Skip to main content
All articles
AIAutomation

Before you add AI to a task, count the checking it creates

Julien Côte3 min read
Drawn cover: "Writing saved. Checking added." above two rows for benefit appeals: AI writing the decision letter, where staff took longer, and AI finding the policy and past cases, where review time dropped.

There's a lot of pressure to add AI to everything at the moment. A thoughtful article in Smashing Magazine this month, by UX researcher Victor Yocco, asks a better question first: does it actually save the person doing the job any effort, once you count the checking?

A story about a feature that made work slower

His opening example is from a large health insurer. The team added AI that wrote a full decision letter for staff who handle benefit appeals. It worked as designed. But the staff stopped using it, and the ones who kept using it took longer on each case. Reading an AI-written decision carefully enough to sign it took about as long as writing one.

The second version left the writing to the staff and used AI to find the relevant policy and past decisions for each case instead. Review time dropped.

"The technology is identical in both versions, and the difference is which part of the task the AI handles."

The checking doesn't disappear

One line in the article is worth pinning above any AI project:

"When AI fills in a field, the user takes on checking it."

If a task already takes a few seconds and almost no thought, handing it to AI can make it harder, because someone now has to check what the machine did. The good candidates are tasks where the thinking is heavy, or where AI catches things a person would miss.

Can your team spot the mistakes?

"Every AI system makes mistakes. The question is whether your user will notice."

Some mistakes are obvious on screen. Others look perfectly reasonable, like opening hours that are slightly wrong. The article cites research on "automation bias": people tend to trust automated output even when it's wrong, and stop checking it closely.

What this means for your enquiries

This is how we decide what to automate for clients. Sorting new enquiries by what the person needs, and asking the questions your team would ask, usually passes the test: it removes real work, and a person still makes the decisions. Writing every reply word for word often doesn't, because someone has to read each one carefully anyway.

Copying an enquiry from your inbox into your CRM is a good example of work that passes. Our short video The job nobody joined your firm to do shows how that works, with a person on your team still checking and sending.

So before adding AI to any task, ask three things. Does it remove work, after you count the checking? Is it better than how your team does it now? And would your team notice when it gets something wrong?

If you'd like to see how this would work for your enquiries, send us your website address and we'll record a free video walkthrough of what we find.

Source: Should You Actually Add AI To That?, Smashing Magazine (Victor Yocco), November 2026

Written by

Julien CôteCôTEDESIGNS

Losing enquiries after hours?

AI enquiry handling starts from £1,500. It answers new enquiries as they arrive, asks the questions your team would ask and passes each one on sorted, with a person checking anything that matters. Tell us how enquiries reach you now and we'll do a free review of what could be automated.