---
title: "What to automate first: a 4-question test for SMEs"
description: "A simple scoring method to pick the first AI automation project: frequency, clarity, cost of the manual work and cost of a mistake."
canonical: https://aikkvnak.hu/en/blog/what-to-automate-first
language: en
publisher: AI KKV-nak
published: 2026-09-19
updated: 2026-09-19
---

# What to automate first: a 4-question test for SMEs

A simple scoring method to pick the first AI automation project: frequency, clarity, cost of the manual work and cost of a mistake.

## Key takeaways

- Pick a process that runs at least 20 times a week with a clear input and output.
- Score candidates on frequency, clarity, manual cost and error cost. The highest total wins.
- The first project should pay for itself within three months; that is the bar, not novelty.

## Why the first project matters more than the tenth

The first automation sets the team's expectations. If it saves ten hours a week and never breaks, the second one is easy to approve. If it is a clever demo that nobody uses, the topic is closed for a year. So the first project should be boring, frequent and measurable.

## The four questions

Score each candidate process from 1 to 5 on each question and add them up.

1. Frequency: how many times a week does it run? 1 is monthly, 5 is dozens of times a day.
2. Clarity: could you write the steps on one page? 1 is 'it depends', 5 is 'anyone could follow it'.
3. Manual cost: hours per week times the hourly cost of the people doing it.
4. Error cost: what happens when it is done late or wrong? 1 is nothing, 5 is a lost customer.

## Typical winners

In our scoping calls the same four processes score highest again and again: missed inbound calls, invoice data entry, first reply to web leads, and support questions that have a documented answer. Each runs daily, is easy to describe and has a visible cost.

## What to measure before you start

Write down the baseline for one week: how many times the process ran, how long it took, how many mistakes. Without it you cannot prove the automation worked, and proof is what unlocks the next project.

## FAQ

**Should the first project be the biggest pain point?**

Only if it also scores high on frequency and clarity. A painful but vague process is a bad first project; fix an easy, frequent one first and return to the painful one with credibility.

**How much should a first AI automation cost?**

A fixed price that pays back within roughly three months from the measured manual cost. If the numbers do not support that, the scope is wrong, not the price.