Skip to content
Condictor Studio
Automation
Automation

Business process automation - where to start

A seven-step checklist for choosing the first process to automate, measuring it before change, and identifying processes not to touch first.

About 2 min readby
A black process track with a central mint automation loop and coral starting point, surrounded by dark manual stations

A good first automation is one process performed often, describable in practice, and measurable before change. The most painful task is not always the best first move: it may have too many exceptions or an expensive error.

1. Find recurring work

For one week, let the team note recurring activities and time spent. Compare observations with available system data. Typical candidates are copying data between systems, entering invoices, answering repeat questions, assembling reports, and checking that nothing was missed.

2. Choose by three criteria

CriterionWhy it matters
Frequencysavings multiply with repetition
Repeatable rulesstable rules are cheaper and easier to predict
Measurabilitywithout a baseline, you cannot know whether the change helped

3. Measure the before state

At minimum, record weekly frequency, minutes per execution, and the rate of errors requiring repair. A baseline distinguishes real improvement from the impression that a new tool helped.

4. Describe reality, including exceptions

Document the process as people actually perform it, not only as the procedure says it should happen. Ask “what do you do when the data does not match?” The answers expose the true process and the exception path.

5. Separate rules from AI

Use regular automation when data is structured and decisions can be expressed as “if A, then B.” Add an AI layer for unstructured text or document interpretation. Use an AI agent only when later steps cannot be fixed in advance. Do not add a model where a rule suffices.

6. Test before autonomous production action

Build a version that works beside the process, test it against historical data, use an approval mode, and only then automate stable, low-risk cases. Each stage needs its own acceptance test and a way to reverse the change.

Four consecutive stages: a version running beside the process, a run on historical data, human approval mode, and full automation
Delivery sequence: from working beside the process to full automation.

7. Add monitoring and ownership

An automation can silently fail when a file format, access token, or provider interface changes. At minimum, use an error alert, execution log, periodic completeness check, and a named process owner.

Do not start with a process that needs redesign, has many exceptions, cannot be described, or is one of many simultaneous changes. Compare annual benefit with implementation, licence, supervision, and exception-handling costs first.

Indicative August 2026 ranges: 2,000–8,000 PLN net for one process, 8,000–25,000 PLN net for several automations or document flow, and 1,500–5,000 PLN net/month for continuous improvement. Start with a First Automated Process or an AI audit.

Maciej Szukalski

Author

Maciej Szukalski

Founder of Condictor · systems architect · research and development

He has designed and built digital products since 2014. He specialises in architecture, research, and applications with automation and intelligence layers.

Have a problem to solve?

Let’s find the right first step

Describe your situation in a few sentences. We’ll return with questions or a concrete proposal for what comes next.

Describe your topic

See also

All articles