Two clay hands, one dark green and one sage, holding a single copper thread between their fingertips

Most of a salesperson’s week never touches selling. The rest goes to logging, drafting, scheduling, chasing, summarising. That's the part the model takes off the salesperson's plate first.

HubSpot teams report 2 hours per salesperson per week back when the model carries the thread.

The more interesting part is what the model does inside the conversation.

What the Research Shows

Anthropic published work on LLM persuasion. With personalisation, LLMs are now about 80% more persuasive than the average human responder on common B2C tasks. With access to the prospect's profile, that number rises past 81%.

That looks like a case for replacing salespeople. The same research shows generic LLMs are bad at the parts of selling that matter. They're too agreeable to negotiate: they concede when the buyer pushes. They don't know when to hold a price or escalate to a human.

What Separates the Model From a Generic LLM

  • Memory: the buyer's last 12 messages are live context.
  • Timing: the model reads the pause, the silence, the second tab, and acts when the thread moves.
  • Judgment: when to push, when to hold, when to hand off.
  • Ground truth: the model arrives trained on the conversations that closed in the buyer's category, instead of the open web.

The Economics That Follow

  • Same headcount, more closed deals. We've seen 3× in 90 days at Libra.
  • Higher reply rates on outbound. The buyer is no longer deciding "is this real?"
  • Salespeople spend their hour on the calls that need a human voice.

What Doesn't Change

High-consideration buyers still want to be seen, heard, guided. The model's job is to read like a person who is paying attention.

Over the next 2 years, the teams that do well will let the model carry the thread. The salesperson stays on the relationship.

Written by Sellence. The team building the persuasion model for financial services.