Automation· June 10, 2026

Routine automation: where to start so the first step pays for itself

Start automating routine work with a single area — the one where manual labor costs the most in money. According to Deloitte's global survey, automating routine operations pays for itself in under a year on average, and a targeted first area pays back faster. Below is the method: how to find that area, price the routine, and show the savings within weeks.

This piece is a close-up of the very first step. The map of the whole journey — choosing between off-the-shelf software and custom development, where AI fits, the requirements of Federal Law 152-FZ, Russia's personal data protection law — is collected in the hub business automation: where to start.

Why "automate everything" is the worst possible first plan

The most ambitious automation projects are the ones that fail most often. Trying to cover every process at once stretches the work over a year: the team burns out, the budget drains, and not a single area is finished. The reverse order is safer — one area in weeks, savings you can measure, and only on that foundation the next step. Then the project pays for itself as it goes, and each result adds credibility to the next one.

"Automate everything" in practice hides a dozen processes with different owners, systems and exceptions. Each one drags along its own integrations and sign-offs, the timelines stack up, and the benefit is postponed until a finish line that often never arrives. A single area breaks that logic: the boundaries are narrow, the result is visible in weeks, and the savings can be measured.

Which area should you start with?

Walk through your processes with three questions: where is the same action repeated, where do mistakes happen most often, and where do people wait the longest. The area that scores three yeses is your main candidate. Usually a day of watching the real work is enough to find it: sit next to the people doing it by hand and write down where the hours go.

In most companies the candidates for the first step are the same:

Reconciliations: bank against the books, warehouse against sales, settlements with counterparties.
Moving data between systems by hand: from email into a spreadsheet, from the spreadsheet into 1C, the accounting and ERP platform most Russian companies run on.
Recurring reports that are assembled from scratch every time out of the same sources.
Standard documents — invoices, contracts, acceptance certificates — filled in by copy-paste.
Notifications and reminders that depend on one specific person remembering them.

Routine like this exists in almost any company: the McKinsey Global Institute estimates that in roughly 60% of occupations at least 30% of work activities can technically be automated with tools that already exist. A simple frame helps you test a specific candidate:

Criterion	Good first area	Poor first area
Frequency	daily or weekly	a few times a year
Rules	clear, "if — then"	every case is decided on its own terms
Process maturity	already runs manually without argument	the team is still arguing about how it should work
Boundaries	one department, one or two systems	half the company and five systems
Effect	measurable in hours and money	"it will be more convenient"

What does your routine cost in money?

The price of routine is calculated like this: frequency of the operation × time per operation × employee rate + the cost of errors. A reconciliation that takes an accountant two days a month is 24 working days a year — more than a month of one specialist's time. Add the errors: a discrepancy found a month later means hours of investigation, and sometimes direct losses. What you get is the sum the business pays for that area every month.

The scale is usually underestimated. In an hh.ru survey (May 2025, 3,283 working respondents), 25% of employees spend more than six hours a week preparing reports that aren't automated, and 38% spend the same amount approving documents and requests. The international picture is similar: in Asana's "Anatomy of Work" study, 58% of office workers' time goes to "work about work" — forwarding status updates, hunting for information, switching between systems.

Once the price of the area is calculated, the conversation about automation turns into an ordinary investment decision with a payback period. That same calculation is the first step in every custom development project we take on: before architecture and estimates, we establish what the routine costs and what target in money the project has to hit.

When will the first step pay for itself?

In months — if the area was chosen by the price of its routine. An industry benchmark: in Deloitte's global survey, companies that deployed robotic process automation for routine operations reached payback in under a year on average. A targeted first area usually pays back faster than the "average project": the investment is smaller, and the savings start in the system's first month of operation.

Keep expectations sober, though: in that same survey, companies at the pilot stage expected payback in nine months, while those that actually deployed came in at around twelve. Even the conservative scenario means that after a year the savings are already net — provided the area was chosen by the money and the metric was fixed before the start.

The principles that carry a first project to the finish

Five rules save a first project: one area at a time, automate only a process that already works, keep a human in the loop for decisions, leave a trace of every action the system takes, and take the next step only after the savings are proven. The rules look dull, but they cut off the main causes of failure — scope creep and accelerated chaos.

One area at a time. A result in weeks instead of a "big system" in a year. Narrow boundaries mean a predictable timeline and a clear acceptance test.
Automate a process that already works. Chaos after automation becomes automated chaos: errors multiply at machine speed and people can no longer catch them in time.
A human stays in the loop wherever decisions are needed. The system prepares data and handles the routine; the person signs off on exceptions and disputed cases.
Every step leaves a trace. Logs and notifications show what the system did and why — trust in it grows on facts.
The next area comes after proven savings. The numbers from the first step fund and justify the second.

Can a quick utility close the routine?

Yes — if the routine lives inside the team and doesn't touch anyone else's data. A file converter, a price-list parser, a generator for a standard report: utilities like these can now be put together in an evening — you describe the task to an AI tool in plain words and it writes the code. The approach is called vibe coding, and for internal tools where the cost of an error is low it's an honest first step into automation.

A breakdown of the method — what you can really build in an evening and where it falls apart — is in the article what vibe coding is. The line runs along the cost of an error: as soon as a utility touches customer data, payments or integrations with 1C and the bank, it needs engineering work — with tests, access rights and contractual accountability.

A separate case is bookkeeping that has grown inside spreadsheets to the point where "only Marina understands that file." The signs that the transition zone has ended are in our piece on when it's time for a business to leave Excel.

Do you need AI to remove routine work?

Not everywhere — and that's fine. Classic automation — integrations, "if — then" rules, schedules — closes most routine work more cheaply and more predictably, with no model training and no surprise answers. AI belongs where the input is unstructured: parsing an incoming email, extracting line items from a PDF invoice, classifying a support request, answering a customer's open-ended question.

The practical takeaway is simple: first the area and the price of the routine, then the tool. Reconciliations and data transfers are almost always solved by connecting systems — we covered how to bring 1C, CRM and the bank into a single loop in our article on integrations. For standard areas there is a catalogue of ready-made solutions: a fast start without building from scratch, adapted to your process.

How do you know the automation has paid for itself?

The metric is fixed before the start: hours spent on the process, number of errors, how long a request takes to get through. After launch the same metric is taken again and compared with the baseline. Without a "before" measurement the effect can't be proven — the project risks being declared a failure even where it works. The target in money or hours is formulated back when the area is being chosen.

The order of the check:

Before the start, record the baseline: how many hours the process takes, how many errors a month, how long the customer waits.
For a while, run the old process in parallel with the new one — the numbers get cross-checked and trust in the system grows.
After a month or two, compare the metric with the baseline and calculate the actual payback.
Show the savings to the team and to management: these are the arguments for the next area.

Surveys give you benchmarks for comparison: according to SberAnalytics and Sber Business Soft (November 2025, 559 respondents), 45% of companies that deployed automation report faster handling of standard tasks, 37% a lower share of manual operations, and 36% fewer errors in documents. Tracking the effect is easier on numbers that update themselves — which indicators to display is covered in our piece on the owner's dashboard.

What comes after the first area?

Choose the next one — by the same routine-pricing formula. The first proven area changes the standing of automation inside the company: now there are measured savings, team trust, and a queue of candidates. After that the areas start connecting to each other, and targeted solutions gradually add up to a single loop — accounting, sales and analytics stop living as separate islands.

At this point it's worth returning to the map of the whole journey — from choosing between off-the-shelf software, a low-code builder and custom development, to the plan for the first 90 days. It's collected in the hub business automation: where to start.

If you'd like to take the first step with a team that does this regularly — tell us about your task: we'll look at your processes, price the routine, and propose an area with a target in money and a payback period. The decision stays yours — and the numbers stay with you either way.

Sources

Frequently asked questions

Our processes live in people's heads and aren't documented — can they still be automated?+

Yes, mapping the processes is part of the project. Before anything is automated, the real order of work is recorded from the people who do it by hand — usually a day or two of observation. What gets automated is the working process as it actually exists in practice, with the gaps closed along the way.

Does automating routine work mean layoffs?+

In practice, almost never. The freed-up hours go to work nobody had time for before: customers, growth, oversight. What disappears is overtime and mistakes, not people.

How do I know the automation has paid for itself?+

Measure the area before you start — hours spent on the process, number of errors, turnaround time — and measure again after launch. The difference converts into money and is compared with the cost of the solution. The target in hours or money is fixed while you are still choosing the area — then the payback can be proven with numbers.

Which process do companies most often automate first?+

Reconciliations and moving data between systems: bank against the books, warehouse against sales, email into a spreadsheet and into 1C. These are the most common and most expensive manual areas, and their rules are clear, so they pay back faster than anything else.

How long does it take to automate one area?+

Weeks. A targeted solution — automated reconciliation, a recurring report, document generation — goes live in a few weeks and runs alongside the old process at first, so the numbers can be checked against each other. Only full ERP-scale systems — accounting, sales and analytics together — take months.