An automation is worth building when the gain shows on paper before any code is written. A simple method: time, errors, total cost, and what cannot be measured.
Almost every automation project starts with a sentence like “we lose too much time on this”. The sentence is true, but it is not a plan. Before you pay someone (or spend your own evenings) to automate a process, it is worth having an estimate on a sheet of paper: what the task costs today, what it would cost afterwards, and what the road between the two costs.
You do not need a financial model. You need four honest questions and the discipline to write down the answers before you see any quote.
Step zero: pick a task, not a department
“Automate sales” cannot be estimated. “Take the leads from the website form into the CRM and send the customer a confirmation” can. The narrower the task, the easier it is to check the estimate afterwards.
A good first task has four traits: it repeats often, its steps look the same every time, it uses data that already sits in some system, and its result can be checked in two minutes.
What to measure: four things
1. Actual time
Not the time on paper, but the real one. Ask someone to time the task five to ten times, at different moments of the week. Note the warm-up time as well: opening files, looking up information, getting back after an interruption. Half of the cost often hides there.
2. Frequency
How many times a day, week or month does the task come up? A ten-minute task that appears twice a month does not justify a project. A two-minute one that appears hundreds of times might.
3. Errors and corrections
What happens when it goes wrong: a quote with an old price, an invoice sent to the wrong customer, a forgotten lead. Note how long the correction takes and, separately, what you lose when the mistake reaches the customer. Do not invent a figure. If you do not know it, write “unknown” and move on to the section about what cannot be measured.
4. Waiting time
Sometimes it is not the work that costs, but the delay: the customer waiting until tomorrow for an answer, or an order sitting in a queue until somebody comes back. Delay has a price, but calculate it with caution and keep it apart from the time saved.
A formula you can do on the corner of a page
The gross annual gain is approximated like this:
- time per run × runs per year = hours spent today;
- hours spent today × cost of an hour of work (salary plus contributions, not just take-home pay) = cost of the task today;
- cost of the task today minus the cost left after automation (checking, exceptions) = gross gain.
For example, in a hypothetical case: a 6-minute task repeated 20 times per working day takes 2 hours a day. If, after automation, a person still checks the result for 15 minutes a day, the difference is 1 hour 45 minutes a day. The numbers are made up only to show the calculation; yours will differ, and if they come out small, that is useful information, not a problem.
The cost of automating: more than the implementation invoice
The second half of the calculation is the one most often left out. On top of the implementation price, add:
- subscriptions and usage of the tools involved (the automation platform, the AI model, the APIs);
- your time and the team’s time to explain the process, test it and correct it;
- maintenance: a workflow connected to other systems breaks when those change;
- supervision: someone has to notice when something fails and deal with the exceptions.
An automation that saves 100 hours a year but needs 60 hours of supervision and fixes has saved 40. Write down both columns.
What cannot be measured (and should be said openly)
Part of the gain does not fit into a formula, and it is better not to pretend it does. A few examples:
- freed attention: the person who no longer copies data can do something else, but that “something else” has to exist;
- calmer teams and fewer mistakes made out of tiredness;
- faster replies to customers, whose effect on sales you cannot isolate precisely;
- lower risk when a key person is away: the task no longer depends on a single individual.
Put these on a separate list, as possible benefits. Do not turn them into money just to get a neat number.
How to check the estimate: a small test before the big project
The best protection against an optimistic estimate is a short pilot. Automate only a fragment (for example one type of request) and measure for two or three weeks with the same tools you used before. Compare with your estimate. If the two differ a lot, you have learned something important while it still costs little.
Decide in advance what result would make you stop. Without a threshold set beforehand, every project finds reasons to carry on.
When automation is not worth it
There are situations where the right answer is “not now”:
- the process is not stable: it changes every few months, so the rules would have to be rewritten all the time;
- the input data is messy and nobody cleans it; fix the data first;
- the task is rare and short;
- mistakes would be serious and nobody is there to check the result.
Saying “no” to a poorly justified automation is also a gain: the money and attention stay available for a better one.
In practice
Start from a single task, time it, write down the full cost of the solution, note separately what cannot be measured, then test in small pieces. If the figures still look good after these steps, you have a project you can defend to anyone, yourself included.
At ME Digital we help small companies do this calculation before any quote; if you want a second opinion, see the AI implementation page or get in touch.