When a business should adopt AI and where to start

More and more often, companies come to us saying “we want to implement AI.” The first question we ask in return is: what work do you want to hand over to it? Often there is no answer yet, and that’s fine. But in that case, you need to start somewhere else.

AI speeds up work that follows rules. If there are no rules and every employee does things their own way, AI will speed up that mess too. That’s why implementation starts with understanding how the company works today.

The whole path looks like this:

1. 1Describe processes
2. 2Clean up data
3. 3Find routine tasks
4. 4Hand over to AI and test
5. 5Connect to 1C and CRM

Step 1. Describe how the work is done

Take one process and write it down step by step: who does what, in what order, where they get the data and who they pass the result to. A diagram on a sheet of paper or an Excel spreadsheet is enough.

For example, the path of a customer request:

• the request comes in by email, WhatsApp or phone;
• a manager enters it into the CRM or a spreadsheet;
• an engineer calculates the cost and writes a commercial proposal;
• a manager approves the proposal and sends it to the customer.

This step alone usually uncovers a lot of waste: the same data is entered twice, files are sent back and forth by email, three people keep one spreadsheet in three different ways. Some of these problems can be solved without any AI at all.

Step 2. Clean up your data

AI works with whatever you give it. If customers are recorded in three spreadsheets under different names, and prices in the price list don’t match those in 1C, the result will be just as inaccurate.

Agree on where each type of data lives: customers and deals in the CRM, products and prices in 1C, documents in one shared folder. Create templates for recurring documents: proposals, completion certificates, reports. This is the boring part of the work, but it determines whether AI will be useful.

Step 3. Find routine tasks that can be described as an algorithm

Now look at the processes you have described and mark the tasks that have four traits:

• they repeat often, every day or every week;
• it is clear what goes in and what should come out;
• the rules can be explained to a new employee in fifteen minutes;
• a mistake is easy to spot and fix.

Almost every company has tasks like these:

• filling in spreadsheets from emails, delivery notes and scanned documents;
• sorting incoming requests: request type, urgency, person in charge;
• drafting commercial proposals and emails from templates;
• answering common customer questions;
• writing meeting minutes from a call recording;
• checking the timesheet against the work schedule and finding discrepancies.

For now, it’s better not to give AI tasks where a mistake is costly and nobody can check it, or work where the rules change every time: negotiations, non-standard decisions, evaluating people.

Step 4. Start with one task and measure the result

Pick one task from the list, the one that takes the most time. Before launch, measure how long it takes now and how often mistakes happen. Then hand it over to AI for a few weeks and compare.

At first, a person checks every result. This shows where AI makes mistakes and which rules need to be clarified. Once there are almost no mistakes left, you can switch to spot checks.

Step 5. Connect AI to your systems

If the pilot pays off, it’s worth connecting AI to the systems where your data already lives: 1C, CRM, Bitrix24, email, messengers. Then employees won’t have to copy text from one window to another, and the result goes straight where it’s needed.

At this step, it’s important to decide what data can be sent to external services. Personal data of customers and employees and trade secrets are best processed on your own server or in services where the contract allows it.

When it’s too early to implement AI

• processes aren’t documented anywhere and everyone works their own way;
• data is scattered across personal spreadsheets and contradicts itself;
• there is no one who is responsible for the result and will check the AI;
• the task comes up once a month, and automating it costs more than doing it by hand.

In these cases, it’s worth going through the first two steps first. They are useful on their own: work becomes more transparent, and new employees are easier to train.

How we can help

We survey a company’s processes, document them, find tasks to automate and implement solutions: from exporting timesheets to 1C to WhatsApp and Telegram connectors for Bitrix24. If you want to understand what in your company can be handed over to AI, tell us about your task.