Came across an ai implementation failure story on reddit yesterday. Company spent £120k on a chatbot that keeps hallucinating. Screenshots show the bot telling customers to restart their computer for billing questions and suggesting they "clear browser cache" to cancel subscriptions. That's not hallucination. That's an AI trained on a support doc mess mixed with custom support info. If you hired someone to go through that, it would probably look something like this: Proper knowledge base audit and cleanup: 40-50 hours. £7,200-£9,000. Retraining on clean, structured documentation: 10-15 hours. £1,800-£2,700. Instead of thinking through the problem and improving their communication, they'll probably throw another £80k and fail in exactly the same way.
xmetrix
1mo
And for what? Not to reduce costs... but to join the ai hype bullshit. #aiformetookthx lol
Leora Knight
1mo
I HATE AI in customer service. A banking app bot once told me that I can sign up for a Dutch account with a UK passport.... No.
Ryan Jose Brosas
1mo
This is why I always tell clients to fix their documentation before touching AI. Garbage in, garbage out. The real cost isn't the £120k they spent, it's the customer trust they're losing every day. Clean data beats fancy models every single time.
Data quality and proper scoping aren’t optiona, they’re the foundation. Rush AI deployment to chase hype, skip the fundamentals, and you’ll get exactly what you paid for: garbage. Any serious partner qualifies requirements of all aspects first, because shortcuts in discovery always surface as failures in delivery.
Lewis Cowles
1mo
someone put in a 10k bid to fix... Look 100k is a tiny project, they should spend whatever to turn the thing off and delete it if this is the ceiling, or go on to burn another few 100k making it better.
Raphael Cohn
1mo
The old adage is still true: Garbage In Garbage Out. If your data is poor your results will be worse.
experienced software polyglot back on the market for ethical work that makes a real difference.
1moOr, or, revolutionary idea, have people who know how to troubleshoot doing the actual work instead of wasting money on AI that will never work, no matter how much it is "tuned and trained".