Which tasks in my business should I never hand fully to AI?

By Ranjit Bhinge, founder of Blur Studio · Updated · 6 min read

Short answer: Do not fully hand to AI any task where a wrong answer is costly and hard to undo: signing contracts, moving money, firing or hiring decisions, legal or medical advice, and handling an angry customer alone. AI, meaning software that writes or decides using patterns learned from large amounts of text, is fine for drafting and sorting on those tasks, with a person approving. Use a two-question test: what does a wrong answer cost, and can a person undo it?

Ranjit Bhinge

You want to know where AI is safe to use in your business and where a human has to stay in charge.

I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.

If you want a sorted list for your own business, the AI Strategy Sprint separates what to automate from what to keep human in 3 days.

The tasks to keep human are the ones where a mistake is expensive and cannot be taken back. Everything else is a candidate for AI, sometimes fully, more often as a first draft that a person approves. This page gives you a test you can run on any task, then applies it to 15 common ones.

Why the caution? AI tools write fluent text and can still be wrong, and they do not know when they are wrong. They can state an invented price, date or policy in a confident tone. So the question is never "can AI do this?". It is "what happens if it gets it wrong, and who catches it?"

The two-question test

  1. What does a wrong answer cost, compared with the time saved? A mistyped subject line costs nothing. A wrong refund policy sent to 200 customers costs real money. If one error would cost more than a month of savings, keep a person in charge.
  2. Can a person undo it? A draft that sits in a folder can be fixed. A payment that has left the bank, a signed contract or a firing letter cannot be fixed by deleting it.

Score a task on both and sort it into one of three groups:

Wrong answer costCan it be undone?Group
LowYesAutomate. AI does it, you spot check
Medium or highYes, with effortDraft only. AI drafts, a person approves
HighNoKeep human. AI may assist with research, a person decides

Fifteen common tasks, sorted

TaskGroupWhy
Sorting incoming email into foldersAutomateWrong sort is a minor annoyance, easy to fix
Summarising long documents or meeting notesAutomateLow cost; spot check against the source
Tagging and sorting support ticketsAutomateA wrong tag is cheap to correct
Drafting social posts and newslettersDraft onlyPublic and hard to retract, so a person reads it first
Quote and proposal first draftsDraft onlyA wrong price becomes a promise to a customer
Routine customer replies (order status, opening hours)Draft only, or automate with guardrailsSafe when answers come from your own approved text
Bookkeeping categoriesDraft onlyErrors compound into tax and reporting mistakes
Reviewing a contractDraft onlyAI can flag clauses; a person with legal judgment decides
Replying to an upset or complaining customerDraft only, then humanTone and trust matter more than speed
Pricing and discount decisionsDraft onlyAI can show options; you own the margin
Approving or sending payments and refundsKeep humanMoney that has left is rarely recoverable
Signing contracts or agreeing termsKeep humanBinding and hard to reverse
Hiring, firing and performance decisionsKeep humanLegal exposure and real harm to people
Legal, medical or tax advice to customersKeep humanProfessional duty; a wrong answer can hurt someone
Anything involving a client's confidential data in an unapproved toolKeep humanBreach of trust or contract

Treat the "group" as a starting point. A cheap task can move up a group if your business is small enough that one error hurts, and a risky one can move down once you have a good check step in place.

The hard cases in more detail

Contracts

AI is good at reading a contract and listing what looks unusual: a long notice period, an automatic renewal, a liability cap. It is not responsible for the result. A person who understands your risk should read the flagged clauses. My guide to AI contract review shows what the tools can and cannot do. For anything large, ask a lawyer.

Upset customers

A fast, polite draft helps. But an angry customer wants to feel heard by a person who can decide something. Let AI write the first version of the reply, then change it, add the decision and send it yourself. An automated reply to a complaint that misses the point makes things worse.

Money moves

Let software prepare the payment list, flag duplicates and match invoices. Have a person approve and release. If you automate payments at all, put a limit on the amount and a second approver above it.

Hiring

Using AI to screen candidates carries legal risk. New York City's Local Law 144, for example, lets employers use an automated employment decision tool only if it has had a bias audit within the past year, the audit results are public, and candidates and employees get the required notices. Other places have their own rules, and the law is changing. If you use AI anywhere near hiring, ask a lawyer first. A person should make the decision.

Things you have told customers

If AI writes on your behalf, it speaks for you. In a 2024 announcement, the chair of the US Federal Trade Commission (FTC) said there is "no AI exemption from the laws on the books." A chatbot that invents a refund policy can become your problem, not the software's. Keep a chatbot to answers it can take from your own approved text, and give people an easy way to reach a person.

What AI is good for as a starting point

If you want something safe to try this month, pick from the low-cost, easy-to-undo group:

  • Summarising meeting notes into action items.
  • Sorting and labelling email or support tickets.
  • Rewriting a rough draft to be clearer.
  • Pulling dates and amounts out of documents into a spreadsheet, then checking a sample.
  • Drafting the first version of routine replies for a person to approve.

The AI for small business guide and the AI workflow automation guide have more starting points, and business process automation explains the wider idea. If a task keeps coming up, an AI employee that does the routine part and hands the rest to a person may be worth a look.

How to keep a person in the loop without slowing down

  • Approve by exception. Let the AI handle clear cases and send only unclear ones to a person.
  • Make the check a step, not a hope. The workflow should not send anything until a named person clicks approve.
  • Sample. For low-risk tasks, review one in ten and watch the error rate.
  • Keep the source visible. The reviewer should see what the AI used, so they can check it fast.
  • Log it. Keep a record of what the AI did and who approved it.

The US National Institute of Standards and Technology (NIST) publishes a voluntary AI Risk Management Framework for managing AI risks. For a small business, the plain version is: decide who owns it, know where it is used, check how it performs, and fix what goes wrong. My guide to human-in-the-loop AI workflows covers the build.

Who should be extra careful

If your work is regulated, such as health, law, finance or work with children, move every task up one group and ask a professional what applies. If you handle other people's confidential data, check which AI tools you may use at all. My guide to AI tools that do not train on your data is a place to start. OpenAI (maker of ChatGPT, a chat assistant) states that by default it does not train its models on business data from ChatGPT Business, ChatGPT Enterprise or its API, and the FTC's data security guidance tells businesses that hold sensitive customer or employee data to collect only what they need, keep it protected and dispose of it securely.

How I'd approach this

I would write down ten tasks, run the two questions on each, and put them in three columns. Start by automating two from the first column, and watch them for a month. Do not touch the third column. The aim is not to use AI everywhere. It is to use it where mistakes are cheap, and keep your judgment where they are not.

I would also review the list every few months, because tools improve and your checks get better. A task can move from "keep human" to "draft only" once you have a reliable review step. It rarely moves to "automate". If you want a sorted list for your own business, the AI Strategy Sprint is a paid 3-day first step where we find where AI could create real new value in your business and write down the exact steps. It starts with a free 15-minute AI brainstorm call.

Questions people ask

What should never be automated with AI?

Anything where a wrong answer is costly and cannot be undone: signing contracts, releasing payments, hiring and firing decisions, and legal, medical or tax advice. AI can help with research and drafts on these, but a person decides. See the table above for 15 common tasks.

Can AI handle customer service on its own?

For simple, repeat questions answered from your own approved text, often yes, with an easy route to a person. For complaints, refunds and anything unusual, a person should decide. See the AI customer service agent guide.

Is it safe to use AI for hiring?

It carries legal risk. Some places, such as New York City, require bias audits and notices for automated hiring tools. Ask a lawyer before using AI to screen people, and keep a person making the decision.

Can I trust AI to review contracts?

It is useful for flagging unusual clauses fast, and unreliable as the final word. Read the AI contract review guide, and have a lawyer review anything important.

How do I decide what to automate first?

Pick a task that is frequent, cheap if wrong and easy to undo, such as sorting email or summarising notes. Run it for a month, review a sample, then pick the next one. The AI implementation guide explains the steps.

More in this topic: Problem solving: AI strategy and adoption (24 guides) →

Sources

Ready for the next step?

A free 15-minute AI brainstorm. We find 3 broad areas where AI could help your business.

Ranjit Bhinge You talk to me, Ranjit. I reply within a few hours.
See the AI Strategy Sprint

Not sure what you need? Tell me what is going on. I will point you the right way, even if it is not me.

Blur Studio · 30 minute call

Book a 30 minute call

You talk to me, Ranjit. If you do not need me, I will tell you.

  • We talk about your business and what is going on.
  • You leave with a clear next step, even if it is not me.
  • I reply to every message within a few hours.
Loading the calendar…