To document a process so it can be automated, write down what starts it, every step in order (one action per line, with who does it and in which system), the data each step uses, the rules behind every decision, the exceptions, and what "done" looks like. Do it by watching the person who actually does the work, not from memory, and capture real numbers: how often it happens and how long each step takes. A process written this way can be handed to a developer, an automation platform or an AI tool and turned into a working automation without weeks of guesswork.
Download the free printable worksheet (PDF) to fill in as you go, or follow the steps below.
Why most process documents can't be automated
Many businesses have procedures written down somewhere, usually as a few paragraphs in a shared document: "Enter the order, check inventory, send the confirmation." That's fine for training a person, who fills the gaps with common sense. Software can't.
To automate a process, someone has to know:
- Exactly what starts it. An email arriving? A form submission? A date on the calendar?
- Exactly what happens at each step. "Enter the order" might be eight separate actions across three systems.
- What data moves where. Which fields come from the email, which get looked up, which get created.
- How decisions are made. "Big orders get checked" isn't a rule. "Orders over $5,000 need manager approval" is.
- What happens when things go wrong. Missing information, duplicates, a customer who isn't in the system yet.
The gaps between "how we describe it" and "how it actually works" are where automation projects go over budget. Closing them before anything is built is the cheapest step in the whole project.
Step 1: Pick one process and set its boundaries
Start with a single process that happens often and takes real time. Our guide on what a small business should automate first explains how to choose.
Then define the edges:
- Trigger: the event that starts it ("A customer submits the quote form").
- Done: the result that ends it ("The quote is sent and logged in the CRM").
- Owner: the person responsible for it today.
A process without clear edges grows as you document it. If you find yourself writing about invoicing while documenting order entry, that's a second process. Note it and keep going.
Step 2: Watch it happen
Sit with the person who does the work (or have them share their screen) and watch them handle three to five real examples. Ask them to talk through what they're doing and why. Recording the session, with their permission, makes the next steps much easier.
This matters because the way a manager describes a process and the way it actually happens are rarely the same. Watching uncovers the extra spreadsheet, the sticky note, the "I always double-check this because it was wrong once," and the step that only one person knows about.
Step 3: List every step, one action per line
Write each step as a verb plus an object, with the person and the system:
| # | Step | Who | System |
|---|---|---|---|
| 1 | Open the new order email | Office manager | Outlook |
| 2 | Copy customer name and PO number | Office manager | Outlook → QuickBooks |
| 3 | Look up the customer | Office manager | QuickBooks |
"Process the order" is too big; "Click the Save button" is too small. Aim for steps someone could check off one at a time. If a step involves judgment, mark it with a star. Those are the places to think hardest about whether to automate or keep a person in the loop.
Step 4: Capture the data at each step
For each step, note what information it uses and where that information comes from:
- Inputs: fields read from the email, form, file or another system
- Lookups: information pulled from a system (a customer record, a price, a stock level)
- Outputs: records created or updated, messages sent, files saved
This is the part most process documents skip, and it's the part an automation depends on most. It also shows where data is re-typed between systems, which is usually the first thing worth automating. If the same customer details are entered in three places, that's three chances for errors and one clear opportunity.
Step 5: Turn decisions into written rules
Every place someone decides something, write it as an if/then rule with real numbers:
- If the order total is over $5,000, then a manager approves it before it's entered.
- If the customer isn't in QuickBooks, then create them using the details from the form.
- If the requested date is less than three business days away, then flag it as a rush.
If a decision can't be written as a rule because "it depends," ask what it depends on. Often the answer turns into two or three rules. When it truly comes down to experience and judgment, that step stays with a person, and the automation prepares everything they need to decide quickly.
Step 6: List the exceptions
Ask: what goes wrong, what's unusual, and how often? Common examples:
- Required information is missing
- The same request arrives twice
- A customer, product or account doesn't exist yet
- An attachment can't be read
- Something needs a manager's sign-off
For each one, write how often it happens and what the person does about it today. You don't have to automate every exception. A good automation handles the common cases and sends the rare ones to a person with a clear note. But you do have to know they exist, or the automation will fail on them quietly.
Step 7: Measure time, volume and waiting
Write down:
- How often the process runs (per day or week)
- How long each run takes, and which steps take the longest
- Where work waits: for an approval, for a reply, in an inbox overnight
Waiting is often where the biggest gains are. A process that takes 15 minutes of work might take three days from start to finish because it sits in queues. These numbers also let you estimate the payoff. Our guide to what AI automation costs shows how to turn hours saved into a yearly figure.
Step 8: Check it with the people who do the work
Hand the finished document to the person who does the process and ask them to follow it exactly for the next few examples. Anywhere they hesitate, skip something or do something not on the page is a gap to fill. Then ask a second person who does the same job, because two people often do the "same" process differently. Agree on one way before automating it.
A short worked example
Here's what a documented process might look like for a small distributor handling purchase orders by email. (This is an illustration, not a specific client.)
- Trigger: a customer emails a purchase order as a PDF.
- Done: the order is in the accounting system and the customer has a confirmation with a ship date.
- Volume: about 40 orders a week, around 12 minutes each, roughly 8 hours a week in total.
- Steps: open the email; download the PDF; read the customer, PO number, items, quantities and requested date; look up the customer; check stock for each item; enter the sales order; email a confirmation; file the PDF.
- Rules: orders over $5,000 need approval; rush requests under three business days are flagged; new customers are created from the PO details.
- Exceptions: an item number doesn't match the catalog (about 1 in 10); a PO arrives as a photo instead of a PDF; duplicate POs.
- Waiting: orders that arrive after 3 p.m. aren't entered until the next morning.
From that one page, it's clear what to automate: AI reads the PDF and fills in the fields, the workflow looks up the customer and stock, creates the order and sends the confirmation, and only mismatched items, large orders and rushes go to a person. It's also clear what has to be fixed first: a clean item list the AI can match against.
Tools that help
You don't need special software. A spreadsheet or the printable worksheet is enough for most processes. A few things make it easier:
- Screen recordings of the work being done, so you can rewatch the details
- A simple flowchart for processes with several branches
- Sample files: a few real (anonymized) emails, forms or PDFs the process handles, which developers and AI tools need for testing
What to do with your documented process
Once it's written down, you have three options:
- Fix it before automating it. Documentation often reveals steps that can simply be removed.
- Automate it yourself with a platform like Make, Zapier or n8n if it's straightforward. Our comparison of the three helps you choose.
- Hand it to a developer or automation partner. A well-documented process gets a faster, more accurate estimate and a better result.
For the bigger picture of whether your business is ready, see our AI readiness checklist.
Want this done for you?
Documenting processes is the first half of our AI Readiness Audit. Over two weeks, we sit with the people who do the work, map how requests, data and documents actually move through your business, including the steps nobody has written down, and hand you a ranked list of automations with estimated hours saved and the cost to build each one. The price is fixed and agreed before we start, and the roadmap is yours to keep.
Book a free call and tell us which process eats the most hours. We'll tell you what it would take to automate it.