The Chatbot Company That Bet Against the Chatbot Hype

In January 2023, a company most people had never heard of raised $40 million and named a new CEO on the same day. The headlines called Inbenta a chatbot startup. It was not a startup. It had been building conversational AI since 2005, back when the phrase would have drawn blank stares in most boardrooms. What looked like a debut was really the moment the market finally caught up to a company that had been running the same playbook for eighteen years.

That timing gap is the whole story of Inbenta. The generative-AI boom convinced every enterprise it needed a bot on its support page. It also introduced them to a new fear: the bot that confidently makes things up. Inbenta had spent nearly two decades on the unglamorous half of that problem - getting a machine to understand what a customer is actually asking, and to answer without inventing. When the fear arrived, the patience became a product.

01 / What it doesAnswering the questions companies are tired of answering

At its core, Inbenta automates customer conversations. When you type a question into a bank's help widget, an airline's chat window, or a software company's support page, there is a decent chance Inbenta is reading the sentence, working out what you mean, and pulling the right answer - in any of more than 100 languages. When it cannot resolve the request, it hands you to a human with the context already attached.

The scale is the part that surprises people. The platform handles more than two billion interactions a year. That is not two billion polished demos; it is two billion real customers, many of them frustrated, typing in shorthand, slang, and typos, expecting an answer that is correct the first time.

02 / The differenceThe unfashionable technology behind the accuracy

Most of the current AI wave runs on prediction: feed a model enough text and it guesses the next word well enough to sound human. It is powerful, and it also occasionally invents facts with total confidence. Inbenta's founder, Jordi Torras, a computational linguist, bet on a different approach - one that was deeply out of fashion for most of the company's life.

Inbenta calls it neuro-symbolic AI. In plain terms, the machine does not only learn statistics from data; it also carries a large, hand-built understanding of language - what words mean, how they relate, how meaning shifts with context. Torras has described the goal as letting AI "work efficiently with symbolic representations as humans do, not merely with numeric data." Today the company layers generative models on top, but the lexicon underneath is what keeps answers grounded.

Symbolic AI can work efficiently with symbolic representations as humans do, not merely with numeric data.Jordi Torras, Founder & Chief Innovation Officer

This is why regulated industries pick up the phone. A bank cannot ship a support bot that improvises an answer about your mortgage. Inbenta's referenced customer list - DocuSign, HP, BMW, Samsung, Deutsche Bank, Nationwide Building Society - skews toward companies for whom a wrong answer is a real liability. The pitch is not "our AI is the most creative." It is "our AI does not get you sued."

Note  Figures reflect the company's stated positioning; the shape is directional, not audited.

03 / ProductsFive modules, one conversation

Inbenta is not a single chatbot. It is a stack of modules a company can turn on as needed, all sharing the same understanding of the customer's intent.

The Learn module is a small strategy lesson in itself. It arrived through Inbenta's 2023 acquisition of Horizn, a company that built embeddable, interactive product demos for banks. The logic is quietly clever: a lot of "support" is not a question at all, it is a customer who does not know how to use the app. Showing beats explaining, and Horizn's demos reportedly kept roughly 80% of those cases in self-service instead of pushing them to an agent.

04 / Who uses it & how they payThe business of the boring, useful question

Inbenta sells to enterprises as software-as-a-service, priced broadly by the volume of interactions and the modules in play, with services wrapped around the bigger rollouts. Its customers cluster in finance, e-commerce, telecom, insurance, government, healthcare, and travel - industries with huge support volumes and low tolerance for error. Third-party estimates put annual revenue in the neighborhood of $53 million, on total funding of roughly $64.5 million.

The AI hype cycle rewards flashy launches. This business is measured in deflected tickets, resolved conversations, and agents who do not have to answer the same question twice.

The value proposition is easy to describe and hard to fake. Every large company sits on a mountain of the same repeated questions - where is my order, how do I reset this, what does this fee mean. Each one, answered by a human, costs money and time. Push enough of them into accurate self-service and the math changes quickly: agents get freed for the hard cases, customers get an answer at 2 a.m., and the support line stops being a cost center that only grows. Inbenta's job is to sit in that gap between "the demo was amazing" and "we put it in front of real customers" - and to survive the second one in a hundred languages.

05 / ExpertiseLinguists, not just data scientists

Part of what makes Inbenta unusual is who builds it. The company grew out of computational linguistics, and that heritage still shows. Rather than treating language as raw numbers to be crunched, its team encodes meaning - a lexicon that knows a "return" at a retailer and a "return" on an investment are different animals. That work is slow and deeply unglamorous, and it is precisely the part that is hard to shortcut by throwing a bigger model at the problem. The company holds patents in the area, including a semantic search engine built on meaning-text theory.

06 / The marketAn 18-year head start that finally means something

The competitive field is crowded now - Zendesk AI, Kore.ai, Ada, Cognigy, Yellow.ai, IBM's watsonx Assistant, and the AI features arriving inside Salesforce and Intercom all want the same enterprise budget. Most of them are newer to the problem. Inbenta's argument is not that it is the loudest; it is that it has been answering real customer questions at scale since before the category had a name, and that the accuracy layer it built out of necessity is exactly what nervous enterprises now want.

The recognition has followed. Inbenta won the 2025 TSIA STAR Award as Digital Customer Success Innovator of the Year, and took Best Customer Engagement Programme at the 2024 PAY360 Awards for work with Nationwide Building Society that more than doubled digital adoption among the building society's mobile customers.

07 / The road hereBarcelona to Allen, Texas

The company that began in a Barcelona office now runs seven of them, from Toronto to Toulouse to Akita, Japan, with headquarters in Allen, Texas. The founder no longer runs day-to-day operations; Melissa Solis does. But the original bet - that real understanding beats brute-force guessing - is still the thing Inbenta sells.