Short answer: Automate work that is low-risk, repetitive and easy to undo. For work that needs judgement, let AI prepare it and have a named person approve it. Escalate anything high-stakes to people, and keep AI out of that decision.
I'm Vinay Pasricha, founder of GoodSpace AI and author of AI for Business Leaders. The question I hear most from owners isn't "should we use AI?" It's "where?" My answer is a simple sort, task by task, into three groups.
The three groups
Automate. Work that is repetitive, low-risk and reversible. If AI gets it wrong, someone notices quickly and fixing it costs little. Examples: turning meeting notes into a summary, moving data from forms into a sheet, filing invoices, sending reminders about overdue tasks.
Assist. Work that needs judgement, but where most of the effort is preparation. AI drafts; a named person reviews and decides. Examples: a quote that a salesperson checks before it goes out, a monthly report that the finance head signs off, a shortlist of candidates that a manager reads before choosing whom to meet.
Escalate. Work where a mistake is expensive, hard to undo, or touches a person's livelihood, safety or legal standing. People decide; AI stays out of the decision, though it can still help someone find information. Examples: letting someone go, pricing a large contract, answering a complaint about safety.
Three questions for each task
To sort a task, ask three things:
- What happens if it's wrong? An embarrassing typo is not the same as a wrong price to your biggest customer.
- Can it be undone? A draft can be deleted. A message to a customer, or money that has moved, often can't.
- Does it need judgement about a person, money or a relationship? If yes, it is at least Assist.
Most tasks land in Assist. That's fine. Assist is where most of the value is in the early months, because your people stop doing preparation and spend their time on decisions.
The rule that makes Assist work
"A person reviews it" only works if one person's name is on the approval. If nobody in particular is responsible, people stop reading what AI prepared and simply pass it on. That's automation without admitting it. Write down who approves each Assist task, and what they check.
Two common mistakes
- Automating a broken process. If nobody can explain how a task should be done, AI will do it wrong faster. Fix the steps first.
- Automating what is visible instead of what is valuable. A chatbot on the website is visible. Freeing your best salesperson from three hours of daily data entry is valuable.
One thing to do this week
List ten tasks your team does every week. Run them through the free Automate, Assist, Escalate tool: it asks a few questions about each task, sorts it, and gives you a one-page PDF. Pick one task from the Automate group and start there. Pick one from Assist next, with a named approver.
Where this comes from
This answer comes from Automate, Assist, Escalate, one of the frameworks in AI for Business Leaders. The book is in hardcover and on Kindle.
If you want to test a specific task, ask the books in your own words on Ask Vinay's books.
Vinay Pasricha is the founder of GoodSpace AI and the author of six books, including AI for Business Leaders.
