Where to start automating your business when everything feels urgent

Most owners automate the wrong task first. They pick the most visible one, the most painful one, or the one a salesperson told them about. Then they find the time saved is marginal and the system has added a layer of complexity. Here is the method we apply, with our clients and on our own company.

The real problem is not a missing tool

An average small company already runs between eight and fifteen pieces of software. An inbox, a spreadsheet, an invoicing tool, a calendar, a half-filled CRM, two industry applications, and a WhatsApp group acting as the command centre. The problem is almost never a missing tool. The problem is that nothing talks to anything else, and a human being is used as the network cable between each application.

That human cable has a cost. It is slow, it forgets, it gets sick, it goes on holiday. Above all it is expensive: the owner or their right hand often spends a third of the week copying information from one place to another.

Step 1: measure before deciding

For five working days, log every repetitive task in a plain table with three columns: the name of the task, how often it happens, how long it takes each time. Nothing else. No dedicated tool, no time tracking software, a spreadsheet is enough.

After five days, work out the monthly time for each line and multiply it by the loaded hourly cost of the person doing it. You now have the real cost of each task. It is the only number that lets you arbitrate, and it is almost always a surprise.

We ran this exercise on our own property management business before selling anything to a client. One task alone, preparing the daily schedule for the field teams, accounted for more than three hours a day spread across several people. Nobody had noticed, because those three hours were sliced into dozens of invisible micro-actions.

Step 2: rank by frequency, not by how much you hate it

This is the most common mistake. People want to automate the task they hate, when they should automate the task they repeat. A painful task done once a month costs you twelve occurrences a year. A trivial task done six times a day costs you more than fifteen hundred.

So cross two axes:

• Frequency. How many times a week does the task come back?
• Stability. Have the rules of the task changed in the last six months?

The first task to automate is the one in the top right corner: very frequent and very stable. An unstable task, with rules that shift every month, produces an automation you will be repairing constantly. Wait until it settles, or settle it first.

Step 3: write the process by hand before building it

Before any automation, write the process in plain language, step by step, as if you were training someone starting on Monday. If you cannot write it clearly, the process does not really exist yet: it lives in one person's head, as a set of exceptions and habits.

This step almost always reveals two things. First, that the task contains implicit decisions nobody has ever formalised. Second, that part of the work serves no purpose and can simply be deleted. Deleting a task costs nothing and pays back immediately. It is always the best automation available.

Step 4: build small, measure fast

A good first automation is built in a few days and measured within the first week. If your project needs three months before the first visible result, the scope is too wide. Cut it down.

In practice the first block usually looks like this: a trigger (a booking comes in, a form is submitted, a date arrives), a condition (if the amount is above a threshold, if the client is in a given category), an action (create the record, send the message, update the dashboard). Three elements. Nothing spectacular, and that is exactly what makes it reliable.

Then measure the only thing that matters: human time recovered per week. Not the number of runs, not the number of active scenarios. Hours.

Where does artificial intelligence fit in?

AI is not a starting point, it is one building block among others. Most of the gains in a small company come from simple deterministic rules that always produce the same result. AI becomes useful when something has to be interpreted that rules cannot describe: classifying a customer message, extracting information from a document, summarising a conversation, drafting a first version.

Putting AI everywhere costs more and makes the system unpredictable. The right reflex is the opposite: solve everything you can with rules, and reserve AI for the places where judgement is genuinely required. Always with a human validating anything that commits the company.

The three mistakes we see most often

1. Automating a broken process. Automation repairs nothing, it accelerates. A shaky process, automated, produces errors faster.
2. Building everything at once. The big global project ships late, costs three times the budget, and nobody uses it. One block at a time, put into production, measured.
3. Skipping documentation. An automation nobody understands becomes a black box. The day it breaks, it blocks the company. Every block needs its page: what it does, what triggers it, who to call.

Where to start, concretely

If you take one thing away: open a spreadsheet, keep the log for five days, work out the cost of each repetitive task, and attack the one that is both the most frequent and the most stable. Within a week you will know whether the subject deserves real investment, and you will know it with numbers rather than intuition.