Yellow.ai Is Paid by the Minute to Kill the Phone Call
Yellow.ai bills by the minute of phone call it handles, and spends its R&D budget making those calls unnecessary. Its CEO explains why that isn't a problem yet.
On this page
Yellow.ai charges its customers by the minute of phone call its software handles. It also builds machine learning designed to stop those phone calls happening. Latka spotted the tension and asked the question directly.
If I’m paying you to handle telephony for me and I’m paying you per minute of call handling, but you’re also trying to use AI machine learning to prevent the call in the first place — don’t you sort of compete with yourself there?
Nathan Latka, interviewing Raghu Ravinutala
“Oh, absolutely, Nathan.”
The answer is a market-size argument, not a pricing one. There are roughly 400 billion phone calls made worldwide every year, and by Ravinutala’s reckoning fewer than 0.1 percent of them have any automation at all. Cannibalising your own voice revenue only matters if you are near the edge of the market.
The full answer
Ravinutala did not soften it. “The overall objective of the company is to automate the majority of the available interactions at the company. And we clearly believe that digital interactions are superior to voice interactions in many cases. We recommend customers to do that, though it cannibalises a potentially higher revenue in voice handling.”
Then the reason he can afford to: “There are 400 billion calls made every single year in the world, and today less than 0.1 percent of them have any kind of automation. So it’ll not disappear in a while… we as a company don’t need to worry about one product cannibalising. That’s thinking too narrow at this point.”
The pricing that creates the tension is genuinely usage-linked rather than seat-based. “Our pricing is directly linked to the number of conversations or interactions that are automated. So the more phone calls that our virtual assistant answers, the subscription price goes up pretty linearly.” Voice averages $0.50 to $1.00 per minute of call handling, varying by country; WhatsApp and Google Business Messaging are priced per conversation session.
Two companies, in two geographies
Yellow.ai runs two distinct go-to-market motions, and the gap between what they land at is three- to sixfold.
Lands customers at $30,000 to $40,000 in annual recurring revenue. Over 90 percent of the customer count sits here.
Lands at $120,000 to $250,000, averaging around $130,000. A minority of logos, and the faster-expanding half.
Asia Pacific still produces 70 to 80 percent of ARR despite the price gap, simply on volume. But the expansion curve in the US is what Ravinutala is excited about: “Some of our customers landed at $30,000 or $40,000 in the US market and grew to $600,000, $700,000.”
The first US customer came inbound
The most transferable part of the interview is how an India-based company entered the United States, because the answer is that it did almost nothing.
- The first US customer was inbound. “We got an inbound based on some of our press releases. It was actually closed by a rep sitting out of India.”
- Reach half a million in ARR before hiring anyone locally. “We landed our first customer when we didn’t have any person in the US market. By the time we hired a person in the US, we had a few customers and they had some references to build on.”
- Hire senior first, not junior. The first US hire was an SVP of Sales — not an SDR, not a rep.
- Have them build the function and carry a bag. Early months were “figuring out how to create a demand gen function… making a marketing hire, setting up an SDR function, hiring the initial sales reps — and of course doing player-coach, handling the leads and meetings coming in while doing so.”
Within a year that team went from one person to 28 or 30 in North America.
The range he would give
Asked for revenue, Ravinutala declined to name a figure but handed over a band and invited Latka to pick from it: “I’m not sharing the broad revenue metrics. I’ll give a very broad range so you can pick a number. We’re somewhere in the twenties to thirties — 25 to 35 is where you can take a broad range.”
Growth over the prior twelve months: 140 to 150 percent. Target for the year: “40 to 60 is what you can broadly assume that we are targeting.”
The GetLatka profile records $30 million for August 2022 — the middle of the range he gave — and correctly carries it as an estimate rather than a disclosure.
Net revenue retention runs 150 to 160 percent, which Latka called world class. Ravinutala immediately qualified what that does and does not buy you, which is the most numerate thing anyone says on the tape.
Just to give calculations, if someone were at $8 million to $10 million, 150 percent NRR will only take them to $14 or $15 million. You still build new revenue on top of it.
Raghu Ravinutala, founder and CEO, Yellow.ai
Profitable at a million, then deliberately not
Yellow.ai bootstrapped to roughly $1 million in ARR and was profitable getting there. It took two and a half to three years, and Ravinutala remembers it as “the toughest, toughest part I ever had.” Only then did they raise: a $4 million Series A in 2019 (“we raised it in India, so in India it was a Series A”), $20 million Series B in 2020, $78 million Series C in 2021. Around $104 million total.
Asked what makes the business capital-intensive, he gave the honest answer rather than the technical one.
“This is a market that we are creating — I don’t think there was a lot of market that was existing, and we were going behind very fast growth rates. This could be a completely profitable company if we were taking lower growth rates. We were profitable when we were at a million dollars. Much of the capital burn happens through sales and marketing, where you invest upfront in capacity and invest in getting into newer geographies to drive those growth rates a little bit ahead of the curve.”
By August 2022, with valuations compressing, more than 90 percent of the Series C was still in the bank — enough, he said, for two to three years. The response to the downturn was three specific dials rather than a headcount cut:
- Sales productivity — improve the numbers rather than adding capacity.
- Fewer geographies at once — “limit the number of geographies that we are trying to expand at the same time. You want to go by one.”
- Conviction over experiments — “a lot more focus on the current feature set rather than a lot of experiments. A lot of investment behind high-conviction areas, and tuning down investments in experimental areas.”
Where the engineers are
The team was about 850 people, roughly 300 of them engineers — and 99 percent of those engineers are in India, headquartered in Bangalore. That ratio, on an enterprise product selling six-figure contracts into North America, is a large part of why the unit economics work at all.
On valuation, Latka reasoned out around $700 million from the Series C. Ravinutala trimmed it: “It’s not exactly 10 percent, it’s a little bit more than that. So the valuation you would expect is not at 700 — a little lesser than that.” The profile records $500 million, flagged as an estimate.
The first customer, incidentally, was Asia’s largest paints company. And the models that now train on billions of conversations a day started with none at all: “When we started in 2016 it was humans that were training these models, because there weren’t any conversations — we were just starting off. It was initially seeded with a lot of manual training and labelling.”
Asked what he wished he had known at twenty, Ravinutala gave the answer of a founder who spent three years bootstrapping to a million: “Start your business, take risks much earlier in your life.”
Sources — Raghu Ravinutala interviewed by Nathan Latka, recorded 10 August 2022. Revenue, headcount, customer, valuation and funding figures from the Yellow.ai profile on GetLatka, with dates as recorded.
