What to automate in a company first
Most failed automation projects do not fail on the technology. They fail because the wrong process was chosen to start with. Here is how to pick the right one.
The first process to automate should meet four conditions at once: it repeats often, it runs the same way every time, its input is structured, and nobody has to make an expert judgement inside it.
In practice that almost always means retyping data between two systems — from an email into a spreadsheet, from a shop into invoicing, from a form into a CRM. It is not the most interesting task in the company, but it is where the investment comes back fastest and where the least can go wrong.
Do not start with what hurts most if it is also complicated. The first project's main job is to prove this works — to you and to your team.
Four criteria for a good first process
1. It repeats
At least several times a week. The higher the frequency, the faster the payback — there is no point automating a one-off task.
2. It runs the same way every time
If there is a sentence beginning “but when it's from this supplier, then differently”, and there are five such exceptions, start with another process.
3. The input is structured
A form, an order, a row in an export. Unstructured input can be handled too, but it costs more and belongs to the second round.
4. It needs no judgement
If it involves deciding on a price, an exception or a client, automate at most the preparation of the material.
If a process meets all four, it is almost certainly a good first candidate. If it meets two, keep it for later.
A scoring table — choose in ten minutes
Write down the five processes that eat the most time in your company and score each from 0 to 3 in these columns:
| Criterion | 0 points | 1–2 points | 3 points |
|---|---|---|---|
| Frequency | once a month | weekly | daily or several times a day |
| Time per run | under a minute | a few minutes | tens of minutes |
| Sameness of the procedure | different every time | a few exceptions | always the same |
| Structure of the input | free text, paper | an email with an attachment | a form, an export, an API |
| Need for judgement | an expert decides | occasionally needs assessment | none |
| Cost of an error | high (contract, money) | medium | low, fixable |
| Systems available | paper, closed software | export/import | an open API |
How to read it: the process with the highest total is your first candidate. Below a total of 10, postpone automation — it will not pay back. At 15 and above it is worth asking for a quote.
This table is not science, it is a conversation tool. Its main value is that it forces you to name the process concretely — and that is usually the moment it turns out “we do that every day” actually means twice a week.
Typical candidates by type of company
E-shop
- Order → document → stock → customer notification
- Creating products from a supplier's source material
- Answering questions about order status
Services and trades
- Enquiry from a form or message → record → notification → reminder if nobody responded
- Booking a slot and a reminder before it
- A quote built from predefined line items
Accounting and administration
- Incoming invoice from email → data extraction → record
- Monthly figures currently copied from three sources
- Checking whether all expected documents have arrived
B2B and sales
- Lead from the website → CRM → assignment → follow-up
- Summarising a call or an email thread into a note
- A regular report for management
Notice that none of these examples is eye-catching. That is intentional — a first automation should be boring and reliable.
Five mistakes on a first project
Starting with the hardest thing. “Since we're doing it anyway, let's build the whole system.” A big first project has a long road to the first result and a high chance the requirements change along the way.
Automating a process that should rather disappear. Before automating, it is always worth asking whether that step has to happen at all. Sometimes the best automation is removing the step.
Not handling errors. Automation that fails silently on a problem is worse than manual work. Every process needs a notification and a record.
Not telling people. If the team does not know something now runs by itself, they will keep doing it — or, worse, stop trusting the system at the first exception.
Not measuring the result. Without knowing how long it took before, you cannot say whether it paid off — and you have no basis for deciding on the second step.
When it is not worth automating anything
There are situations where the honest answer is “not now”:
- The process is changing right now. Switching shop platform, accounting system or supplier? Wait. Automation built on something that disappears in a month is money thrown away.
- Nobody can describe the process precisely. If two people disagree about how it works, it has to be settled first. Automation will not settle it, it will only set one version in concrete.
- The volume is small. As a rough rule: if a process does not take at least five hours a month, deployment and maintenance will cost more than they save.
- The problem is capacity, not routine. If you are short of people for expert work, automating the routine helps but does not replace them.
What to do this week
- Log interruptions for three days. Every time somebody retypes data or answers the same question, one line in a spreadsheet.
- Sort by time. It usually turns out the biggest drain is not the thing people talk about in the company.
- Score the three biggest using the table above.
- Describe the highest-scoring one step by step — who, what, in which tool. That description is the entire brief a supplier needs to quote.
Most companies skip the first step and guess instead. Three days of logging is the cheapest investment in the whole project.
Frequently asked questions
Which process should be automated first?
The one that repeats often, runs the same way every time, has structured input and needs no expert decision. In practice that tends to be retyping data between two systems — not the most interesting task, but the one with the fastest payback and the lowest risk.
From what volume does automation pay off?
As a rule of thumb, from around five hours a month spent on the process. Below that, deployment and maintenance usually cost more than the saving. The exception is processes where an error costs a lot of money — there it pays off at lower volume too.
Should we automate what hurts most?
Not necessarily. The most painful process tends to be the most complicated one too. The first project's main job is to show this works — so it is better to start with something simple with a fast result and come back to the hard thing with confidence and experience.
How do we find out how much time a process really takes?
Log every interruption and every data retyping for three days — one line in a spreadsheet. Estimates are badly wrong in both directions, and this simple log is the cheapest investment in the whole project.
What if our process keeps changing?
Then wait with the automation. A solution built on a procedure that changes in a month will need rebuilding faster than it can save anything. Stabilise the process first, automate it second.
Do we have to automate the whole process at once?
No, and usually it is better not to. Often it is enough to automate the most tedious stretch — moving the data, say — and leave the rest with a person. Partial automation with a clear boundary is more reliable than an ambitious one that is occasionally wrong.
What does a first automation project cost?
One bounded process starts at €390. The exact price depends on the number of integrations and the quality of your data. If you have the process described step by step, you can get a price almost immediately.
Related services
Related articles
Related work
Scored a process and not sure what next?
Send me its step-by-step description. Free of charge I will tell you whether it is worth automating, what of it stays with a person, and what price range it falls into.
Get a free consultation →