AI & Work

Before AI acts on its own, name who answers for it

Deloitte published a survey of 3,500 leaders across 25 countries at the end of October. Two findings sit next to each other in its press release: 40% of companies now let AI execute decisions without human approval, at least to a moderate degree, and 23% say accountability remains unclear when an AI system produces a negative outcome.

The survey does not say whether these are the same companies. Read together, they still raise a fair question for any organisation: when AI is allowed to act, has anyone decided who answers for what it does?

The work is changing faster than who decides

The same Deloitte release reports that 60% of companies have redesigned jobs around AI to at least a moderate degree, and that only 15% report a significant shift in decision-making authority as a result. The two figures use different thresholds, so they do not compare exactly. They do suggest that tasks are being redesigned more widely than authority is.

That gap matters when something goes wrong. The question "who owns this?" lands on a reporting structure that may not have been updated for the AI's part in the work.

Three questions to settle first

If your team wants AI to act without a person approving each step, write down three answers before you switch anything on.

1. What exactly may it decide alone? Name the decision, the limit and the exception. For example: it may approve refunds under a set amount, for orders placed in the last 30 days, and passes everything else to a person.

2. Who owns the outcome? Name one person. This is the person who gets the call when the AI gets it wrong, and who can change the rules it works by.

3. How will you find out it went wrong? Decide what gets checked after the fact: a sample of decisions each week, every decision above a threshold, or every complaint. If nobody reviews anything, the first sign of a problem is likely to come from a customer.

Settling them needs someone to write the answers down and keep them current as the AI is given more to do.

The best example of this is from the time I set up Jira for my team to automate the entire process of communication within the team for editorial content. It was necessary to know who owned the system, because when it broke (not if), the teams knew who to reach out to. It broke. They called me. This saved them time as their experience (as I found out later) with other automations was there was someone who built it, but no one owned it, and it led to delays, issues, and therefore a drop in adoption. With my system and my accountability, the team got more efficient: 130% more efficient in 6 months! While this is not an AI story, this certainly relates to the accountability.

I made a related point in Using AI is not the same as running it: the value of AI depends on the arrangements around the tool. Human oversight is one of those arrangements, and it is part of how I approach AI enablement.

Think of one task in your organisation that AI now does without a person approving each step. Could you name, today, the one person who answers for it?

Send feedback

Goes to my inbox. Nothing is published.

All posts