Navi Kang

Founder-Operator | Building AI-enabled experimentation systems for GTM & growth | Northpeak

[Navi Kang @NaviKang1]

The bar (per ProductLed) - 60 seconds to first value.

The gap between 8% and 20% CVR is almost entirely a Day 1 problem.

[Navi Kang @NaviKang1]

AI-native products convert at 15-20%. Most SaaS converts at 8%.

The difference is what happens in the first session.

[Navi Kang @NaviKang1]

Why does the first hour matter this much?

ChartMogul: trial-to-paid conversions cluster around Day 7. Win or lose that week. There's almost no recovery after a bad first session.

[Navi Kang @NaviKang1]

The hardest thing in product is changing behavior.

Cursor never tried. They found where developers already lived and made it much better.

Got me wondering, which knowledge work category has a dominant workflow that somebody needs to make dramatically better?

[Navi Kang @NaviKang1]

SpaceX offered $60B for a 3-year-old coding startup. Days before they were closing a $50B funding round.

They're buying the workflow developers live in.

What's interesting about Cursor -

[Navi Kang @NaviKang1]

Now, SpaceX already has Colossus, one of the largest GPU clusters in the world. They're not short on compute.

What they want: the workflow developers live in, plus joint development of "the world's best programming and knowledge work AI."

That's a platform bet.

[Navi Kang @NaviKang1]

Fix the ugly workflow. Measure what changes. Then expand.

The real upside: you get back work you never had time for in the first place.

What's the ugliest, highest-friction workflow on your team right now?

[Navi Kang @NaviKang1]

Don't start with the transformational AI workflow. Start with the ugly one.

The one your team already hates. Slow, manual, full of copy/paste and rework.

That's where AI pays off first.

[Navi Kang @NaviKang1]

What this looks like in practice:

DoorDash rebuilt support workflows: dev time -50%, latency under 2.5s, testing capacity 50x.

The Judge Group: contract review from 6 hours to 60 minutes.

Boyd: service ticket close time 5x faster.

[Navi Kang @NaviKang1]

The goal of experimentation is not to run experiments.

It's to make better decisions with less wasted capital.

Knowing when NOT to test separates great operators from experiment addicts.

What's your 5th condition?

[Navi Kang @NaviKang1]

Stop running A/B tests. (At least in these 4 situations.)

Most optimization content tells you what to test.

Nobody tells you when NOT to.

[Navi Kang @NaviKang1]

4. You have a preference disguised as a hypothesis.

"My CEO wants to try a new headline" is not a hypothesis.

A hypothesis explains WHY the change would move the metric. If you can't articulate the mechanism, you're guessing with extra steps.