Skip to content

guide

Mapping a process before you automate it

Irfan4 min read

How do you map a business process before automating it?

Map a process by following one real instance end to end, recording every trigger, decision, handoff and exception as it actually happens rather than as the documentation describes it. Capture who does each step, what information they need, what they do when the information is missing, and where the process waits. The exceptions are the deliverable — they are what breaks automations, and they are invisible in any description written from memory.

Most automations that fail in production do not fail on the happy path. They fail on the third Tuesday, when a record arrives without a field somebody assumed was always present, and nobody finds out for a week.

Those cases are entirely discoverable in advance. They just are not discoverable by asking someone to describe the process, because people describe processes as they are supposed to work. The exceptions live in muscle memory, and they only surface when you follow a real instance through.

What a map has to contain

Not a flowchart. A record of what actually happens.

ElementThe question it answers
TriggerWhat starts this, exactly? A time, an event, a person noticing something?
StepsWhat happens, in order, with who does it
DecisionsEvery point where the path forks, and the rule that decides
InputsWhat information each step needs, and where it comes from
HandoffsEvery point where it moves between people or systems
WaitingWhere it sits, and for how long
ExceptionsWhat happens when the expected thing does not happen
OutcomeHow you know it finished, and who confirms

The last two carry most of the value. The first six are usually already understood.

Follow one real instance

The single most useful thing you can do is take one live case and follow it from trigger to completion, recording what happens rather than what should.

Three things reliably surface that no interview produces:

  • Undocumented steps. Someone checks something before proceeding. It is not in any process description because it is not officially part of the process. Automate without it and you have removed a check nobody knew existed
  • Actual waiting time. Two hours of work spread over four days. The four days are the customer's experience, and they are the thing worth fixing
  • Silent workarounds. A spreadsheet that reconciles two systems. A message someone sends to unblock things. These exist because the official process does not work, and automating the official process leaves the workaround stranded

Recording decisions properly

Every decision point needs a rule that can be written down. Where one cannot be written, that is the finding.

What you hearWhat it meansWhat to do
"If X then Y, otherwise Z"A ruleRecord it and move on
"Usually Y, but sometimes Z"An unstated conditionKeep asking until it surfaces
"It depends"Judgement, not procedureRecord who decides and what they weigh
"Ask Sarah"Sarah holds undocumented rulesInterview Sarah specifically
"We just know"Tacit knowledgeFollow three instances and infer the pattern

"It depends, ask Sarah" is not a failure of the mapping. It is the most valuable thing you will find. It tells you exactly where the process depends on a person, and therefore exactly where automation should stop and hand over rather than attempt to continue.

Mapping frequently shows that the process itself is the problem rather than the manual effort, which is why

the process redesign that usually comes before automation.

Exceptions are the deliverable

Treat the exception list as the primary output, not an appendix.

For each step, ask:

  • What if the input is missing?
  • What if it is malformed or in the wrong format?
  • What if the person responsible is unavailable?
  • What if the external system is down?
  • What if this has already been done?
  • What if it needs to be undone?

That last one — reversal — is the most commonly forgotten and the most expensive. An automation that creates records, sends notifications and updates three systems needs a defined answer for what happens when it ran on the wrong record. If nobody has thought about it, the answer is discovered under pressure.

Separate waiting from working

Record these as different things, because they have different fixes.

Working time is when someone is actively doing something. Automation reduces it by doing the work faster.

Waiting time is when nothing is happening — awaiting approval, awaiting a reply, awaiting a batch job. Automation reduces it by removing the wait entirely, which is usually the larger win.

A process with twenty minutes of work spread across three days is a waiting problem. Automating the twenty minutes saves twenty minutes. Removing the waiting saves three days, and the customer only ever noticed the three days.

What the map tells you

When it is finished, it answers the automation question directly:

FindingImplication
Every decision has a written ruleGenuinely automatable
Some decisions are judgementAutomate around them — hand off, do not attempt
Steps exist that nobody can justifyRedesign first. The automation may become unnecessary
Most elapsed time is waitingThe win is removing waits, not speeding up work
Exceptions outnumber the happy pathNot stable enough. Fix the upstream cause
The process differs per personStandardise before automating, or you encode one person's version

The fifth row is a stop signal. If exceptions outnumber the main path, the process is not a process yet — it is a set of related judgements, and automating it produces a system that escalates most of its cases.

A workable format

Formal notation is rarely worth it. This is enough for almost every business process and, crucially, people will actually read it:

PROCESS: [name]
OWNER:   [who is accountable]
RUNS:    [frequency]

TRIGGER
  [what starts it]

STEPS
  1. [action] — [who] — [inputs needed]
     Decision: [if applicable, with the rule]
     Exception: [what happens when it goes wrong]
     Waits: [duration, if it waits here]
  2. …

EXCEPTIONS (consolidated)
  [every failure mode found, with current handling]

JUDGEMENT POINTS
  [decisions that are not rules, and who makes them]

OUTCOME
  [how you know it finished]
  [who confirms]

OBSERVED
  Instances followed: [how many]
  Working time: [x]
  Elapsed time: [y]
  [the gap between x and y is where the opportunity is]

The last block matters. Working time versus elapsed time is the number that justifies the project, and it is measurable rather than estimated.

What to do with it afterwards

Keep the map. It is more durable than any automation built from it.

Platforms change and workflows get rebuilt, but the process understanding transfers directly — as noted when comparing automation platforms, teams whose process knowledge exists only inside a tool are far more locked in than the pricing page suggests.

The map is also what makes the next conversation possible: whether this should be automated at all, which is a different question and worth asking before anyone opens a platform.

Want this handled properly?

You now know what the work involves. If you would rather not do it yourself, that is what we do.

Key takeaways

  • Follow a real instance. Processes described from memory omit the exceptions that break automations.
  • The exceptions are the deliverable, not an appendix to it.
  • Record waiting time separately from working time — most delay is waiting, and automation fixes waiting best.
  • "It depends, ask someone" means judgement, not rules. Note who, and what they consider.
  • Map before choosing a platform. Choosing first makes you design the process around the tool.
FAQ

Questions about this

For a single process, hours rather than weeks — usually one or two sessions with the people who run it, plus following a live instance. If mapping takes longer than that, the scope is probably several processes wearing one name, and splitting them is the first finding.

Related services

More on AI Automation

TopicsProcess mappingWorkflow automationAutomation
Chat with us