Writing a process flow description: trigger, steps, variants and handoffs before someone else takes over
A process flow description is quick to write as long as you run the process yourself. A few notes are enough, because you carry the rest in your head. It gets difficult the moment someone else is supposed to take over the flow: a new colleague, a stand-in during vacation, or an AI employee. That's when you see what's missing. When does it start? What's different in this one special case? Whom do you ask when a document doesn't fit? In this article, we show how to describe a process flow so that it can be carried out, by someone who can't ask you. As an example, we use a real process from a tax advisory firm.
When a process flow is ready for handover
When you write a flow down for yourself, it reminds you of the order of things, and nothing more. It's ready for handover when someone else can carry it out without asking you. How to check that with a simple test is covered in How to write an SOP or work instruction that a person and an AI employee can follow.
If the flow goes to an AI employee, one more thing belongs in it: what happens in doubt. With us, an AI employee doesn't guess at a case it can't match with certainty. It hands the case to a person on your team through the escalation path, and the answer from the feedback session becomes a new rule in the flow.
For this to work, the flow needs three things:
- Clear edges. When does the flow start, and when is it finished?
- Every deviation is in it. Every case that's handled differently has its own path.
- A fixed path for uncertainty. Instead of guessing, whoever runs the flow knows whom to turn to.
The six parts of a process flow
We describe every flow in six parts. The first five are the five points that are also on our home page. The sixth is the escalation path.
- Trigger: what starts the flow, a fixed schedule or an event.
- Start: what exactly is on the table when the work begins.
- Steps per variant: what gets done, and how the cases differ.
- End: how you can tell the flow is finished.
- Handoff and report: who gets the result and what you learn at the end.
- Escalation path: what happens when something is missing or doesn't fit.
How to capture these parts in a process profile, our word for a process description, is covered in How to document business processes: the template for handing a process to an AI employee. Here, it's about the third part, the steps themselves, and about the places where flows break in practice.
Writing process steps that can be carried out
A process step is an action with a result. Three rules make the difference between a reminder and an instruction.
One step, one action. "Check and book the invoice" is two steps. If you write them together, you hide the question of what happens when the check fails.
The verb first, the result with it. "File the notice in the client's folder" says what to do and how you can tell it's done. "Filing" on its own doesn't.
No words only you understand. "As usual", "the usual stuff", "the folder from last time" are knowledge in your head. Spell out which folder, which field, which rule.
A comparison:
| Reminder | Step that can be carried out |
|---|---|
| Go through the mail | Open every new letter and read who it's from |
| Forward the important stuff | Send letters with a deadline to the person responsible |
| File the rest | File every remaining letter in the folder for its case |
The right-hand column is longer. In return, someone who can't ask you can carry it out.
Decision points as questions
At some points, the flow splits. That's where a question goes. Write it so that it can be answered with yes or no, and write down how things continue for both answers.
Poor: "Forward depending on the case." Better: "Does the letter have a deadline? Yes: send it to the person responsible. No: file it."
If you can't answer a question with yes or no, there are several cases behind it. Those cases then belong in the list of variants.
Variants in a table instead of in side clauses
Variants are the core of every flow. A variant is any case in which the person on your team does something differently: a different sender, a different layout, a different place to file, a different recipient. A different name or a different amount is not a variant; those are just different values.
When you write a flow down for yourself, variants quickly end up in side clauses: "File the invoice, unless it comes on paper, in which case type it up first, and for one particular supplier also check the order number." After three side clauses like that, nobody understands the step anymore.
So pull the variants out of the step and into a table, one row per variant, just like in the process profile. The step itself stays short, for example "File the invoice according to the rule for its variant".
| Variant | How do you recognize it? | How does it arrive? | What's different in the work? | Where is it filed, who hears about it? |
|---|---|---|---|---|
| … | … | … | … | … |
| … | … | … | … | … |
That keeps the table readable even when the number of variants grows.
And don't wait until the table is complete. The frequent variants have to be in it. For the rare ones you've overlooked, there's the escalation path.
What happens when something is missing
Every flow will sooner or later hit a case that hasn't come up before. A document is illegible. A piece of information is missing. A match isn't clear. For that moment, you need a fixed path, for people and for AI employees alike.
That's why every place where something can be missing gets the same exit: the case goes to a person on your team, and until then it waits and is listed as open in the report. The other cases keep going. In the feedback session afterwards, the AI employee is trained on the new variant, and it's saved as a rule.
In the flow, it looks like this: at every question whose answer can be unclear, there's a third exit next to yes and no, "unclear: escalation path". That sounds like more effort. In practice, it's the reason a flow can later be allowed to run on its own.
Handoffs: where process flows break
Many flows pass through several stations. One person takes something in, another works on it, a third files it. At every handoff, something can be left lying around, because nobody feels responsible.
So describe every handoff explicitly:
- What is handed over, and in what state?
- To whom, and who stands in?
- How does the recipient find out about it?
- How does the sender know it has arrived?
In Owner Autonomy, handoffs like these come in once your first process runs on Autopilot. Where a case passes through several stations, the AI employees hand it off to each other.
That only works if the handoff is described as precisely as every other step.
What this looks like for us
At a tax advisory firm, the first process was filing digital tax assessment notices. Up to 30 to 40 arrive there per day, and before, the office team needed about 2 minutes per notice. The assessment notice process distinguishes around 45 rules in total: many types of assessment, letters from the city with different layouts, different types of tax and types of letter.
The basis for the flow was what Kevin saw while watching the office team: which types of assessment come in and how each one gets filed. As a numbered flow, the process looks like this. The steps are our simplified description; the exact rules per type of assessment aren't listed here.
The frame is short: every day, the digital notices are retrieved, and the new notices from that retrieval are the start. The complete process profile with end, report and escalation path is in the example in How to document business processes: the template for handing a process to an AI employee.
- Take the next notice from the retrieval.
- Read what's on it.
- Is the type of assessment known? Yes: continue with the rule for this type. No or unclear: escalation path.
- Fill in the filing fields according to the rule.
- Is the match clear? Yes: file it in the document management system. No: escalation path.
- Is there another notice in the retrieval? Yes: back to step 1. No: continue to the end.
When the retrieval is empty, a completion email goes out. Every escalation path leads to the office team by email, the notice stays open until then, and a feedback session follows.
For the notices that arrive on paper, there's a handoff in front of this flow. They reach the firm's inbox by email, a second AI employee checks every 30 minutes, spots the emails with paper notices and passes them on to the first one. That's exactly the kind of handoff described above. In our simplified description, things then continue from step 2, with the rules for the paper variants.
The flow was signed off after four clean runs, and then monitored side by side for a while. Both processes are at Autonomy Level 2, Autopilot (as of September 2026).
Template: the steps to fill in
The frame around the steps, which means trigger, start, end, report and escalation path, comes from the fill-in version in the article on the process profile. What this article adds is the part in between. As a text template, it looks like this:
- Steps: numbered, each one starts with a verb and ends with a result.
- Question as a step: answerable with yes or no. Yes: continue with which step. No: continue with which step. Unclear: escalation path.
- Finished when: a state you can check.
- Handoff to: a role, and how it finds out (report, email).
Below that goes the variant table from above, one row per variant. If you'd rather see the flow as a picture first, Process visualization: three formats your team understands will help.
Runs without me
Every two weeks, one process that runs without the owner: what came in, what the AI did, where a person decided. Two minutes to read.
Worksheet
Process Potential Check
With five questions and a first process list to fill in, you find the process that could run on its own first. Comes as a PDF by email once you confirm your sign-up.
Strategy session
In a 20-minute strategy session, you tell us where things get stuck. We'll tell you honestly whether an AI employee is a good fit for you.
Book a strategy sessionFrequently asked questions
How detailed does a process flow need to be?
Detailed enough that someone else can carry it out without asking you. For the edges, that means very detailed. For the finer points of each variant, one line is often enough, because you demonstrate those once during training.
What's the difference between a process flow and a process step?
A process step is a single action with a result. The process flow is the sequence of all steps from the trigger to the report, with the branches in between.
Who should write the process flow description?
The person who does the process most often today, together with you. They know the cases, you ask the questions. At the end, they read it through and say: that's right.
What if the flow keeps changing?
Then there are probably variants in it that haven't been written down yet. Every case that comes through the escalation path turns into a new rule. That way, the flow gets more stable over time.
When it's about your own desk first
Not every flow starts in a business with a team. If you want to get your own inbox in order first, Kevin has a system for that: Sorting Your Inbox: The Pile System. This article was about the next step: describing a flow so that someone else can take it over.
Your next step
A flow that someone else can carry out without asking questions is the first step toward a process that no longer depends on one person. Bring a flow to the call that today sits with just one person. Then we'll look together at what it's still missing.