Guide
Which tasks to give AI — and which not to
The hard part of AI isn't the technology — it's knowing where it belongs. Give it the wrong job and you get an expensive mess and unhappy customers. Give it the right job and it quietly pays for itself. Here's a simple way to decide.
By Miguel Alejandro Hayes· Founder, Hayes Projects
The three-question test
Before automating anything, ask three questions. First: does it happen often? Automation only pays back on volume. Second: does it follow clear rules? If the steps are the same every time, a machine can do them. Third: how much human judgment does it need? The less, the better a candidate it is.
Score each task on those three. High frequency, clear rules, low judgment = automate it now. Low frequency, fuzzy rules, high judgment = keep it with a person. Most of your work falls clearly on one side; the interesting cases are in the middle, and those are usually best done as a hybrid — the AI proposes, a person decides.
Good jobs for AI
AI and automation shine at: answering common questions 24/7, qualifying and routing leads, taking a first pass at a document or reply, moving and reconciling data between systems, sending reminders and follow-ups, and summarizing or classifying large volumes of text. These are frequent, rule-based, and low-judgment — exactly the sweet spot.
In a business, that often looks like an agent that answers the phone or WhatsApp after hours, a flow that qualifies inbound leads before they reach sales, or an automation that keeps your CRM and billing in sync. The customer gets a faster response; your team stops doing the rote part.
Bad jobs for AI (keep these with people)
Keep humans on: decisions with real consequences (money, legal, health), anything that depends on relationship or trust, judgment calls with incomplete information, and one-off exceptions where the “rules” don't really exist. Handing these to a machine to save time usually costs more than it saves.
The healthiest setup is almost always a mix: automate the routine so your people have the time and energy for the work that needs a human. AI that knows when to hand off to a person — instead of pretending it can do everything — is the version that actually works.
Frequently asked
Will AI make mistakes with my customers?
+
Any system can. The fix is design: give AI the rule-based work, set clear limits, and make it hand off to a person when it is unsure or the conversation goes off-script. A good build never leaves a customer stranded.
Can I start small?
+
Yes — and you should. Pick one high-frequency, rule-based task, automate it well, measure the result, then expand. Small wins in production beat big rollouts that never ship.
What about data privacy?
+
Your data should stay yours. Whatever you build, you should own it, control how long data is kept, and be able to audit what the system did. We sign an NDA and adapt handling to regulated sectors.
About the author
Miguel Alejandro Hayes — Founder, Hayes Projects
Economist and essayist turned developer. He founded Hayes Projects, a Miami venture studio and custom software lab, to build software that ships, scales and solves real problems.
Meet the founder→Not sure what to automate first? Our free AutoMap tool scores each task for you — or book 20 minutes and we’ll map it together.
Try AutoMap — free →