Three California community colleges are spending up to $500,000 per year on AI chatbots. The chatbots can't correctly name their own college president.
This is what happens when institutions buy AI because they feel like they should, not because they've identified a specific problem it solves well.
The chatbots were supposed to help students with financial aid and admissions questions. They handle generic questions fine. Specific questions? Not so much. And student questions are almost always specific. "What's the deadline for my particular situation?" "Does my transfer credit count for this requirement?" "Who do I talk to about this specific financial aid issue?"
Generic AI fails at specific problems. That's not a technology limitation that will get fixed with better models. It's a deployment problem. These chatbots weren't integrated deeply enough with the colleges' actual systems, databases, and policies to give accurate specific answers.
I've seen this exact pattern at companies. Someone buys an AI tool, points it at some documentation, and expects it to answer employee questions. It works great in the demo (which uses generic questions) and falls apart in production (where every question is specific and contextual).
The fix isn't "don't use AI." The fix is "do the integration work." Connect the AI to real systems. Feed it actual data. Test it against real questions from real users. Budget for the deployment, not just the license.
The $500K these colleges spent probably went mostly to the chatbot vendor. The actual deployment work, connecting it to student records, financial aid databases, and institutional policies, probably got under-resourced.
Good AI deployment is 20% technology and 80% integration. Most buyers flip that ratio.