Series
AI Made This Worse
AI is useful for a lot of work. Not all of it. This series looks at the times it made the job slower, less clear or riskier, and what would have worked better.
Each case is fictional, set at Fernway Group, and built from the kinds of mistake that happen in real offices. Every one follows the same shape: what the person was trying to do, what they asked the AI, what went wrong, a better approach, and one rule to remember.
- Case
The email that got longer
Sam asked AI to polish a short customer email. It came back four times longer, and the one fact the customer needed had moved to the fourth paragraph.
- Case
The source that did not exist
Leah asked AI for evidence to support an office decision. It supplied a precise figure and a report title. Neither was real, and the figure was already in a draft for the Managing Director.
- Case
Five minutes became twenty
Grace used AI to write a payment reminder she normally sends from a template. Between redacting, prompting, rewording and checking, a five-minute job took twenty.
- Case
The spreadsheet it misread
Alex asked AI for the total and the June figure from a short order export. It counted the totals row as an order and read one American-format date as May.
- Case
The complaint reply that sounded fake
Sam used AI to reply to a customer's third complaint. The draft was smooth and sorry, and the customer replied that it read like a bot. It also offered a refund no one had approved.
- Case
The owner nobody agreed
Maya asked AI to tidy up her meeting notes into actions. Where the notes had gaps, the summary filled them in, confidently and wrongly, and the summary went to the whole team.
- Case
The decision AI should not make
Tom used an approved AI tool to shortlist candidates for a sales role. It ranked them confidently, gave reasons that leaned on age, career gaps and background, and Tom couldn't explain the result to anyone.
Want to practise the fixes? Try the practice lab.