DEV Community

Hive80-lab
Hive80-lab

Posted on

Should you automate this? The 4-question filter (with worked examples)

#automation #productivity #devops #smallbusiness

Should you automate this? The 4-question filter (with worked examples)

The first automation a small team builds is usually the wrong one — the exciting one, not the valuable one. A chatbot nobody asked for beats an invoice reminder nobody has to send, in excitement. In value, it's not close.

Before building anything, run every candidate through four questions in order. This is the filter we use for our own recurring ops work; it takes 20 minutes and has killed more bad projects than it has approved.

The 4 questions (in order)

  1. Does this happen at least weekly? Automation pays back through repetition. A once-a-quarter task that takes 10 minutes isn't worth the build, however fun the build is.
  2. Is the task's rule writable in one sentence? "Every completed order gets a confirmation email" — writable. "Review whether this lead looks promising" — not yet. If you can't state the rule in one sentence, you don't understand the task well enough to delegate it to a machine.
  3. Is failure visible within a day? Good candidates fail loudly or observably (email didn't send, invoice missing). Bad candidates fail silently for weeks (data enrichment — wrong output looks like right output).
  4. Can a human check the output in under a minute? If verification is expensive, you've automated the task and manualized the review — the worst trade in automation.

Scoring: four yeses = automate next sprint. Two or fewer = leave it alone for now. Three = fix the weak answer first — usually #2: write the rule sentence until it's actually true.

The 3 red flags (any one means don't automate yet)

  • The process changed in the last month. You'd be automating a snapshot of a moving process. Stabilize first.
  • Nobody currently does the task end-to-end. Automation needs a reference human; without one there's no ground truth to check against.
  • The task exists to catch rare, expensive errors. Human review IS the control here. Automate the logging of those errors, never the judgment.

Worked examples

Candidate Weekly? One-sentence rule? Visible failure? <1min check? Verdict
Invoice overdue reminders Y Y ("invoice unpaid >7d → reminder") Y (email log) Y Automate
Backup verification alert Y Y ("verify restore-test log exists daily") Y Y Automate
Lead scoring Y N N (silent) N Leave alone
Refund approvals N N N N Leave alone — human judgment is the control

The worksheet

Copy this block per candidate. Aim for 3–5 candidates, then pick the top scorer.

Candidate: {{NAME}}
Frequency (per week): {{N}}
Rule in one sentence: "When {{TRIGGER}}, then {{ACTION}}."
How failure shows up within a day: {{SIGNAL}}
Human check (what, how long): {{CHECK}}
Red flags present: {{NONE or WHICH}}
Score: {{X}}/4 → verdict: {{AUTOMATE / FIX RULE FIRST / LEAVE ALONE}}
Enter fullscreen mode Exit fullscreen mode

What stays human, always

Anything involving judgment about people — hiring, pricing decisions, escalating an angry customer. Automate the paperwork around those, never the decision.

And once you've picked a winner, don't skip the boring part: run it in shadow mode for a week, give it a kill switch, and put a monthly review on the calendar. Automations don't die loudly; they rot quietly until someone notices the wrong invoices went out in March.

The full pack — the filter worksheet, 12 workflows ranked by payback, the one-page spec (shadow mode + kill switch + run log), the 2am incident-log automation, 5 AI workflow templates with human checkpoints, and the 10-minute monthly review — is the Automation Starter Pack ($19): https://hive80lab.gumroad.com/l/automation-starter-pack

Launch month: 30% off with code HIVE-LAUNCH30.

More from Hive80 Lab:

Top comments (0)