Business Process Automation: What to Document First — AI Automation article by Rui Codex

Before you start business process automation, document each step as it actually runs, including inputs, outputs, systems, exceptions and decision owners, then involve the people who do the work and who own the data. This current-state map exposes the rules and handoffs any automation must respect.

Key takeaways

  • Map the process as it actually runs today, not as the procedure manual says it should run.
  • Record the trigger, inputs, outputs, systems, decision rules and exceptions for every step.
  • Involve the people who perform the work, one accountable process owner and the owners of the data.
  • Treat gaps found during mapping as signals to fix the process before any software is built.

Why map a process before business process automation?

Mapping a process before automation means documenting it as it runs today, so the automation reproduces the real work rather than the intended one. Done properly, it produces a current-state map, not a wish list. Start with a process that has a clear beginning and end, such as order-to-invoice or supplier onboarding.

The map serves two purposes. It shows what the automation must reproduce, including the awkward steps people handle without thinking. It also shows which steps depend on judgement, which depend on a rule and which depend on a system nobody has documented. Those distinctions decide how much of the process software can take over.

Mapping is also where many projects change direction. A process that looks like three steps on a whiteboard often has eleven branches once someone traces a real month of work. That discovery is cheap at this stage and expensive after an integration is already live.

What should you document for each process?

Document the trigger, inputs, outputs, systems, decision rules, exceptions and timing for every step. These are the minimum details a process needs before software can execute it. Add volumes where you can measure them.

  • Trigger: the event that starts the step, such as an email arriving or a status change in a sales tool.
  • Inputs: every document, field and data source the step reads, with the format it arrives in.
  • Outputs: what the step produces and where it is stored or sent.
  • Systems: each application touched, including spreadsheets and shared mailboxes.
  • Decision rules: the conditions that send work down one branch or another, written as they are applied today.
  • Exceptions: cases that break the normal path, how often they occur and who resolves them.
  • Timing and handoffs: waiting times, deadlines and the moment ownership changes hands.

Be careful with spreadsheets and shared mailboxes. They often hold the real logic of a process, such as a column of manual overrides or a rule one colleague applies from memory. Record them as systems and note who maintains them.

Tip: Ask for one recent real case and map it from start to finish before mapping the general version. Real cases expose the exceptions that a procedure document leaves out.

Who should be involved in the mapping?

Involve the people who perform the work, one accountable process owner, and anyone who owns the data or systems the process touches. Leaving out any one of these groups usually means rework later, because each group sees a different part of the process.

  • Process operators, who know the workarounds and the cases that never reach a manager.
  • The process owner, who settles disputes about the intended rule and approves the target state.
  • IT and system owners, who know the data structures, access rights and integration limits.
  • Data protection or compliance contacts, where the process handles personal data, so retention and access rules are captured early.
  • Managers of adjacent processes, because a handoff from one team is another team's input.

Keep the group small enough to meet regularly. Two or three working sessions with the operators and the owner usually resolve more than a large steering committee does. Ask operators to walk through the work on their own screens rather than describe it from memory.

How do you map the process step by step?

Map in four passes: observe, draw, validate and mark. Each pass has a clear output, and you should not move on until the previous one is agreed.

Pass one: observe the work

Watch two or three real cases run from start to finish, and capture what happens on screen and in conversation. Note every point where someone checks something, waits or copies data between systems. Observation catches the steps people no longer consider steps.

Pass two: draw the current state

Draw a swimlane diagram with one lane per role or system. Keep the notation simple so operators can read it. A diagram only a specialist understands will not be corrected by the people who know the process best.

Pass three: validate with the operators

Walk the diagram through with the people who do the work and ask them to find the step that is missing or wrong. Confirm volumes and the share of cases that take each branch, using system logs or counts where they exist. Where data does not exist, record the estimate and label it as an estimate.

Pass four: mark the automation candidates

Tag each step as rule-based, data transfer, judgement or unclear. Rule-based and data-transfer steps are usually the first candidates. Judgement steps need a human decision point or a reviewed recommendation, and the AI automation service page describes how such steps are designed.

Where does process mapping fail?

Mapping fails when it records the intended process instead of the actual one, when it stops at the happy path, or when nobody owns the result. Each failure has a recognisable symptom and a practical fix.

  • Documenting the ideal: the map matches the procedure manual, and the first live test breaks on a case nobody wrote down. Fix it by mapping recent real cases.
  • Stopping at the happy path: exceptions are treated as rare, yet they consume most of the effort. Fix it by counting exceptions over a representative period.
  • Mapping too deep: every field and click is recorded, and the project stalls. Fix it by stopping at the level where a decision or handoff occurs.
  • No owner: the map sits in a shared folder and changes are never agreed. Fix it by naming one owner who signs off the target state.

The trade-off is between completeness and speed. A shallow map can be built quickly but will need revision once integration starts. A deeper map takes longer and reduces rework. For most processes, a map covering the main path, the most common exceptions and every handoff is enough to begin.

Hypothetical examples of process mapping

Here are two hypothetical scenarios:

Scenario one: a distribution business receives customer orders by email and retypes them into its order system. Operators describe the work as simple, but mapping reveals three order formats, a manual price check for one customer group and a spreadsheet that holds the only list of approved substitute products. The team documents these before choosing an integration approach, and the spreadsheet becomes a named system with an owner.

Scenario two: a services firm wants to automate client onboarding. Interviews suggest a single form and one approval. Observing two onboarding cases shows that the approval depends on a risk check done by another department through a separate email thread. The mapping session brings that department in, and the team decides to automate data collection first and keep the risk decision with a person.

Neither scenario describes a real project, and neither carries results. They illustrate the kind of finding mapping tends to surface: hidden formats, undocumented rules and handoffs to other teams.

How do you choose which process to automate first?

Choose a process whose map is stable, whose volume justifies the effort and whose failure is recoverable. A stable map means operators agree on the steps and the exceptions are known. Strong candidates usually have clear inputs, clear outputs and a named owner.

Avoid starting with a process that is still changing, spans several departments without an owner, or carries serious consequences if an error goes unnoticed. Those processes are better redesigned first. Where the map shows that two systems must exchange data, the system integration service covers that layer, and a process that needs a purpose-built tool fits the custom software development service. The article on how Belgian companies use AI in business processes discusses further process types.

Conclusion

Mapping a business process before automating it is the cheapest point to find the exceptions, owners and handoffs that decide whether automation works. Document the trigger, inputs, outputs, systems, decision rules and exceptions for each step, and involve the operators, the process owner and the data owners before choosing a tool. Rui Codex is independent and offers technical implementation support for AI automation, system integration and custom software across Belgium, the Netherlands and Luxembourg. To review a process map with an engineering team, request a project consultation at the Rui Codex contact page or email info@ruicodex.com.

Frequently Asked Questions

What is the difference between process mapping and process documentation?

Process mapping is the visual, step-by-step picture of how work flows between people and systems. Process documentation adds the detail behind each step, such as rules, data formats, exceptions and owners. Automation projects need both, because the map shows the flow and the documentation shows what each step requires to run.

Do I need to map every exception before automating a process?

No, you need to map the exceptions that occur often enough or cost enough to matter. Rare cases can be handled by a defined escalation path to a named person. Count exceptions over a representative period, then decide which ones the automation must handle and which ones stay with people.

Who should approve the target state of an automated process?

The process owner should approve the target state, because that person is accountable for the outcome and can settle disagreements about the intended rules. Operators, system owners and data owners should review it before approval. Written sign-off prevents later disputes about what the automation was meant to do.

Should process mapping include the personal data a process handles?

Yes, mapping should record which personal data a process reads, stores and sends, and where it goes. Include retention periods, access rights and any third parties involved. A data protection or compliance contact in the mapping session helps capture these points early, although Rui Codex provides technical implementation support rather than legal or compliance advice.

Can a process map be reused after automation starts?

Yes, a current-state map becomes a reference for testing, troubleshooting and future changes. Keep it under version control with an owner and a review date. When the process changes, update the map first so the automation and the documentation stay aligned.

Tags: Business Process Automation Process Mapping Workflow Automation Process Documentation Digital Transformation AI Automation Operations Management Stakeholder Involvement

Need Help Implementing This?

Our team can help you put these insights into practice. From AI automation to custom software development, we build solutions that deliver real results.

Send a project request