How to pick your first workflow to automate

Most automation projects fail before they start, because they start too big. "Automate our operations" is a wish. "Sort and summarize the inbox every morning" is a project. The first one you pick matters more than the tool you pick, because it sets the pattern for everything after it.

Here is the method we use with clients to choose a first project that will work.

Step 1: Collect candidates for one week

Keep a running list for five working days. Every time you or someone on your team does something that feels like "I've done this a hundred times," write it down in one line. Don't judge yet. Typical entries look like:

• Copying details from a form into a spreadsheet and a calendar
• Writing the same three replies to the same three kinds of email
• Pulling numbers from two systems into a Monday report
• Renaming and filing documents that arrive by email
• Chasing people for a status update before a meeting

By Friday you'll have ten to thirty lines. That's your candidate list.

Step 2: Score each candidate on four questions

Give every line a score from 1 to 3 on each question. The scores are rough on purpose; you're ranking, not measuring.

Frequency. How often does it happen? Daily or many times a day scores 3. Weekly scores 2. Monthly scores 1. Automation pays back on repetition.

Rules. How clear are the steps? If you could write the instructions for a new hire on one page, score 3. If it needs a lot of "it depends," score 1. AI handles judgment better than old automation did, but clear rules still make the first build faster and safer.

Risk if wrong. What happens when the automation makes a mistake? If a person will see the result before it goes anywhere, or the mistake is easy to catch and fix, score 3. If an error could reach a customer, touch money or involve sensitive information without anyone looking, score 1.

Visibility. Will people notice when it works? A task the whole team hates scores 3. A chore only one person does, quietly, scores 1. The first automation also has to sell the second one.

Add the four numbers. Anything that scores 10 or above is a strong first project. Anything below 7 waits.

Step 3: Cut the list to one

You'll usually have two or three strong candidates. Pick one by asking which has the shortest path to "running." Prefer the task whose inputs already live in a system you use every day (email, a shared drive, your CRM) over the one that starts on paper or in someone's head.

If two are tied, pick the one that annoys the most people. Momentum matters.

Step 4: Write the workflow as it is today

Before building anything, write down how the task works right now, step by step, including the ugly parts: the second spreadsheet someone keeps on the side, the person who always needs a reminder, the exception that comes up every other week. This is where most of the design decisions hide.

Two questions to answer in writing:

1. Where does a person need to approve? For a first project, the answer is almost always "before anything leaves the building." The automation drafts the reply, prepares the record or assembles the report; a person looks and clicks send. You can loosen this later, once you trust it.
2. What must never go into an AI tool? Customer financial details, health information, passwords, anything under a confidentiality agreement. Decide now, and design the automation so that information stays in the system it already lives in.

Step 5: Measure before you build

Time the task as it works today. Not a guess, a real timing for a week: how many times it happened and how long each one took. Write the number down. This is the only way you'll know later whether the automation worked, and it's the number that makes the case for the next project.

Step 6: Build small, then widen

Build the narrowest version that handles the common case, with every exception routed to a person. Run it alongside the old way for a week or two. Then handle the most common exception, and the next. An automation that handles 80 percent of cases and hands the rest to a human is a success. One that tries to handle everything on day one usually ships late or not at all.

What a good first project looks like

The pattern we see most often: a daily or weekly task, fed by email or a form, with a clear set of steps, where the output is a draft a person reviews. Inbox triage, meeting-notes-to-tasks, report assembly and intake-to-record all fit. Each can be built in the tools most businesses already have, each shows a visible result within a few weeks, and each teaches the team how to work with an automation before you trust one with anything bigger.

Pick one. Score it. Time it. Build it with a checkpoint. Then do the next one.