How to Find the First Workflow to Automate

Pick a first workflow that repeats at least weekly, takes visible time each run, follows rules two people would apply the same way, and has a natural point where a human approves the output before it reaches a client. In most agencies that’s recurring reporting, client status updates, follow-up reminders or new-client intake. Creative and relationship work can wait. Those routine jobs win because they come round often enough for you to check the output, fix the rules and trust the result within a few weeks.
Here’s the usual situation. You know you should automate something. You’ve got a list of twenty candidates in your head, and the one you’re most excited about is the hardest: drafting strategy decks, or writing campaign concepts.
That’s the wrong place to start. Your first workflow is there to prove the process. Can your team describe the rules, review the output and trust it? Choose something boring that runs often and lets you answer that quickly.
This guide gives you a scoring method, a worked example with real agency task categories, the checks that rule a candidate out, and a stop rule for when your pick isn’t working.
What makes a workflow a good first automation?
A good first workflow passes four tests. Fail one and it isn’t your first.
- It repeats often. Weekly or more. Something that runs once a quarter gives you four chances a year to learn whether it works.
- It takes visible time. People can point to the hours. If nobody notices the work disappearing, nobody notices the win either.
- It follows stable rules. Hand the same inputs to two people and you’d get the same output. If the steps change every month, you’ll spend your time rewriting the automation.
- It has a safe approval point. There’s a moment where a person can look at the result before anything goes to a client, gets paid or gets filed.
The first three line up with how researchers describe automation’s sweet spot. A systematic literature review of robotic process automation by Judith Wewerka and Manfred Reichert, covering 63 publications, defines RPA as automating rule-based routine work.
The fourth test comes from how we design installs at Magic Teams. We build human-in-the-loop systems, so every workflow we set up has a named person who approves the output. If a candidate has no obvious place for that check, we don’t pick it as a first project.
Why does the approval point matter so much? Because your first workflow will get things wrong. The approval point decides whether those mistakes are cheap lessons or awkward client emails.
Why do reporting, status updates, reminders and intake usually win?
They’re the kind of coordination work that eats a week without anyone choosing it. Asana’s Anatomy of Work Index 2021, based on 2020 survey data, found knowledge workers spend 60% of their day on “work about work” rather than skilled work.
The interruptions add up too. Microsoft’s 2025 Work Trend Index research, drawn from Microsoft 365 usage data, found employees are interrupted every two minutes during core work hours by a meeting, email or chat. Chasing status and nudging people for replies feed that noise.
There’s also a practical reason. These workflows give you a fast feedback loop:
- Reporting has a fixed structure, and most of the data is already in your tools.
- Status updates run every week, so you get four review cycles a month.
- Follow-up reminders are short, low-risk and easy to check.
- Intake has a defined form, a defined owner and a clear “done” state.
Creative work and relationship calls are different. They depend on taste, timing and trust. Automation can help with drafts there later, once your team trusts the review process. They make poor first projects because nobody agrees on what “correct” looks like.
How do you score candidates with the FTJ score?
FTJ stands for Frequency, Time and Judgment. It’s the scoring method we use on this blog to decide which tasks to automate first, and it’s the same test we apply in our guide to what to automate in a marketing agency. The three questions are the ones that matter most when deciding whether a workflow is safe and worth handing to a system.
For choosing one first workflow, use simple bands so you don’t get stuck estimating minutes:
- Frequency (1-3): monthly or less = 1, weekly = 2, several times a week or across many clients = 3.
- Time (1-3): under 15 minutes a run = 1, 15-60 minutes = 2, over an hour = 3.
- Drain = Frequency x Time. That gives you a range from 1 to 9.
- Judgment (1-5): 1 means the same answer every time, 5 means a senior person’s call.
Then apply the gates. Judgment of 4 or 5 is out. Judgment of 3 stays in only if there’s a clear approval point. Unstable rules are out, and rules that are only partly stable push a candidate down the queue.
Among what’s left, the highest drain wins. If two are close, choose the one whose review cycle comes round more often, because it teaches you faster.
Here’s the same logic as a decision tree.
What does a worked FTJ example look like?
Picture a 15-person marketing agency with about 20 retainer clients. The owner lists seven recurring workflows. The bands below are example inputs to show the method. They aren’t measurements from a client, and they won’t match your agency exactly.
| Workflow | Frequency | Time | Drain | Judgment | Stable rules? | Approval point? | Verdict |
|---|---|---|---|---|---|---|---|
| Weekly client status updates | 3 | 2 | 6 | 2 | Yes | Account manager reviews before send | First pick |
| Monthly campaign reporting | 3 | 3 | 9 | 2 (data), 3 (commentary) | Partly | Strategist signs off | Second |
| Proposal follow-up reminders | 2 | 1 | 2 | 1 | Yes | Sales lead approves template once | Quick add-on |
| New-client intake | 1 | 3 | 3 | 2 | Yes | Ops lead checks the brief | Third |
| Invoice reminders | 2 | 1 | 2 | 2 | Yes | Finance approves each send | Later, with care |
| Creative concepting | 2 | 3 | 6 | 5 | No | Creative director | Keep human |
| Renewal and upsell calls | 1 | 2 | 2 | 5 | No | Owner | Keep human |
Campaign reporting has the biggest drain at 9, because 20 reports a month works out to several runs a week. Creative concepting ties with status updates at 6, then drops out straight away on judgment.
So why isn’t reporting first? Its rules are only partly stable. Ad platforms change which metrics they expose, as we cover in our guide to automating campaign reporting, and the commentary needs a strategist’s judgment. The runs also bunch into the first week of the month, so the team gets one review cycle a month to learn from.
Status updates pass every gate cleanly. They go out every week to every client, so the team gets roughly four review cycles a month to check drafts and tighten the rules. That makes them the first pick.
Reporting comes next, once the review habit exists. It can start with the data assembly while the strategist keeps writing the commentary.
Notice what this table doesn’t claim. It doesn’t promise hours saved. You’ll only know that after you time the work before and after, which is the point of starting with something you can measure.
Here are the drain scores for each workflow side by side, using the example bands above.
The chart shows why drain alone isn’t enough. Creative concepting looks as attractive as status updates, and reporting looks better than both, until you apply the judgment and stable-rules gates.
What should rule a workflow out as your first?
Some candidates look good on paper and still make poor first projects. Run your top pick through this check before you build anything.
- Two people would produce the same output from the same inputs
- A named person can review the output before it leaves the business
- The inputs already live in tools you can connect to
- A wrong output is easy to catch and cheap to fix
- The steps haven't changed in the last few months
If a workflow fails on the fourth point, slow down. Anything touching money, legal documents, regulated client data or professional advice needs a stricter approval step.
If you run a law, accounting or advisory practice, check your professional and data protection obligations. Talk to a qualified adviser before automating any client-facing step.
If it fails on the last point, fix the process first. Automating a messy process just makes the mess faster. Our AI readiness assessment for agencies covers how to tell whether your processes are documented well enough to start.
How do you run the first workflow once you’ve picked it?
Keep the first month small and visible.
- Write the rules down. One page: inputs, steps, output, who approves, what counts as wrong.
- Time the manual version. Two or three runs is enough to get a baseline.
- Run it in draft mode. The system prepares the output and a person approves every one.
- Log every correction. Each edit the reviewer makes is a rule you missed.
- Loosen review only when corrections stop. Some workflows should keep human approval permanently, and that’s fine.
This is the same human-in-the-loop pattern we use in a Magic Teams install. The approval point stays useful long after launch. Reviewers build trust in the system, and they catch the cases the rules didn’t cover.
When should you drop your first pick and try another?
Your correction log tells you. Read it after a few review cycles.
If the corrections are shrinking and repeating, keep going. You’re finding missing rules and closing them. That’s the system working.
If the corrections are different every time, or the reviewer keeps rewriting whole drafts, the rules weren’t as stable as you scored them. Go back to your table, mark that workflow’s rules as “partly stable” and move to the next candidate. A switch after a month costs far less than forcing a workflow that never settles.
Frequently asked questions
Should my first automation be the task I hate most?
Only if it also passes the gates. Founders often hate high-judgment work, like pricing or difficult client emails, and that’s exactly the work that shouldn’t go first. How much you hate a task tracks frequency and time. It tells you nothing about safety.
What if every candidate has low frequency?
Count across clients, not per client. One monthly report is low frequency, but twenty monthly reports is several runs a week. If it’s still rare, batch similar small tasks into one workflow.
Can I start with an AI agent instead of a simple workflow?
You can, but you’ll learn less. A narrow workflow with a clear output shows you where your rules are vague. Broad agents hide that until something goes wrong in front of a client.
How is this different from choosing which tasks to delegate to AI?
Delegation sorts your whole task list into AI, review and human work. Choosing a first workflow means picking one connected process to build end to end. For the wider sort, see how to spot the right tasks to delegate to AI.
Who wrote this?
This guide is by Satya Phanindra Reddy, founder of Magic Teams AI. We install local-first, human-in-the-loop AI operating systems for founder-led agencies and small professional practices, usually in a one-week intensive. Picking the first workflow, and the person who approves it, is one of the first decisions we make with a founder.
If your first workflow has to pull from your project tool, your CRM and your reporting stack at once, the connections between those tools usually matter more than any single automation. That’s the kind of setup Magic Teams helps founders map out.