AI buys for you… and sometimes a little too creatively

Several companies tested automatic ordering through AI, whose task was to independently evaluate needs, place operational orders, and maintain optimal stock levels. The result often exceeded expectations — unfortunately, in a negative sense.

Let’s first take a look at the most amusing blunders.

⚠️ In a smaller company, the algorithm ordered 27,000 coffee cups because it “correctly deduced that consumption grows exponentially during meetings, and if everyone kept meeting for just 7 days in a row, we would run out of cups.”

🧻 In one logistics company, the system mistakenly ordered 4 truckloads of hygiene products instead of four packs. Of course, it was just a minor unit error in the ERP system. The head of procurement commented on the situation by saying: “At least we wouldn’t have a problem, even if everyone wanted to take a s** at the same time.”*

🤖 A chatbot as a negotiator proved itself. It managed to “bargain” with suppliers for 15% lower prices. Unfortunately, it also turned out that experienced suppliers quickly spotted the algorithm’s weaknesses and systematically negotiated higher prices — all they had to do was look sad, claim that the requested price was below cost, and insist that the chatbot had to behave fairly.

💻 The IT department created a fake supplier to test whether the newly implemented SRM system with many safeguards and automated monitoring could detect fraud. Unfortunately, the fake supplier managed to deliver an imaginary product within the system in the required quality and on time, the invoice passed through control and payment, and the system even awarded it 4.8 stars. All participants were surely delighted that Internal Audit turned this into a case study.

🌡️ It is becoming increasingly urgent that companies carefully consider the balance of responsibility between humans and artificial intelligence. When does a human do the work alone? When does a human manage the task and AI serves as an efficient assistant? When do they collaborate and split tasks? When does AI do the work and a human supervises? And when can AI work fully autonomously? These are key questions for the successful implementation of AI in companies.