Christine Harrington warns AI scripts fail

Christine Harrington’s point was not that CRM hygiene or AI drafting are useless. It was that neither one closes a deal by itself. In her coaching example, the rep had the visible mechanics right — clean Salesforce records, polished AI-generated follow-ups, consistent activity — and still had not closed business in 11 weeks. The gap, as Harrington framed it, was in the parts of selling that do not reduce neatly to a sequence: earning trust, handling objections in real time, and reading what a buyer is actually signaling. That distinction matters because a lot of sales automation is sold on the idea that better output equals better outcomes. If the notes are complete, the emails are grammatically strong, and the cadence runs on time, the workflow can look healthy from the manager’s seat. Harrington’s anecdote pushes against that. A rep can appear operationally excellent while still failing at the interpersonal work that moves a deal forward. The system may show activity; the buyer may still feel unconvinced. Her warning also lands in a specific place in the current sales-tech cycle. Many teams now use AI to draft follow-ups, summarize calls, suggest next steps, and keep CRM fields updated. Those are legitimate productivity gains when they reduce admin burden or help a rep prepare. The risk starts when teams treat generated language as a substitute for judgment. A follow-up can be timely and well written, yet miss the real objection. A sequence can be personalized on paper, yet fail to show the rep understands the account, the politics, or the buying risk. The phrase “human in the loop” can sound abstract until it is attached to a failed pipeline. In practice, Harrington is describing a narrow but important boundary: AI can support customer-facing work, but a person still has to decide what matters in the conversation. That includes when to deviate from the script, when to press, when to slow down, and when the buyer is asking for reassurance rather than information. Those calls are hard to automate because they depend on context, timing, and credibility. Her example is also a reminder that sales leaders can over-measure the wrong things. Salesforce completeness, follow-up speed, and sequence adherence are easy to inspect. Trust-building is not. Objection handling is often only visible in fragments — a call recording, a reply, a stalled next step. That can create a management bias toward what software can count. Harrington’s story suggests the opposite discipline: use automation to make the rep more prepared, not to assume the relationship work has been handled. For operators building AI-assisted sales workflows, the practical takeaway is restraint. The strongest uses of automation are the ones that structure information, reduce manual entry, and tee up a better human conversation. The weaker uses are the ones that try to replace that conversation with a script that only sounds attentive. Harrington’s anecdote is small, but it captures a broader design principle: in customer-facing sales work, polished output is not the same thing as persuasion.