saved per occurrence by the frequency, over a month. Compare that figure with the set-up time and the time spent correcting errors. If the gain does not exceed the effort over a few months, move on to another task.

Do you have to automate everything to scale? No. Automating transfer tasks frees up your capacity. Tasks that require judgment, a relationship or responsibility stay with you or with a person, even when your activity grows.

Stuck on a specific case? Write to me through the contact page.

Get the AI news

Sign up and you get the guide "The useful AI automations in 2027", then at most one email a week with my new articles and tutorials.

At most one email a week, plus the guide for free. Unsubscribe in one click, the guide stays yours. Privacy

Related articles

Related resources