How a Two-Person Team Closes 1,000 Enterprise Deals a Month: Tanay of WhisperFlow on Building an AI Sales Machine

When Think School's Ganesh Prasad first heard that WhisperFlow signs roughly 1,000 enterprise clients a month with a two-person sales team, he assumed it was clickbait. It wasn't. In a wide-ranging conversation, Tanay — CEO and co-founder of Whisper — opened up his entire sales system, demonstrated the tools live, and made the case that almost any repeatable sales task can now be automated if you approach it the right way.

Everything below reflects what was said in the episode. The metrics and claims belong to the speakers.

The Numbers, and the Five-Step Chain Behind Them

Prasad sets the stage with WhisperFlow's stats: a voice-to-text AI that polishes natural speech into ready-to-send writing, launched in March 2025, now with 2.5 million downloads, 270 of the Fortune 500 as users — Reid Hoffman and Steven Bartlett among them — and a $2 billion valuation. The interesting part isn't the numbers, it's the machine underneath.

Tanay starts by breaking conventional B2B sales into five stages: outreach, discovery, conversion, onboarding, and post-sales. Outreach is finding the right buyer and the right message. Discovery is learning the customer so deeply that "instead of selling them a product, you're solving their problem" — which he calls the job of the best salesperson. Conversion is the signed contract. Onboarding is getting the client to real value. Post-sales is support and upsell. His warning: the process is intensely people-driven, and if any single link breaks, the whole chain goes down. Onboarding, he notes, is where companies routinely lose 20–40% of clients to churn, simply because they hand over a manual and walk away.

The Customer Discovery Engine

If there's one thing to build first, Tanay says, it's a "customer discovery engine." He demonstrates it on a tool called hex.tech: a single analytics system wired into every data source — their own users, usage analytics, every marketing channel, and even sales and support conversations — and powered by an LLM. Anyone at the company can ask it a plain-language question like "find all power users who are VP of Engineering at companies between 500 and 5,000 people, sorted by company size," and get an accurate answer in minutes.

The strategy that feeds it is product-led growth: become consumer-facing first, gather thousands of individual users, then draw a web outward to see which companies those users sit inside and who the most influential users are within each. He adds a sharp tactic — de-anonymizing website visitors, so you learn that, say, six people from a target company browsed your pricing page today.

Sell to the Person Who Holds the Budget

Rather than cold-calling 100 random prospects and hoping two respond, Tanay targets whoever actually controls the purchase. For engineering tools, that's the VP of Engineering — and since engineers already love Whisper, a curated list of heavy-usage VPs converts at a dramatically higher rate. He defines "power user" concretely (more than 20,000 words dictated, for instance) so the system surfaces the strongest candidates automatically.

What's striking is his claim about who can do this. Asked whether it takes a veteran who has met dozens of VPs of Engineering, Tanay says the opposite: he had never sold to anyone until six months earlier. His method is hypothesis-driven. Form a clear guess ("engineers love the product, so their VPs probably buy"), talk to a few of those people as a blank slate purely to understand them — not to sell — and the right approach becomes obvious. All you need, he argues, is curiosity and the discipline to frame everything as a testable hypothesis.

Every Persona Is a Different Game

A message that lands with one buyer flops with another. Tanay learned that engineers hate sales calls and respond best to a direct, relevant email that lets them explore on their own terms. Marketing buyers are different again — a five-person marketing team behaves nothing like a 500-person one, and a conservative bank's CMO is worlds apart from a creative consumer brand like CRED. The takeaway is to narrow down to a persona specific enough to message at scale, then rebuild the proposition from scratch each time you move to a new segment.

This is exactly where he draws the line between humans and machines. Once you understand a persona, the repetitive messaging can be fully automated — "there's no person involved in this loop at all." But discovering a new persona has to be done by a human, and specifically one with high emotional intelligence, not a pushy closer. Tanay describes his hiring test: five minutes into a mock pitch, he deliberately acts disengaged and checks whether the candidate notices and pulls him back in. Ninety percent, he says, failed to even realize he'd checked out.

A Live Build: Turning Claude Into a Sponsorship Engine

For much of the episode, Tanay builds a real system on the spot — an outbound engine to find YouTube sponsors for Think School — using Claude. A single detailed prompt spins up multiple sub-agents in research mode: one categorizes the last 50 videos, another finds current and similar sponsors, another profiles the audience, and a final one matches companies to video types. He walks through targeting Razorpay, whose audience of Indian business builders overlaps almost perfectly with Think School's, and whose looming IPO creates a "now-or-never" urgency for brand reach. The system even helps identify the right point of contact — distinguishing an SME-business leader who'd act on the pitch from a senior brand thinker who'd let it get lost.

His automation advice is methodical. Ask the model to explain every line it writes, so it catches its own mistakes and teaches you to build "taste." Run the workflow manually a few times first. Then distill everything into a single instruction file — a "Claude.md" — that captures all the context so future runs don't start from zero. To turn it into a repeatable app, hand it to an engineer or use a no-code platform like Emergent (he recommends a powerful model such as Opus), and build in human checkpoints that you only remove once you trust the output.

Treat It Like Software, Not Magic

Perhaps his most important message is a reality check. Automating a meaningful vertical takes about a month the first time — not an afternoon — and drops to roughly a week once you've done it a few times. Deploy in phases, verify the output for the first several runs, and go fully autonomous only after your checks consistently pass. Feed the system your own documentation, too, so it doesn't burn tokens rediscovering what you already know; if you've worked with 200 sponsors over four years, that history is gold.

Conversion, he admits, isn't automatable today. But he sees a near future where a company's voice agent negotiates directly with a prospect's voice agent — which is why, he insists, the moment to start collecting your conversations, emails, and learnings is now. For anyone hiring to lead this work, he looks for hands-on AI experience, personal projects that signal ownership, and a computer science background. The throughline of the whole conversation: AI doesn't replace the thinking, it scales it — and the companies that win won't be the ones using the most tools, but the ones building the smartest systems.

---

Originally published on Think School. Watch the full episode: https://www.youtube.com/watch?v=-eDZpzT8WJM