SALES JUDGMENT CHALLENGE 005
THE
OVERRIDE
An AI pricing agent recommends a maximum discount of 8% on a major renewal. The rep believes the customer will walk unless she offers more. She overrides it, offers 12%. The customer signs.
What would you do?
THE SITUATION
The deal closed. That's not proof the override was needed.
A company has introduced an AI pricing agent for its sales organization. It uses customer history, product usage, prior discounts, competitor data and current demand to recommend the maximum discount a salesperson should offer.
A salesperson is negotiating a major renewal. AI recommendation: maximum discount 8%.
She believes the customer will walk unless she offers 12%. She overrides the recommendation. She offers 12%. The customer signs.
The system can use successful overrides as feedback to improve future recommendations. Her manager has to decide whether this override should become part of that learning.
Nobody knows if 8% would also have worked.
MAKE THE CALL
What do you do?
Choose before you continue.
WATCH THE CHALLENGE
Coming November 9.
The video for this challenge publishes November 9. Check back then — or work through the situation and questions below in the meantime.
THE JUDGMENT PROBLEM
Sufficient is not the same as necessary.
This isn't a story about a salesperson being right. It's a story about a piece of evidence being mistaken for proof.
The deal closing doesn't tell you the override was necessary. It tells you 12% was sufficient. Those are not the same claim, and the gap between them is exactly where the margin was spent.
If this override is treated as a positive training signal, the system may learn that moving beyond its 8% recommendation was associated with success. But the outcome never established that the extra discount caused the success.
Nobody decided that was a valid lesson. It became one by default, because the outcome looked good and nobody asked what the outcome actually proved.
THE PROOF PROBLEM
An outcome is not evidence about the alternative.
The customer renewed at 12%. Whether they'd have renewed at 8% was never tried, and never will be.
AI can learn from outcomes perfectly. Judgment determines whether an outcome actually proves what it's being used to teach.
THE JUDGMENT DIFFERENCE
Sufficient is not the same as necessary.
The customer signed at 12%.
Would 8% have worked just as well?
A successful override — proof it was needed? It depends on evidence nobody collected.
BETTER QUESTIONS
Before this override teaches the system, ask:
What evidence would make us more confident the extra discount actually mattered?
Are we looking at one successful override, or a repeatable pattern?
What happens if we teach the system from this example and our interpretation is wrong?
This isn't about whether the salesperson was right. It's about what the outcome actually proves — and a closed deal alone can't answer that.
SO, WHAT WOULD I DO?
Each option is defensible — depending on what's at stake if I'm wrong.
A protects the salesperson's frontline read of the customer — and risks teaching the model from a conclusion nobody actually verified. B adds scrutiny, but the manager still can't see the counterfactual either. C keeps the model clean of unproven signals, at the cost of discarding frontline information that might genuinely be valuable. D waits for a pattern before drawing a lesson, but delays learning from information that may already be valuable.
This isn't about which option is correct. It's about knowing what the outcome actually proves before deciding what it should teach.
THE SALES JUDGMENT TAKEAWAY
Don't mistake a successful outcome for proof of a good decision.
Ask what the outcome actually established — not just what happened.
Know what you're teaching before you teach it.
A good outcome can support a decision without proving the decision was good.
Confusing the two is how organizations teach themselves lessons the evidence never established.
BRING THE CHALLENGE TO YOUR TEAM
Make judgment visible.
Judgment Challenges can be explored with your team through a private 90-minute Judgment Lab, turning the scenario into a practical conversation about AI feedback loops, evidence and what a good outcome actually proves.