AI was given a "sales territory" - how did it do?
- John Hsieh
- Nov 21, 2024
- 1 min read
Updated: Jul 9
Anthropic gave Claude a “sales territory”. What happened next says a lot about AI’s ability to handle the nuances of selling to real people. As part of a recent experiment, Anthropic put Claude in charge of a vending machine business.
It had everything a typical rep might need to succeed:
Enablement on sales goals and process (in the form of a detailed AI Prompt)
Dedicated territory (via the vending machine)
Way to prospect and engage with buyers (via slack)
Customer success team to drive success (with an Ops team named “Andon Labs”)
But what it couldn’t provide AI was the unpredictable nuance of selling to real people. That’s where things got interesting.
For example, Claude ignored a clear buying signal for a niche soda request that would’ve delivered an 85% profit margin!
When pressed on price, it offered unnecessary discounts without understanding buyer intent.
And when sales stalled, it didn’t adapt. It just repeated the same logic, expecting different results.
How did it do at the end of the experiment? As you can see from the chart below, the mistakes Claude made were costly!
Turns out AI, just like many reps, struggles without coaching, feedback, and the ability to navigate the unpredictable nuances of real-world selling. It’s a reminder that tools alone aren’t enough to drive performance in sales. Not for machines. Not for humans.
🙌 Huge credit to Anthropic for publishing this kind of research.
It’s refreshing to see someone push the limits of AI in the wild instead of just adding to the endless hype.
Link to the full research paper: