Spotting Repeatable Workflows to Automate
The biggest ROI win for a manager is not in any single prompt. It is in spotting the repeatable workflows on your team and automating them — partially or fully. This lesson teaches you how to find those workflows and decide which ones are worth automating.
The goal is not to automate everything. The goal is to find the top three workflows that, if automated, give your team back the most time without breaking anything important.
What You'll Learn
- How to map your team's recurring workflows in a 60-minute exercise
- The three-axis scoring system: Volume, Variance, Value
- When AI automation pays back vs. when it doesn't
- Tools for stitching workflows together (Zapier, Make, Power Automate, n8n)
- How to run a one-week pilot before committing
- A first-quarter automation roadmap template
What Counts as a "Workflow"
A workflow is any process your team does more than once a week that has roughly the same shape every time. Common examples on knowledge-work teams:
- Triage of an inbound queue (support tickets, sales leads, applicant resumes)
- Weekly reporting (status updates, exec dashboards, KPI emails)
- Meeting lifecycle (agenda → meeting → transcript → action items → follow-up)
- Document lifecycle (draft → review → approval → publish → notify)
- Customer or stakeholder communication (welcome, follow-up, renewal reminder)
- Internal request fulfillment (access provisioning, expense approvals, onboarding tasks)
Notice the pattern: each workflow has a trigger, a set of steps, inputs, and an output. AI automation slots into the "steps" — usually the drafting, summarization, or triage step.
The Workflow Mapping Exercise (60 Minutes)
Block one hour. Whiteboard or shared doc. Walk through this with your team or your lead reports.
Step 1 — List the triggers (10 minutes). What events recur on your team? "Every Monday morning." "Every new ticket." "Every sales call ends." "Every new hire's first day." List them. Aim for 15-25 triggers.
Step 2 — For each trigger, list the steps (20 minutes). What does someone on the team do in response? Write each step. Be honest — include the "I copy this and paste it into Slack" steps you don't usually call out.
Step 3 — Mark the AI-able steps (10 minutes). For each step, mark with a star if AI could plausibly do it or assist with it. Draft writing, summarization, classification, extraction, and translation are AI-friendly. Decisions, sensitive comms, and judgment calls are not.
Step 4 — Score the workflow on three axes (15 minutes). Use the Volume / Variance / Value scoring below.
Step 5 — Pick the top three (5 minutes). Those are your automation candidates.
The Three-Axis Scoring System
Score every candidate workflow on three axes, each 1-5.
Volume (how often does this happen?)
- 1 = monthly or less
- 2 = weekly
- 3 = a few times per week
- 4 = daily
- 5 = many times per day
Variance (how similar is each instance to the last?)
- 1 = every instance is completely different
- 2 = roughly similar shape
- 3 = same shape, different content
- 4 = same template, just fill in fields
- 5 = nearly identical, one or two variables change
Value (what's the time or quality payoff if you automate?)
- 1 = trivial — saves a few minutes per week
- 2 = nice-to-have — saves 30 minutes per week
- 3 = real — saves 2-3 hours per week
- 4 = significant — saves a half day per week, or removes a bottleneck
- 5 = transformative — frees a full FTE-day per week or unlocks a metric improvement
Total score = Volume + Variance + Value (out of 15).
| Score | Verdict |
|---|---|
| 12-15 | Automate first. ROI is obvious. |
| 9-11 | Automate after the first wave. |
| 6-8 | Consider AI-assist only (Pattern 2 from the delegation lesson). |
| 3-5 | Leave it. Not worth the build cost. |
Most teams find two or three workflows that score 12+. Start there.
When Automation Does Not Pay Back
Be honest about the cases where building automation is worse than leaving the workflow alone:
- The workflow is changing. If your team is mid-reorg or the process is being redesigned, automating now locks in tomorrow's broken process.
- Variance is genuinely high. If every instance needs real judgment, automation will produce confidently wrong output and your team will resent the cleanup.
- The volume is too low. Building a Zapier flow takes a half day. If the workflow runs four times a year, your time is better spent elsewhere.
- Compliance risk outweighs the savings. Any workflow touching regulated data (PHI, PCI, EU personal data) needs legal review before you automate it.
- The "human in the relationship" matters. A customer might appreciate a slightly slower, slightly less polished, but clearly-from-a-human note over a perfect AI-drafted one.
If you have any doubt, run the pilot (next section) before committing.
The Toolkit: How Managers Stitch Workflows Together
You do not need engineering to automate most knowledge-work workflows. Three tool categories cover 90% of cases.
1. Embedded AI (Copilot, Gemini, ChatGPT projects). Best for "use AI inside the tool I already use." Examples: Copilot summarizing an Outlook thread, Gemini drafting in Docs, Claude Projects retaining team context.
2. No-code automation platforms.
- Zapier — easiest to learn, strongest catalog of integrations, monthly plans for small teams. Best for "when X happens in tool A, do Y in tool B with AI in the middle."
- Make (formerly Integromat) — more powerful flow logic, visual canvas, slightly steeper learning curve, cheaper at higher volumes.
- Microsoft Power Automate — best fit if your team is on Microsoft 365 and your IT department already supports it. Integrates with Copilot studio for agent-style workflows.
- n8n — open-source alternative. Self-hostable for teams with privacy requirements.
3. AI-specific specialty tools. Fireflies and Otter for meetings, Gong for sales calls, Lattice/15Five for performance check-ins, Granola for managerial meeting notes. These pre-build entire workflows for one job.
For most middle managers, the right starter stack is one AI-assistant (ChatGPT or Claude), one embedded tool (Copilot or Gemini), and one automation glue tool (Zapier or Power Automate).
The One-Week Pilot
Before you commit to building an automated workflow for your whole team, run a one-week pilot.
Day 1 — Define success. What metric should improve? "Status update prep time drops from 90 minutes to 15." "Ticket response time drops from 4 hours to 30 minutes." Be specific.
Day 2 — Build the minimum viable version. Cheap and ugly. Maybe just one prompt template and a manual paste. No fancy integrations yet.
Days 3-7 — Run it. Use it on real work. Log every time it failed or produced output you had to fix.
Day 8 — Decide. Did the metric move? Were the failure modes survivable? Three outcomes:
- It worked → scale to the team, build the proper integration.
- It half-worked → keep it as a manual prompt template, do not invest in automation yet.
- It didn't work → kill it and try a different workflow.
Most pilots take ~5 hours of manager time. That is the cost of avoiding a bad three-month investment.
A First-Quarter Automation Roadmap
This is a real schedule you can adopt this quarter.
Week 1. Run the workflow mapping exercise. Pick three candidates.
Week 2. Pilot candidate #1.
Week 3. Decide. Pilot candidate #2.
Week 4. Decide. Pilot candidate #3.
Week 5. Decide. Roll out winners. Each winner gets a prompt library entry and a one-page SOP.
Weeks 6-7. Train the team. One 30-minute session per workflow. Reports practice on real tasks.
Weeks 8-10. Measure. Same metric you defined at the start. Report results to your skip-level.
Weeks 11-12. Review and refine. Anything still rough? Adjust. Already-stable workflows move to "maintenance mode" — quarterly review only.
By the end of the quarter you will have two to three real workflow wins, a measurable ROI story for your skip-level, and a team that has actually changed how it works.
Key Takeaways
- A workflow is anything your team does more than once a week with roughly the same shape
- Score candidate workflows on Volume + Variance + Value; pursue 12+ scores first
- Run a one-week pilot before committing to any full automation
- The right manager starter stack: one AI assistant, one embedded tool, one automation glue tool
- Some workflows should not be automated: regulated data, in-flux processes, low volume, relationship-critical
- A 90-day automation roadmap with three pilots delivers measurable ROI you can report up