Automation· July 5, 2026

Business automation: where to start and what to automate first

One in four employees in Russia spends more than six hours a week on reports nobody has automated — that is data from an hh.ru survey conducted in 2025. Almost a full working day a week goes into moving numbers around. Business automation does not start with buying a system; it starts with answering one question: which manual area is costing you the most.

This guide maps the whole route: how to tell it is time, what to automate first, which path to take (an off-the-shelf system, a no-code builder or custom development), where to put AI, and how to avoid ending up among the 70% of digital transformations that, according to BCG, fall short of their goals.

What does business automation actually deliver?

In short: automation means an operation is entered once, at the moment it happens, and from there it flows into accounting, reports and analytics by itself. Everything employees currently carry by hand between spreadsheets, emails and systems starts happening without them — faster and without errors.

The effects are measurable. In a study by SberAnalytics and Sber Business Soft (November 2025, a survey of 559 respondents), companies that adopted automation report: routine tasks are completed faster — 45% of companies; the share of manual operations has dropped — 37%; there are fewer errors in documents and reports — 36%. The most commonly automated areas are document flow and request handling (70%) and accounting and financial records (55%).

For an owner this translates into three things: employees do the work instead of moving data; the numbers reconcile without manual checks; and the state of the business is visible right now, not in last month's report.

How do you know it is time to automate?

If answering "how much did we earn this month" takes half an hour of reconciliation and three people, it is time. That moment arrives earlier than it seems: manual work grows quietly alongside the business, and a company gets used to it as the norm until it starts costing serious money.

The typical symptoms:

The same figure is entered by hand in two or more places — and it is different in each of them.
There is an employee whose job is half "reconciling" spreadsheets.
Requests live in managers' personal chats, and nobody has the full picture of the customer base.
One person leaving paralyses the process: only they know "how the file works".
The owner's report is out of date by the time it is ready.

If you recognise yourself in two or more of these, you are already paying the "Excel tax": in employee time, in bookkeeping errors, and in decisions made on stale numbers. A detailed breakdown with seven signs is in the article on when a business should move off Excel.

Why you cannot automate chaos

What gets automated is a process that already works — a rule that a sizeable share of projects breaks against. If requests get lost, responsibilities are blurred and records are kept from memory, then automation only speeds the chaos up: the mess multiplies at machine speed and people can no longer catch the errors.

"AI is a magnifying glass. If there is a mess under it, you will see a very large and very detailed mess" — Stas Cheprasov, entrepreneur, from an account of automating a small business on vc.ru.

That is why step zero of any project is taking the process apart: where the data is born, who enters it, who needs it next. We sit down with the people doing the work by hand and write down the real process — usually a day or two, not months of drafting procedures. The real process almost always differs from how the manager pictures it.

What to automate first: do the maths in money

The first thing to automate is the area with the highest cost of manual work. You calculate it like this: frequency of the operation × time per operation × the employee's rate + the cost of errors. The formula takes the emotion out of the choice: the winner is the area with the largest monthly loss. That turns the automation decision into an ordinary investment decision — with a payback period.

An example. Reconciling the bank against the books takes the accountant two days a month — that is 24 working days a year, a full month of work. Add the cost of errors: a discrepancy found a month later means hours of investigation, and sometimes real money. An area like that pays back automation faster than any other.

The statistics agree: the first things companies automate are usually document flow and request handling, accounting, and customer support — high-volume, repetitive processes that are expensive to run. The classic candidates for a first step:

Reconciliations: bank against books, stock against sales, settlements with counterparties.
Moving data between systems: from email into a spreadsheet, from a spreadsheet into 1C, the accounting and ERP platform most Russian companies run on.
Standard documents: invoices, contracts and acts filled in by copy-paste.
First contact from customers: standard questions, qualification, booking.

How to choose the area and avoid sinking into a "year-long project" is covered in a separate method: automating routine work: where to start so it pays back on the very first step.

If you would rather skip doing the maths yourself — we do it in the first consultation: we go through the processes, find where money and time are leaking, and set the goal in money. That is how every one of our custom development projects begins.

Off-the-shelf, no-code or custom development: choosing the path

The path depends on two things: how standard your process is, and what share of turnover you are prepared to spend on automation. Everyone selling an option pulls in their own direction — for a platform vendor, customisation is "something almost nobody needs"; for developers, builders are "toys". Here is a neutral frame:

Path	When it fits	The hidden cost
An off-the-shelf system	Standard processes: accounting, EDI (legally binding electronic document exchange between companies in Russia), a standard CRM	The process bends to someone else's logic; refinements go in as workarounds; the subscription lasts forever
Builders (no-code)	Testing hypotheses, simple internal tools	Platform prices rise, complexity hits a ceiling; the platform shutting down means losing the product
Custom development	The process is non-standard and makes money; you need deep integrations and control over data	More expensive and slower at the start; requires a contractor who knows what they are doing

Testing a hypothesis even before a builder is often faster with vibe coding — AI writes a prototype from your description in an evening. The boundary is the same as with no-code: as soon as real data, payments and integrations appear, you need custom development.

A budget benchmark: for a company with revenue in the hundreds of millions, spending 1–2% of turnover on automation is acceptable. For a business with a turnover of 5–10 million, the same spend becomes 10–20% of turnover — critical, and a "grown-up" system is contraindicated for a company that size. The place to start is a single targeted solution.

The arrangement that works in practice is a hybrid: a standard core where the process is standard, and development where the process is what sets you apart from competitors. We build on that principle: the catalogue of ready-made solutions covers the fast start, and custom development covers what makes your business your business.

Where you need AI, and where classic automation is enough

AI is not needed everywhere, and that is fine. Rules, integrations and schedules cover the bulk of routine work more cheaply and more predictably — with no hallucinations and no fine-tuning. A model makes sense where the input is unstructured and the volume is high: free-form customer questions, emails, documents in arbitrary formats.

Task	What works
Approvals, document routing, notifications	Classic: if-then rules
Syncing systems, reports, auto-generated documents	Classic: integrations and templates
First-line support, triaging incoming requests	AI: free-form questions, qualification
Extracting data from emails and documents	AI: unstructured input
Decisions with a high cost of error (money, legal)	A human, with the system preparing the data

The economics of AI on the first line have already been calculated: according to a Sber study, AI assistants in support and call centres cut payroll costs by up to 30–40%. In our case study with a property developer, a first-line bot took over the flow of initial enquiries entirely — managers only work with warm leads. When such an agent pays for itself and when it stays a toy is covered in the article on the first-line AI agent.

What does automation cost, and how do you calculate payback?

There is no single price list for automation — and that is normal. The cost depends on how many areas are involved, how deep the integrations go, and whether you are building a targeted solution or a complete environment. Do not trust fixed "all-in" price tags: behind them is either a template that will bend your process, or extra charges as the project goes on.

The right frame is investment. You have calculated the cost of manual work in an area (the formula above) — now compare it with the price of the solution and you have a payback period. A project with no clear payback period is not worth starting, however impressive the technology looks.

We describe our approach to estimating on the page about development pricing: first the audit and the goal in money, then a quote for the specific task. That way the client pays for a result they calculated themselves.

How not to lose data: Federal Law 152-FZ and turnover-based fines

Since May 2025, a personal data breach has stopped being an abstract risk. A first breach costs a company a fine of 3–15 million rubles; a repeat breach brings a turnover-based fine of 1–3% of annual revenue, from 20 to 500 million. These are the rules currently in force under the Administrative Offences Code after the 2025 amendments, and an automation project is exactly the moment when data starts moving between systems.

For a project this means three requirements for the contractor: data and hosting inside Russia, a contract with an NDA and a clear split of liability, and role-based access — who sees and changes what. Uploading your customer base into public foreign AI services "just to test" is a way to join the breach statistics.

We build environments on the client's own infrastructure or on servers in Russia, in line with Federal Law 152-FZ, Russia's personal data protection law. Sensitive data does not leave your perimeter without approval — that is written into the contract.

What an automation project looks like: the stages

A project that reaches a result is built the same way in almost any niche: first the audit and the goal in money, then the architecture, then rollout in iterations and ongoing support. On paper this sounds obvious; in practice the projects that reach a result are the ones where none of the four stages was skipped. Our process looks like this:

Understand the task. We look at the processes first-hand and find where money and time are leaking. We set the goal in money — without it the project can neither be estimated nor accepted.
Design the system. Architecture for your niche and scale: what we build, what we integrate, in what order. This is also where the fate of the "zoo of services" is decided — what to keep and connect, what to replace.
Build and roll out. The system is assembled in iterations: the first working area appears within weeks, rather than "everything at once at the end". Data is migrated and the team is trained.
Support and develop. The business changes and the environment is extended: new integrations, new areas, new automations as you grow.

Connecting the systems you already have is a separate conversation: 1C in accounting, a CRM in sales, the bank, messengers. Replacing everything with a single "mega-system" is usually unnecessary — the better move is to connect what already works into one environment. How that is done technically is covered in the article on integrations.

Why 70% of transformations miss their goals — and how to be in the 30%

According to BCG research, 70% of digital transformations fail to meet their stated goals — and automation projects are no exception. The reasons for failure repeat so consistently that you can check them off as a list before you even start. Here are the five most expensive mistakes we see most often:

Digitising chaos. The process did not work manually — now it does not work automatically, only faster.
Copying "what everyone else does". Someone else's funnels and enterprise systems inside a small business produce an expensive system nobody uses.
Building a patchwork quilt. The systems are not connected and there is no single source of data.
Not defining metrics before the start. There is nothing to prove the effect with — and the project is written off as a failure even where it worked.
Not appointing an owner. Scenarios break, data drifts apart, and nobody notices until it is too late.

"One customer gets created three times. Every system has its own fields, its own formats, its own logic" — Dmitry Goroshko, from an analysis on vc.ru of how automation starts working against a business.

Every item on that list is closed off during the audit and design stages — which is why we do not start development without them.

The first 90 days: an action plan

In three months you can go from "everything by hand" to your first automated area with a calculated payback — without buying systems on step one and without stopping work. The plan below is drawn from our own process and assumes you are working through it alongside your normal workload:

Weeks 1–2: walk through your processes asking "where is the same action repeated", "where do mistakes happen most often", "where do people wait longest". Note down 3–5 candidates.
Weeks 2–3: calculate the cost of manual work for each candidate using the formula frequency × time × rate + cost of errors. Sort by money.
Weeks 3–4: pick the single leading area. Define the success metric before you start: hours, errors, speed — exactly what will change and by how much.
Month 2: launch the first area — with a targeted solution or with development. The old process runs in parallel for a while: the numbers are cross-checked and trust in the system grows.
Month 3: compare the metric against the "before" baseline and calculate the actual payback. Only then choose the next area.

Tracking the effect is easier with numbers that update themselves: which indicators an owner should see daily is covered in the piece on the owner's dashboard.

If you would like to walk this route with a team that has 300+ automation and AI projects in Russian companies behind it — tell us about your task: we will go through your processes, calculate the cost of your manual work, and propose a plan with a goal in money.

Sources

Frequently asked questions

Is automation only for large companies?+

Not anymore. In the 2025 SberAnalytics study, small businesses account for 22% of the companies automating their processes: by Sberbank's assessment, automation has stopped being the preserve of large players. For a small company there is one rule: start by automating the single most expensive area, not by buying a "big system".

How long does an automation project take?+

Automating one specific area takes weeks. A full environment (an ERP ecosystem tying together accounting, sales and analytics) takes months, but we roll it out in iterations: the benefit shows up during the project, not only at the very end.

Can we start with an off-the-shelf solution instead of custom development?+

Yes, if your process is a standard one. For a fast start we have a catalogue of ready-made solutions — an inventory bot, a first-line AI agent, a financial calendar and others. Each one is adapted to your processes, and when the business hits their limits, the environment is extended with development.

What happens when the process changes after the rollout?+

The system has to change along with the business — which is why we support and develop the environment after handover: refinements, new integrations and new automations as you grow. A system without support goes stale in a year or eighteen months.