Ariel Cohen
Key Views & Dialogues
20VC: From $6.2BN Market Cap to $2.8BN: What Is Not Translating About Navan’s Public Story | Are Any Public Company CEOs Actually Happy? | Why Navan Built It’s Own Customer Service AI and What it Could Mean For Customer Service AI with Ariel Cohen
- 🗓️ Date：
2026-02-07｜ 🎙️ Show：20VC
$6.2B IPO market cap fell to $2.2B, with Cohen arguing investors mistake Navan’s consumption model for SaaS as go-to-market costs precede returns. Cognition and Ava handled 55% of chats during the airport shutdown weekend, showing complex vertical AI requires zero hallucinations; unreleased Navan Edge and travel infrastructure are the catalysts and moat to monitor.
View Dialogue Notes & Key Takeaways
Navan went public last month at a $6.2B market cap and now has a $2.2B market cap. Ariel Cohen’s defense of the timing: the post-COVID capital structure was one driver, but a payments business “may end up not where you want to be” if it stays private too long, and enterprise customers “want to know that you’ll be around.” He started non-deal roadshows in April the same day the tariffs were announced — his conclusion is that timing a market is really, really hard; you follow through and run a good business.
The market’s two mispricings, per Cohen: Navan looks like SaaS but is a consumption business — go-to-market cost and commissions hit the P&L today while the payoff arrives “in the years to come,” so investors see spend without the return (he claims “extremely efficient” GTM and churn without sharing numbers). Second, the blanket “everything software is dead” mood leaves “not a lot of patience to see the differences between the companies.”
The current Navan AI-platform push began in a panic: roughly four years ago Cohen saw Sam Altman demo ChatGPT pre-release at a Napa conference, called his co-founder Ilan and said “If we will not build our own platform right now, we are so dead.” That platform (Cognition + the Ava agent) now handles the hardest support case in travel — during the JFK/NY airport shutdown weekend, Ava ran 55% of chats and worst-case call wait was 16 minutes.
On build-vs-buy for customer-service AI, Cohen is categorical: neither Sierra-tier vendors nor raw Anthropic/OpenAI APIs are “even remotely close” to what a complex regulated vertical needs, because zero hallucination tolerance is the spec — send a traveler to the wrong gate and “you’re gonna also get a lawsuit.” The read-through for vertical AI investors: the cute LLM booking demos are “not even remotely close to being there.”
Harry’s Salesforce pushback is the best exchange: distribution is all that matters, so an AI sales rep plus Salesforce’s reach makes it “a screaming buy.” Cohen’s counter — “I’ve never met a salesperson that told me that they like to use Salesforce” — and if users hate you, distribution only delays disruption, as he saw selling HP software (“Walmart to sell servers”). His proof that user love is the moat: in Navan’s history it lost six enterprise customers, and five of them came back.
Cohen calls LLM infrastructure “commoditized” and “not an important discussion” — Harry counters that Harvey and Lagora both flipped from diehard OpenAI to diehard Anthropic on Opus 4.5 quality, the opposite of commoditization. Cohen’s honest non-answer: “I actually don’t have a clue” how an investor picks — Navan’s Cognition routes between models and many Edge transactions are with Google, while Anthropic’s dev-infrastructure lead looks “similar to what Microsoft did in the ’90s.”
The vibe-coding data point worth trading on: co-founder Ilan vibe-coded Navan’s expense product in six hours over a weekend, and ERP integrations to NetSuite “are not [hard]. They are not.” Cohen’s implied map: app-layer software and integrations are now easier to vibe-code, while travel’s licensed, fragmented plumbing is “almost impossible to disrupt” — which redraws where moats (and engineering budgets) live.
On Jason Lemkin’s line that no public-company CEO is happy: Cohen concedes share price “one hundred percent” correlates with employee morale and that staff refresh the stock app all day; his job is pointing them to the two-, three-, ten-year frame, Bezos-letter style. He checks the price twice a day, can’t sell anyway, and notes the same resilience test as COVID — when employees left for Zoom at its peak, he asked, “do you think more people than the people that live on this planet will use Zoom?”