Business process automation: what to automate first and how to count the payoff

Every company has a few jobs that someone does by hand every day: copying an enquiry from WhatsApp into the CRM, issuing an invoice, chasing a payment, updating stock on Kaspi, building the Monday report. None of them is hard. Together they eat hours a week from people who were hired to do something else.

Business process automation (BPA) means handing those repeatable steps to software: the systems you already use pass data to each other and trigger the next step on their own, and a person steps in only where a decision is needed. The short answer to "where to start": automate the process that happens most often, follows clear rules and costs the most time today, usually incoming requests, invoices and data sync between systems. Count its cost before you start, automate one process, measure, then move to the next.

This guide covers which processes to automate first, the three ways to do it, how to calculate the payoff and a step-by-step plan.

What business process automation is (and is not)

A business process is a chain of steps with a trigger and a result: "an enquiry arrives → it is logged → someone is assigned → the customer gets a confirmation". Automation takes over the steps that follow fixed rules.

It is not:

• A new all-in-one system. In most cases the CRM, 1C and the website stay; what changes is that they talk to each other.
• AI by default. Most steps follow simple rules ("if paid, send the receipt"). AI is useful only where a step needs judgement, such as reading a free-text message or an invoice photo.
• A one-off project. Systems and rules change, so automations need an owner.

Where manual work costs the most

Incoming requests

A request arrives in WhatsApp, Instagram or by email, and the manager copies it into the CRM, or forgets to, because they were busy. The request is lost. With automation, a message or a form on the site creates a card in amoCRM or Bitrix24, assigns an owner and sends the customer a confirmation. Nothing waits for someone to have a free minute.

Invoices and documents

Issue an invoice, send it, check the payment, remind, prepare the closing documents: every step needs a person. Automation issues the invoice when a deal moves to "won", matches payments against the bank statement, sends a reminder after a set number of days and prepares the documents for closed deals in the accounting system. In Kazakhstan, VAT invoices are issued in the ESF e-invoicing system: standard 1C configurations for Kazakhstan already exchange data with it, and from a CRM or your own system it is done through the ESF API.

Data sync between systems

CRM, 1C, the website, MoySklad, a Kaspi store: each keeps its own records. Someone makes sure stock on Kaspi matches the warehouse and that Kaspi orders reach 1C, usually by hand, several times a day. An integration does it on every change: orders arrive through the Kaspi seller API, prices and stock go back. Our page on 1C integration shows how such a link works.

Reporting

The weekly sales, pipeline, stock or receivables report: someone collects numbers from several places into one Excel sheet. A dashboard that reads the same sources shows the numbers when you open it.

Notifications and reminders

Appointment reminders, order status updates, an expiring contract, a customer who has not ordered in three months. Set up once, sent on time to every customer, not only to those someone remembered. In Kazakhstan, customers mostly expect these messages in WhatsApp.

Rules or AI: which steps need a model

Most steps in the list above need no AI at all. A useful test for each step:

Use rules where they are clear and a model only where free text or documents have to be understood. That keeps the automation cheap, predictable and easy to check. When a process does need AI steps, that is AI workflow automation, and we explain the difference between a chatbot and an agent in what an AI agent is. When a single step needs a model of its own, such as reading supplier invoices, a custom AI agent plugged into the workflow can handle it.

Three ways to automate

Start with what is built in. Move to a no-code platform when a process crosses tools. Go custom when the chain touches 1C or your own systems, has many exceptions, or the no-code bill and maintenance start to exceed what a proper integration would cost. Before choosing a platform, check whether it has a ready connector for your 1C, Kaspi and MoySklad: if not, the link will have to be written anyway. Often the answer is a mix: we build on platforms such as n8n where they fit and write code only where a platform would be the more expensive route.

For linking several systems into one flow, see how to unite scattered services in one system.

How to calculate the payoff before you automate

Do the maths first, otherwise you cannot choose what to automate or tell whether it worked.

1. Pick one process. Count how many times a week it happens and how many minutes it takes each time.
2. Multiply by the hourly cost of the person who does it (salary, taxes and overheads divided by working hours).
3. Add the cost of errors: a lost enquiry, a late invoice, a wrong stock level on Kaspi.
4. Compare with the cost of automating it plus running costs.

An illustrative example, your numbers will differ. A manager spends 1.5 hours a day copying orders from Kaspi into 1C. Over 21 working days that is about 31 hours a month, almost four full working days. Multiply those hours by the manager's hourly cost and compare with the one-off price of the link (a standard link between two systems starts from 270 000 ₸) plus hosting. You will see how many months it takes to pay back, and the manager's time goes back to customers.

Two things the formula does not show but you should count: speed (a request answered in a minute, not tomorrow) and fewer errors from retyping.

A step-by-step plan

Step 1: list the manual work

Ask each person what they do every day that feels like copying. The expensive processes are often the ones everyone stopped noticing.

Step 2: score and choose one

For each process note frequency, time per run, cost of errors and how clear the rules are. Start with the one that is frequent, rule-based and expensive, not the most impressive one.

Step 3: fix the process on paper

Write the steps, the trigger, the result and every exception ("what if the payment fails?"). If people disagree on how the process works, settle that before any software.

Step 4: connect, don't replace

Link the systems you already have. Choose one source of truth for each type of data: stock in the warehouse system, customers in the CRM, money in 1C.

Step 5: plan the exceptions

A good automation knows what to do when something goes wrong: the product ran out, the payment failed, the address is invalid. It handles the case or notifies the right person, instead of stopping silently.

Step 6: measure and move on

Compare time, errors and speed with your baseline after a month. Then take the next process from the list.

Mistakes to avoid

• Automating a broken process. Automation makes a bad process run faster, not better. Fix the logic first.
• Starting with the biggest process. It takes months and the team loses faith. One small win first.
• No owner. An integration nobody watches breaks the day a system updates its API.
• Silent failures. Every automation needs alerts when it fails, or you find out from an angry customer.
• Too many tools. Five no-code chains built by five people are a system nobody understands. Keep a list of what runs where.

Business process automation checklist

• List of repetitive manual tasks from every team member
• Frequency, time and error cost for each
• First process chosen: frequent, rule-based, expensive
• Steps, trigger, result and exceptions written down
• Source of truth defined for each type of data
• Route chosen: built-in, no-code or custom
• Baseline measured: time, errors, response speed
• Alerts for failures and a named owner
• Date for the review after one month

When to bring in a team

If the process lives inside one tool, use its built-in automations: you do not need us. When it crosses several systems, involves documents or free text, or the no-code setup has become fragile, a fixed-scope project is usually cheaper than another year of workarounds. We build process automation from 270 000 ₸: the price is fixed in the contract, the first workflow goes live within a few weeks, the work carries a six-month warranty, and the code and access stay with you. What drives the price is covered on our automation pricing page.