How I scope automation projects (and avoid building the wrong thing)

The failure mode I see most

A client says: "I want to automate my onboarding."

You build it. They approve it. Three months later the workflow is disabled and nobody uses it. What happened?

The process you automated wasn't the bottleneck. The real problem was two steps upstream — a messy CRM that made it impossible to trigger the automation reliably. You built a perfect solution to the wrong problem.

The 5-question discovery framework

Before touching n8n, I run through these five questions:

1. What does this process cost today?

Time × frequency × person's hourly rate. If the math doesn't justify the build, stop here. Automation has maintenance costs too.

"We spend 2 hours/week on this." → At $50/hr, that's $5,200/year. A 20-hour build at my rate is $3,000. ROI in 7 months. ✓

2. How stable is this process?

Automating a process that changes every quarter is expensive in maintenance. The best candidates are:

• Triggered by a clear event (form submit, new record, scheduled time)
• Have consistent input formats
• Produce consistent output formats

3. What breaks when the automation fails?

Some automations are invisible when they work and catastrophic when they don't. Build appropriate alerting before you care about it.

4. Who maintains it after I leave?

If the answer is "nobody," you need to either simplify the workflow dramatically or document it obsessively. I default to: build it so a non-technical person can disable individual steps without breaking everything.

5. What's the happy path vs. the edge cases?

Start with the happy path. Always. Edge cases are endless. If the happy path covers 80% of cases, ship it. Build edge case handling in iteration 2 based on actual production data.

What a good discovery call looks like

1. Map the current process — I ask them to walk me through it step by step while I take notes. No assumptions.
2. Find the pain — "Which step takes the most time?" "Which step has the most errors?" These aren't the same question.
3. Define done — What does success look like in 3 months? "Saved time" is not a definition. "The SDR spends <30 min/day on lead research" is.
4. Identify the data sources — Where does the input come from? Is it clean? Are there API access issues?
5. Scope minimum viable automation — What's the smallest thing I can build that proves this works?

The rule I never break

Don't automate a process you don't understand completely.

Spend 30 minutes doing the process manually before you build anything. This always surfaces assumptions that would have caused a bug two weeks into production.