Which Task to Start With: Choosing Your First Process to Automate

Which task should an SME automate first?

Short answer: start with the task that comes up most often, follows rules you can write down, costs real money when it goes wrong, and whose output can be checked in thirty seconds. A task with those four traits makes a good first process. A spectacular but vague one makes a bad one.

Most small businesses do not lack automation ideas. They have ten, often more, and each has a champion on the team. The real obstacle is the choice: without a method, people pick the most visible task or the one that irritates them most, and end up with a project that is too broad and never finishes. This article gives a simple scoring grid, a worked example, the most common mistakes, and the cases where it is better to automate nothing for now.

Why the first process decides what comes next

The first automated process acts as proof. If it runs within a few weeks and saves visible time, the team follows and the second one is easier to launch. If it stalls, the internal conclusion becomes “AI does not work here”, when the real problem was the scope.

Guides published by automation vendors agree on one point: a single process and a short pilot. Several mention four to eight weeks for a first pilot. Those are vendor benchmarks, so treat them as orders of magnitude, not commitments. What matters is the principle: a scope small enough to finish, real enough to measure.

The five criteria of a good first process

These are the five questions we ask a business owner before picking a task. Score each from 1 to 3: 1 if the answer is no, 2 if it is partly, 3 if it is clearly yes.

1. Frequency. Does the task come up every day or every week? A yearly task almost never justifies the upfront effort.
2. Written rules. Could a new hire do it from a one-page instruction? If the answer depends on the client’s mood or a manager’s hunch, the task is not ready.
3. Cost of an error or a delay. What is lost when the task is forgotten or done badly: an invoice paid late, a client waiting three days, a wrong report? The more visible the cost, the more visible the gain.
4. Accessible data. Is the starting information already in a tool you use: mailbox, calendar, CRM, ERP, shared spreadsheet? If it lives on paper or in one person’s head, it has to come out first.
5. Quick check. Can a person verify the result in under a minute? A draft reminder is read in ten seconds. A strategic analysis is not.

The total runs from 5 to 15. Below 9, the task is not ripe: clarify the process before touching it. Between 9 and 12, it is possible but needs preparation. Above 12, it is a first-rank candidate.

A scoring example, task by task

The table below is an illustrative example for a service business that handles invoicing and client follow-up. The scores are ours for this case; yours will differ, which is the whole point of the exercise.

In this example, chasing unpaid invoices wins: it comes up every week, its rules fit on a page (deadlines, tone, escalation), a delay translates straight into cash flow, the data sits in the accounts and a draft is quick to read. Choosing a supplier is the counter-example: rare, vague, hard to verify. It is not a starting task for automation.

Put a number on the task before choosing it

A score is not enough; you also need an order of magnitude in hours. The formula is simple: duration of one occurrence, times occurrences per week, times weeks worked.

Take a hypothetical: an assistant spends 20 minutes a day chasing invoices and logging replies. That is a little over an hour and a half a week, or about 80 hours a year over 48 weeks. At a fully loaded cost of 60 francs an hour, that is roughly 4,800 francs of time per year. The figure is an example, not an average: measure yours with a week of real tracking, not a memory-based guess.

The calculation leaves out the hidden cost: invoices chased too late, clients annoyed by a clumsy message, the mental load of remembering. It still gives a floor. If the floor is a few dozen hours a year, the gain will be modest. If it reaches several hundred hours, the task deserves to go first.

Processes that make good candidates

Some tasks almost always top the list in small businesses, because they tick all five criteria:

• Follow-ups (chasing unpaid invoices, following up on a quote): frequent, clear rules, verifiable output.
• Reading and filing documents: high volume, but you must plan who decides when the reading is uncertain.
• First sorting of inbound requests: a message received in the evening should not wait until morning to be sorted.
• Preparing recurring reports: the figures already exist, assembling them is what takes time.

What they share: the AI prepares, a person validates. That split is what makes a first project safe. You will find the detailed use cases in the processes the studio automates.

When not to automate, or not yet

This section matters as much as the others: automating too early costs more than waiting.

• The process is not stable. If the team does it three different ways, automation will freeze the confusion. Put the rules down first.
• The task is rare. A few occurrences a year will not repay the preparation time.
• The result is hard to verify. If nobody can say within a minute whether the output is good, you will not know either when it degrades.
• The data is missing or scattered. The first job is then collection, not automation.
• The error is irreversible and serious. A binding decision with no possible review is not ground for a first trial.

In these cases the right answer is sometimes to install nothing and simplify the process by hand. A simpler process already costs less, before any AI.

The method in six steps

1. List your repetitive tasks. Ask each person to note for one week what they redo several times. Ten to fifteen lines are enough.
2. Put hours on each line. Use the formula above, with measured durations.
3. Score the five criteria. A score from 1 to 3, by two people if possible, to limit bias.
4. Pick one task, not three. Take the best-scoring one and resist launching several projects in parallel.
5. Define the success measure before starting. Hours saved, response time, error rate: one indicator, recorded before and after.
6. Run a short pilot and review it. A few weeks, one person validating, a review at the end. Only then move to the next task.

If you would rather not run this inventory alone, the express diagnostic exists for exactly this: spotting the task that costs the most hours and saying whether it suits automation.

The most common mistakes

• Picking the most visible task. The loudest is not the costliest. The hour log shows it.
• Starting with the tool. People choose a platform, then look for a use. Do the reverse: process first, tool second.
• Aiming for zero human involvement. For a first project, keeping a person who validates is a safety choice, not a failure.
• Forgetting to measure. With no figure before, you cannot prove the gain after, and the project loses support.
• Automating an already broken process. A broken process, automated, produces errors faster.

And after the first process?

Once the pilot has been reviewed, the grid becomes a queue: tasks already scored stay ranked, and each new project benefits from what was learned on the first. The data is already connected, the rules written, and the person who validates knows the circuit. The second process therefore starts faster, provided you do not skip the measurement step.

Revisit the grid every quarter as well. A task that scores low today because its rules were vague may become a candidate once the team has clarified them.

In short

The best first process is rarely the most ambitious one. It is the one that comes up often, whose rules fit on a page, whose errors cost something and whose output is checked fast. Score your tasks on five criteria, put hours on them, choose a single one and measure before moving on.