Craig Stoss and Jen McCorkle have never met. One runs partner solutions at a support-tech vendor; the other has spent almost thirty years in data and analytics. Two separate Supportman Live Chat episodes, recorded months apart, landed on the identical rule for where AI's job in support actually ends.
Pattern recognition, nothing more
Stoss's episode is about the unglamorous work of actually using support data — correlating a CRM, a help desk, product analytics, and an order system that each hold one piece of the same story. His read on what the AI in that stack is actually doing:
Both guests draw the same line: AI's job ends at the handoff. Supportman's AI QA reviews 100% of closed Intercom conversations and still routes anything uncertain to a human — the pattern-matching does the first pass, a person makes the call.
Free trial · no card · about 2 minutes to set up
"Current AI is very good pattern recognition, nothing more. If the noise has a pattern you can teach it, it can filter it. Most noise doesn't."
— Craig Stoss, VP of Partner Solutions at Kodif
That's a narrower claim than most AI pitches make, and Stoss means it narrowly. Teach the model a pattern and it filters that pattern. It doesn't develop judgment about the cases nobody has seen yet — which, in support, is most of the interesting ones.
Always give people a way out to a human
Jen McCorkle's episode is a governance checklist for scaling AI into a support org — what to ask a vendor before you buy, and the sentiment-analysis failure mode where "cancel" scores negative even when cancelling a technician visit meant the agent had just fixed the problem. Her line on where automation has to stop:
"Don't let a bot fake empathy. Customers can tell, and it costs you the credibility you want the human to have. Always give people a way out to a human."
— Jen McCorkle, data-driven leadership expert
Same conclusion from the opposite direction. Stoss is describing what the model is capable of; McCorkle is describing what a customer needs to feel, regardless of what the model is capable of. Both put the exit ramp in the same place.
Where the line actually sits
Neither guest is anti-AI — Stoss's episode is a crawl-walk-run adoption plan, and McCorkle spends half of hers untangling the acronyms so leaders stop being intimidated by them. The line they draw isn't "AI versus human," it's a handoff: pattern-matching does the first pass, and a person makes the call on anything the pattern doesn't cover.
That's the same design decision behind Supportman's AI QA. It reviews 100% of closed Intercom conversations instead of the 1–2% a manual sample can reach — but it's still scoring against a rubric a person wrote, and anything uncertain routes to a human reviewer rather than getting silently marked "fine." The pattern-recognition does the volume; the handoff is still where the trust gets built.