Process Mapping: What to Do Before You Start Automating
Automation does not speed up an unmapped process; it scales its mess. The six steps of process mapping and the five data points to measure.
In short: process mapping means writing out the work to be automated step by step and measuring each step for duration, owner, error rate and how far it can be delegated. Automating an unmapped process does little more than make the existing mess run faster. According to McKinsey's 2025 survey, only 21% of organizations using generative AI have redesigned a single workflow, and that is precisely the change most strongly correlated with impact on earnings.
Why automation without a map does not work
Automating a process means speeding it up as it stands. If the process contains an unnecessary approval step, that step becomes faster after automation and remains just as unnecessary. If the same fact is keyed into three separate places, automation speeds up all three entries and the inconsistency survives intact.
McKinsey's data shows the distinction: 88% of organizations use AI in at least one function, but only 39% can demonstrate an impact on earnings. The variable that closes the gap is whether the workflow was redesigned. And redesign begins with a map.
Process mapping in six steps
1. Draw the boundary. Write in one sentence where the process starts and where it ends — "starts when a customer request arrives by email, ends when the quote is sent." A mapping exercise with a vague boundary never finishes.
2. Write the steps as they are. Not what should happen, but what actually happens today. In most organizations these two differ, and the difference is the improvement opportunity itself. Draw the steps out alongside the person doing the work, at their desk.
3. Measure five data points per step.
| Data point | Why it is needed |
|---|---|
| Duration | The baseline for any savings calculation |
| Owner | Who has to be asked about a change |
| Input and output | Which system carries which data |
| Error rate | The size of the rework |
| Waiting time | Visible when the step itself is short but the handover is long |
Without these five, the choice of which step is worth automating is guesswork.
4. Sort the steps into three groups. Those that can be delegated to AI, those that must stay under human approval, and those that will not be touched at all. The third group is usually the largest, and accepting that speeds the project up.
5. Remove the unnecessary steps. This is the most valuable work you can do before automating, and it costs nothing. Every step removed is code that never has to be written.
6. Write the new flow and define the measurement. Once the changed flow is drawn, put in writing which metric will be measured on days 30, 60 and 90. A redesign without a defined measurement is a redesign whose result cannot be argued.
The three findings mapping surfaces most often
Shadow processes. Steps that are absent from the official flow but that everyone uses — usually a spreadsheet somebody keeps privately, or a group chat. If they never reach the map, the system runs incomplete once automation goes live.
Approval pile-ups. Large amounts of work waiting on one person's approval. The step is short but the wait is long, so it stays invisible in a duration measurement and only appears once waiting time is measured.
The same data entered many times. The same fact keyed by hand into different systems. This is where automation pays back fastest, and usually the easiest thing to fix.
How long mapping takes
Two weeks is realistic for one process in one department. The time splits roughly as: a week of observation and interviews, a few days of measurement, and a few days to write and validate the flow.
Mapping exercises that run longer usually suffer from a boundary that was not drawn tightly enough. Rather than mapping every process in a department at once, starting with the highest-volume process delivers a result faster.
Frequently asked questions
Who should do the process mapping? The person doing the work describes it, the process owner validates it, the consultant sets up the method and writes it down. Maps drawn from a manager's account alone do not reflect the real flow.
Do we need software for it? No. A spreadsheet and a simple flow diagram are enough for most processes. Tool choice is not what determines the quality of a map.
Can mapping conclude that no automation is needed? Yes, and that is a good outcome. In some processes, removing unnecessary steps delivers enough of a gain on its own. A gain achieved without writing software is the cheapest gain there is.
How often should the map be updated? Whenever the process changes, and at least once a year. A map that is not updated becomes misleading within months.
Sources
- McKinsey & Company, The State of AI: Agents, innovation, and transformation (2025) — 88% of organizations use AI in at least one function, only 39% can show an impact on earnings; only 21% of generative AI users have redesigned a workflow.
- Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 (June 2025) — unclear business value is named among the reasons for cancellation.
Closing
Process mapping is the cheapest and most decisive step in an automation project. Two weeks of mapping prevents months of building the wrong thing.
See our process automation service or look at the enterprise AI consulting side.
Let's find out where your organization stands
In a free 30-minute discovery call, we'll talk through your current state and the right first move.
Book a Call