Automation decision framework

Pick what to automate first, then pick the tool. Most teams reverse that and pay for a platform that encodes chaos.

List three candidate loops. Score them. Automate the winner’s capture path, not all three.

Choose the platform after the winner is written down. Feature lists are how you buy the wrong seat.

flowchart LR
    C["Candidates"] --> S["Score"]
    S --> W["One winner"]
    W --> T["Tool"]
    T --> B["Build capture only"]
Cream scorecard on a wooden desk labeled Score, Tool, and Build, with a terracotta GO stamp
If it does not fit on a card, you are not choosing. You are browsing.

Rank three candidates

Score 1 (painful but rare) to 5 (daily, rule-based, currently copy-paste). Highest total goes first. Ties break toward the loop that touches money or a client.

CandidateVolumeRule-clarityTime savedRisk if wrongTotal
Inbound lead intake
Publish / ship checklist
Invoice or payment chase
  • Winner named in one sentence

    New /contact row to a Sheet, ping Simeon, same-day human reply.

  • The other two parked with a date, not “someday”

Then pick the platform

n8n, Zapier, and Make can all POST a form to a sheet. The difference is where the work lives, how you handle secrets, and whether you will outgrow the task model. Deep comparison is in the guide. This table is the pick.

If this is trueStart hereDo not start here
You want the workflow as a file you can version, self-host, or inspectn8nA black-box zap you cannot diff
Non-technical owners must edit mapping in a vendor UI this weekZapier or MakeA half-finished n8n instance nobody will open
The step needs a large language model (LLM) draftAny of the three, plus a human-in-the-loop (HITL) pauseSilent send from the model
Secrets, production data, or a kill switch mattern8n on infra you control, or a vendor with env isolation you understandKeys pasted into a node and screenshotted into Slack
  • Platform written next to the winner (one name)
  • One line on why the other two lost, so you do not reopen the debate next month

Build shape (first two weeks)

  • Validate, store, notify are deterministic (no model)
  • If a model is involved, it proposes JSON. Schema still decides pass/fail
  • Send, charge, delete, or post: named human, timestamped
  • Global pause exists (workflow inactive or env flag) before the first live run

Related guides