Process automation

Where to start with automation if nothing in your company is automated

Aleksandr Lukashevich August 12, 2026

The most common mistake at the start is beginning with a tool. A company picks a service, buys a subscription, someone spends two weeks setting it up, and then it turns out they automated something that happens four times a month.

Below is the order of steps for the first two weeks. It's boring and needs no budget and no developers. But afterwards you know what to tackle.

What "nothing is automated" means

It's usually not a blank slate. Usually it looks like this: there's a CRM that people fill in by hand. There are conversations in three messengers. There's a spreadsheet where orders are collected. And there's a person who keeps all of it in their head and reconciles it once a week.

So the systems are already in place. There's just a person working between them. Automation doesn't start with buying a new system. It starts with seeing exactly where a person is acting as an adapter.

Week one: measure

We don't implement anything. We only measure. The goal for the week is to get numbers you can compare against later.

Days 1–2. Export from the CRM

If you have a CRM, you need three exports from it: all deals with statuses and the date of the last action, all tasks with their names, and all records with a "source" field. If there's no CRM, take whatever spreadsheet or notebook you keep records in.

Days 3–4. Look at the conversations

How many client messages came in over the month, and through which channels. How many of them arrived outside working hours. How many went unanswered for more than an hour. You don't need precision here. You need the order of magnitude.

Day 5. Ask your people

One question for everyone who works with clients: what do you do every day, and what annoys you most? Not "what should we automate". People answer that one with stock phrases. "What annoys you" gets an honest, precise answer.

Three numbers to start with

You only need to pull three metrics from the exports. We ran this kind of audit for a print manufacturer and got this picture:

Stalled deals show that no one is closing the process, and that's pure work for automation. Repeating tasks tell you directly what to do first: if the same action is done thousands of times a year, it belongs in a rule, not in a manager's head. Records without tags are a sign that decisions are being made blind.

Once you've counted these three numbers, you've done more than half the work. All that's left is to choose.

What this looks like in a real business

An audit breakdown: which numbers came up, what we automated first and what we got in five weeks.

Read the case

Week two: pick one process

Exactly one. Not three, not "an area", not "all of support". One process with a beginning, an end and a clear result.

Then answer four questions about it:

1. How many times a month does it happen? Fewer than fifty? Put it aside. The savings won't pay for the setup.
2. How long does it take each time? Multiply by the count and you get the cost of the process in hours.
3. What needs to be in place for it to work? If the data is missing or messy, fix that first.
4. What happens if the automation gets it wrong? If the answer is "the client will be upset", go ahead. If it's "we'll lose money", only with a person confirming.

How to pick the right first process

A good first candidate looks like this: it happens often, takes little time each time, the data for it already exists, and the cost of a mistake is low.

Usually it's one of three things:

• Collecting incoming messages in one place. Not replying, just making sure messages from all channels end up in one queue and nobody gets lost.
• Answering five standard questions. The ones people ask every day and that already have a written answer.
• A mechanical action in the CRM. Creating a task, filling in the source, closing what's been hanging for three months.

A bad first candidate is whatever seems most painful. The most painful thing is usually also the most complex: people, money and exceptions are all tangled up in it. You can't start there. Otherwise the first failure will shut down the topic of automation in the company for a year.

Four mistakes at the start

Starting by buying a tool

The tool is chosen for the process, not the other way around. Until the process is described, every service will "not fit", and that's true, because there's nothing to fit.

Automating chaos

If half the records in your CRM have no phone number, the automation will dutifully send messages into the void. Data hygiene first, rules second.

Not measuring the "before"

The most frustrating mistake. Three months later everything works, but you have nothing to prove it got better, and the project looks like an expense rather than an investment.

Launching for everyone at once

The first process goes live on part of the flow: one channel, one client segment, one shift. There will be mistakes. Let them happen on ten requests, not a thousand.

Who should own this

Not "the IT guy". Process automation is the job of whoever knows the processes: the head of a department, a senior manager, the operations director. A technical person is needed later, for the connection work.

And you need one person responsible. Not a committee. A process that three people are responsible for has no owner, which means in a month nobody will remember why a given decision was made.

What comes next

After the first process you have something you didn't have before: an understanding of how long a change takes and what effect it has. That lets you choose the next step by the numbers, not by gut feeling.

Next usually comes expanding within the same area (more scenarios, more channels) and only then neighboring processes. Keep the urge to take on everything at once in check: the more seams between pieces of automation, the more places where things break.