AI in Practice: From Demos to Production

The most common question this year has been what AI can do for us. We participated in several attempts, from document Q&A to ticket triage. The distance between demo and production turns out to be several hurdles.

Real-world thresholds

Data quality: internal documents are messy and stale, so retrieval often returns answers from three years ago.

Cost of errors differs. A wrong sentence in chat is harmless; a wrong quote, contract clause or support reply is not.

Habits: employees do not switch because a tool is advanced, only when it clearly saves time.

What has worked

Internal knowledge search: staff looking up policies and processes, where tolerance is high and gains direct.

First drafts of repetitive copy: marketing and multilingual product descriptions drafted by AI, finalised by people.

Assisted log and ticket classification: machine first, human confirmation, with measurable throughput gains.

Adoption depends less on model capability than on scenario choice and organisational readiness. High tolerance, high repetition, clear payoff beats talk of frontiers.