The most common mistake is starting with a sweeping «let us automate everything» project. It takes long to roll out, costs a lot and falls apart halfway because the effect never has time to show. It is far safer to take one process, drive it to a result and expand from there.
Here are five processes companies start with most often — roughly in order of how quickly they pay off.
1. Customer replies in messengers
The most obvious candidate. People write on WhatsApp in the evening and at weekends, and get an answer on Monday morning — if at all. Half of the typical questions repeat day after day: price, availability, timing, delivery terms.
What it gives: requests stop getting lost overnight, the manager is freed from repetition, first-response time drops to seconds.
When it fits: more than 20-30 requests a day and a noticeable share of repeating questions.
2. Triage of incoming requests
Requests arrive by email, from the website form, from direct messages and messengers. A manager moves them into the CRM by hand, losing some and creating duplicates along the way.
What it gives: AI reads the incoming message, extracts the name, contact, the essence of the request and the budget, creates a record with fields filled in and flags the urgent ones.
When it fits: requests come from three or more channels and moving data takes at least an hour of someone's day.
3. Working with documents
Invoices, delivery notes, contracts, acts. They get retyped into spreadsheets by hand — slowly and with errors that accounting later hunts down.
What it gives: data is extracted from PDFs, scans and photos and goes into 1C or spreadsheets. A person reviews the doubtful cases instead of typing every line.
When it fits: from a few dozen documents a month with repeating fields.
4. Reports and digests
A manager asks for «the sales summary» and someone spends half a day assembling it from three systems. By the time it is ready it is already out of date.
What it gives: a daily digest arrives by itself — to email or a messenger, with the numbers and a short note on what changed.
When it fits: the data sits in systems you can access and the report is needed regularly in one format.
5. Internal knowledge base
Staff pull colleagues away with questions about policies, client terms and internal procedures. Newcomers especially, in their first months.
What it gives: an assistant answers from your documents and links to the source so the answer can be verified.
When it fits: the company has more than 15-20 people and accumulated policies.
How to pick your first process
Take three criteria and score the candidates against them.
- Volume. The process repeats daily, not once a month
- Measurability. It is clear what to count the effect in: hours, requests, errors
- Data readiness. The materials exist in digital form, not only in an employee's head
The process that passes all three is the right place to start. If none passes the third, the place to start is not AI but putting your data in order. We say so honestly when that is the case.
How to work out whether it is worth it
Multiply the hours the process eats per month by the cost of an employee hour. Add the cost of errors and lost requests if you can estimate it. Compare that with the price of automation plus monthly running costs. If payback comes out under a year, the task is almost certainly worth doing.
Happy to run the maths on your numbers: message us on WhatsApp and we will review the processes and say what to automate first and what to leave alone.