Skip to content
AI

LLMs for customer service — when they help, when they hurt

In 2024 three different clients asked us to add an LLM to their stack. Two succeeded; the third failed badly. The difference wasn't the model — it was the shape of the problem.

LLMs are excellent when the question can take many forms and the answer has room for interpretation. They're poor when the question is specific and the answer must be exactly correct.

JOIN THE CONVERSATION