Most teams don't abandon AI support bots because the AI "isn't smart enough." They abandon them because they never defined the boundaries.

Here's the pattern we keep seeing:

What AI should never touch:

• Refunds and billing disputes

• Security or account-compromise reports

• Anything that smells like a churn threat

What AI should always own:

• Password resets

• "Where do I find my invoice / CSV export"

• Plan and pricing questions

• Basic how-to and setup questions

The teams that succeed with AI support don't ask "can the AI handle this?" — they ask "should the AI handle this?" and draw a hard line up front. Everything in the long tail (the boring, repetitive, high-volume stuff) goes to AI. Everything with financial or trust stakes goes to a human, every time.

If your AI bot is failing, it's rarely a model problem. It's almost always a missing escalation policy.

Curious how others here are drawing that line — what's on your "never automate" list?