There is a familiar shape to the good customer-support-automation origin story, and Verloop.io fits it almost too neatly. In the middle of the last decade, Gaurav Singh and Piyush Mishra were running a concierge startup called MagicTiger, the kind of business where a customer texts "book me a table" and a human somewhere makes it happen. To make that happen at any scale you need people - lots of them - answering chats. Singh has said he was managing a contact operation of roughly 100 to 200 people, and the thing about 200 people answering chats is that most of what they type is the same handful of answers, over and over, forever.
So the team did what engineers do when confronted with repetition: they automated the repetitive parts. And then they noticed the automation was arguably more valuable than the concierge service it was attached to. In 2015 they spun that realization into a company. The legal entity is still called MagicLane, a fossil of the original venture, which is a nice detail: Verloop.io is a business that grew out of a workaround.
The most repetitive part of your job is usually the part somebody will eventually sell as software.
What Verloop.io sells, stripped of the category jargon, is a layer that sits between a company and its customers and handles conversations. Early on that meant a chatbot on a website or a WhatsApp number - the sort of thing that answers "where is my order" without waking up a human. Over ten years it accreted the rest of the stack around that core: live chat, IVR and call routing, ticketing, a knowledge base, real-time analytics, and something on the order of a thousand integrations so it can actually read your CRM and know who it's talking to.
What it does - and who pays for it
The business model is the boring, durable kind: B2B SaaS, sold by subscription and by usage, priced against how many conversations and calls you run through it. There is a free starter tier for companies that want to try it, and enterprise pricing for the companies that actually move the revenue. Verloop.io says it serves more than 200 businesses - a mix of unicorns and Fortune 500 companies - across India, Southeast Asia, the Middle East, and the United States. Over its history the client roster has included names like Nykaa, Livpure, Portea, and Apollo Munich. The sectors are exactly the ones with high-volume, repetitive support: e-commerce, banking, insurance, travel, logistics, real estate, edtech.
The geographic spread explains one of the company's genuinely interesting technical bets. If you're building support automation for India and MENA and Southeast Asia, "supports English" is not a feature, it's a rounding error. Verloop.io built for 80-plus languages, which is the sort of requirement that shapes your whole architecture rather than sitting in a settings menu. Multilingual isn't the demo; it's the moat.
Eighty languages is not a checkbox. For a company selling into India, SEA and MENA, it's the whole product.
The listening problem
In 2024 Verloop.io leaned hard into voice, shipping enterprise voice AI agents that place and answer actual phone calls. The marketing writes itself - "AI that talks to your customers" - but the hard part of voice AI was never the talking. Text-to-speech is a solved-enough problem. The hard part is listening: parsing a caller with an accent, over background noise, saying a product name your model has never seen, and still landing on what they actually want. Verloop.io says its speech-recognition stack was trained on roughly 1,000 hours of real calls to do exactly that - filter noise, handle accents, catch intent. It is an unglamorous number that tells you where the real work went.
The other 2024 launch is the one that most cleanly captures the company's whole thesis. It's called Sparks, and it does auto-QA: quality assurance on support conversations. Traditionally, quality assurance in a contact center means a supervisor pulls a small random sample - two percent, maybe - listens to it, and grades the agents. Sampling is what you do when reading everything is impossibly expensive. Sparks reads everything. It evaluates 100 percent of conversations across voice and chat against a company's own scorecards, runs sentiment analysis, and checks compliance.
When the cost of reading every conversation drops to nearly zero, sampling starts to look quaint.
This is a small idea with a large consequence. Once you can afford to look at all of it, you stop managing to a sample and start managing to the whole. That's a different job. It is also, not coincidentally, exactly the kind of thing that's only possible now, with generative AI cheap enough to run against every transcript.
The lean exit
Here is the part that runs against the usual startup script. Verloop.io raised, over its entire life, something in the neighborhood of eight million dollars - a $3 million Series A in 2018 led by IDFC Parampara Fund with Kris Gopalakrishnan of Infosys, then a roughly $5 million extension in 2020 with Falcon Edge Capital and others, on top of a Y Combinator acceleration back in 2017. That is not a blitzscaling number. In a decade when conversational AI companies raised nine figures and burned most of it, Verloop.io stayed comparatively lean and built something enterprises kept paying for.
In July 2026 it got bought. Nurix AI, the enterprise AI company led by Myntra and Cult.fit founder Mukesh Bansal, acquired Verloop.io for an undisclosed sum. The logic is the tidy kind that makes acquisitions happen: Nurix had a voice AI platform called NuPlay; Verloop.io had a mature chat automation platform, a decade of enterprise trust, and a multilingual engine. Bolt them together and you get one company that does voice and chat, which is what enterprise buyers increasingly say they want - one platform, not ten point tools. The combined operation, the companies said, will automate more than 20 million customer interactions a month. Singh joined Nurix's leadership to run product strategy, enterprise go-to-market, and AI agent development.
It's worth sitting with the symmetry. The company began because its founders were running a large team of humans doing repetitive conversational work and realized they could automate it. It ends its independent life as part of a bigger bet on exactly that - automating conversational work - at a scale of tens of millions of interactions a month. Singh, who by his own account fell in love with coding at 13 in the small town of Rewa, spent a decade turning a contact-center annoyance into infrastructure, and then handed it to someone building the next layer up.
None of this required a superlative. Verloop.io didn't have the flashiest chatbot demo or the largest funding round. It picked one genuinely annoying problem - the repetitive conversation - and worked it completely: more channels, more languages, then voice, then reading every transcript instead of a sample. Unsexy completeness is a strategy. It also, as of this summer, turns out to be an exit.
