AI Just Broke Math. Your Sales Process Is Next.

TL;DR

What can AI do to help fix a broken outbound sales process?

AI tools have no ego attached to your existing process. You can feed sequence performance data into a model and prompt it to find the assumption the data disproves, which gets you the honest read your internal team might be avoiding.

Why do sales teams keep running outbound sequences that stopped working?

The person who built the sequence is usually still on the team, or a VP already sold leadership on the approach. That attachment to the original decision makes it hard to see what the data is actually showing.

How do you find the broken assumption in your outbound motion?

Identify the one belief your outbound motion can't survive being wrong about, then design a test where the null hypothesis is that you've been wrong the whole time. Run it for 30 days and look at the results without explaining them away.

An OpenAI model just disproved a geometry conjecture that had stood unchallenged for years. Mathematicians weren't lazy. They were just too close to it. Sound familiar?

I've watched founders run the same broken outbound sequence for eighteen months because it 'used to work'. The sequence stopped converting around month four. But nobody pulled the thread because the motion felt proven. It had history. It had effort behind it. Effort is not the same as evidence.

The Conjecture You're Running On

Every outbound team has a conjecture at its center. Something like: 'our ICP responds to case study emails' or 'Tuesday morning sends get the best reply rates' or 'a five-touch sequence is enough'. These aren't strategies. They're beliefs that calcified into process.

The problem isn't that the belief was wrong when you formed it. The problem is that you stopped testing it. You built a whole motion on top of it. You hired around it. You reported pipeline numbers to your board based on it.

And now you're defending it in weekly standups instead of questioning it.

What AI Actually Showed Us Here

The geometry story isn't really about AI being smart. It's about what happens when you bring a system into a domain that has no ego invested in the existing answer. The model didn't care that the conjecture was old. It didn't care that credentialed people believed it. It just ran the problem and found a counterexample.

That's the thing about counterexamples. You only need one. One data point that breaks the rule is enough to collapse the whole assumption.

I had a moment like that in 2022. We were convinced that our best reply rates came from highly personalized first lines. We spent real money on researchers writing custom openers. Then we A/B tested a dead-simple problem-statement email with zero personalization. It outperformed our 'premium' sequence by 34%. One counterexample. Whole playbook changed.

The Ego Problem in Outbound

Here's why most teams don't find their counterexample. The person who built the sequence is still on the team. Or the VP sold the board on the approach six months ago. Or the SDR who runs it is your top performer and you don't want to mess with their confidence.

Ego doesn't mean arrogance. It means attachment. And attachment kills your ability to see what's actually happening in the data.

AI tools don't have this problem. That's genuinely useful. Not because AI writes better emails (it often doesn't). But because a well-prompted model will look at your sequence data and tell you the thing your RevOps lead is tiptoeing around.

I've started running our sequence performance data through models with a simple prompt: 'Find me the assumption this data disproves.' The outputs aren't always right. But they're always worth reading.

What to Actually Do With This

Pick the one belief your outbound motion can't survive being wrong about. The load-bearing conjecture. For most teams it's something about sequence length, persona targeting, or channel mix.

Then design a test that could actually disprove it. Not a test designed to confirm it with slightly different copy. A real test where the null hypothesis is 'we've been wrong about this the whole time'.

Run it for thirty days. Look at the data without explaining it away. If it breaks your conjecture, good. That's the most valuable thing that happened to your pipeline this quarter.

The mathematicians who accepted that geometry conjecture weren't bad at math. They just needed something with no stake in the outcome to look at the problem fresh. You can do that yourself if you're willing to be wrong about something you've been confident about for a long time.

What's the conjecture your outbound motion is built on? And when did you last actually try to break it?