In 2014, a piece of software in New York started taking phone calls for banks. It could follow a conversation when the caller changed the subject, pick the thread back up, and - this was the strange part - actually finish the task. Not route the ticket. Not "connect you with an agent." Finish it. The software was called Amelia, named after Amelia Earhart, and it was built by a company almost nobody outside enterprise IT had heard of: IPsoft. Eight years before ChatGPT turned "conversational AI" into dinner-table vocabulary, Amelia was already clocking in for shifts.
That timing is the whole story. Amelia was early to digital employees, early to what we now call agentic AI, early to the idea that the phone call - not the chat window - is where enterprise AI earns its keep. Being early bought it Fortune 500 logos, analyst awards, and a seat in the server rooms of 15 global banks. It also meant spending two decades explaining a category before the category had a name.
The professor who left the whiteboard
Amelia begins with Chetan Dube, a mathematics professor at NYU's Courant Institute who left academia in 1998 to found IPsoft. The first product was not a talking agent. It was autonomic software - programs that monitored and repaired enterprise IT systems on their own, the digital equivalent of a building that fixes its own plumbing. For sixteen years, that was the business: unglamorous, invisible, and installed in exactly the kind of companies that never rip out something that works.
Those sixteen years matter more than they look. When Amelia the conversational agent arrived in 2014, IPsoft was not a startup pitching banks from a WeWork. It was a vendor those banks already trusted with their infrastructure. The agent walked in through a door the plumbing had already opened.
By 2020, the product had outgrown the parent. IPsoft renamed itself Amelia - a 22-year-old company taking the name of its own six-year-old product. Customers had made the decision for them: nobody was buying "IPsoft," everybody was buying her.
What Amelia actually does
Strip away the vocabulary and Amelia sells one thing: resolution. A customer calls a bank about a settlement query, an employee needs a password reset, a patient wants to reorder a prescription. Amelia's agents pick up - by voice or chat - understand the request in natural language, and execute it against the company's real systems. The conversation is the interface; the product is the completed task.
This is the line that separated Amelia from the chatbot generation. A chatbot follows a decision tree; wander off the script and it politely collapses. Amelia was built to handle digressions - you can interrupt your own address change to ask about your balance, and it will answer, then steer you back. It reads context, senses sentiment, and knows when to hand a frustrated human to another human. In the platform's current form under SoundHound, Amelia 7.0, this has matured into what the industry now calls agentic AI: autonomous agents that reason through multi-step tasks and act without a person in the loop.
The engineering choice underneath is the interesting part. Amelia 7.0 is built on a framework SoundHound calls Agentic+, a deliberate hybrid: generative AI handles the understanding, but deterministic workflows handle the doing. In a consumer app, a hallucinated answer is a screenshot on social media. In a bank's call center, it is a compliance incident. Amelia's bet is that regulated industries want creativity at the ears and predictability at the hands - paired with Polaris, SoundHound's speech recognition engine, which the company says reaches 99% intent recognition accuracy.
How an Amelia call resolves itself
Who picks up when Amelia answers
Amelia's customer list is a tour of industries where a dropped call costs real money. BNP Paribas Securities Services built NOA, its virtual assistant for custody, settlement and market queries, on Amelia in 2023. Telefonica runs it in contact centers. BBVA, Resorts World Las Vegas, Aeromexico, ASICS and Visionworks all appear in its deployment history. NTT DATA announced in 2024 that it would deploy Amelia alongside Commure Engage for patient engagement in healthcare. In late 2025, the French insurance broker Apivia Courtage brought Amelia 7 agents into its contact centers.
The pattern across those names: high call volume, regulated processes, and requests that are repetitive for the company but urgent for the person calling. That is Amelia's habitat. The business model matches - platform licensing and subscriptions sold to large enterprises, directly and through integrators and cloud marketplaces, wrapped with the integration work it takes to plug an AI agent into a core banking system without setting off alarms.
The $189 million question
Here is the number that makes Amelia a case study and not just a company profile. Over its life, Amelia raised more than $189 million - including a $175 million strategic investment from BuildGroup and Monroe Capital in March 2023, a raise advised by Goldman Sachs and TD Cowen. Seventeen months later, in August 2024, SoundHound AI acquired the whole company for roughly $80 million in cash and stock.
Selling for less than half your total capital raised is nobody's fairy-tale ending. But the deal logic ran deeper than the sticker. SoundHound had world-class voice technology and strongholds in automotive and restaurants; Amelia had two decades of enterprise workflow automation and nearly 200 brands in healthcare, insurance, finance and retail - industries SoundHound wanted and did not have. One company had the voice, the other had the rooms it needed to speak in. Within nine months, the combined team shipped Amelia 7.0.
Money in, money out
Where it stands in a crowded room
The market Amelia helped invent is now the loudest room in software. IBM, Google, Amazon and Microsoft all sell enterprise conversational AI. Kore.ai and Cognigy fight for the same contact centers. A wave of LLM-native startups like Sierra and Parloa pitch agents built from scratch on frontier models. Against all of them, Amelia's differentiation is not a benchmark score. It is mileage.
Gartner named Amelia a Leader in its Magic Quadrant for Enterprise Conversational AI Platforms two years running; Everest Group did the same three years in a row. Analyst plaques are decoration, but they proxy for something concrete: reference customers in industries where procurement takes eighteen months and failure makes the news. A startup can match Amelia's demo. It cannot match a deployment history that predates the transformer paper - or, now, the distribution muscle of a NASDAQ-listed parent shipping voice AI into cars, drive-throughs and call centers from one stack.
The open question is the one every pre-LLM AI company faces: does two decades of hard-won enterprise plumbing beat a foundation model with a fresh coat of venture capital? Amelia's answer, in effect, is that the plumbing was always the hard part. Anyone can talk now. The companies that win will be the ones allowed to touch the systems - and trusted to finish the task, the way a certain digital employee has been doing since 2014.
- 1998IPsoft founded in New York
Chetan Dube leaves NYU's Courant Institute to build autonomic IT software.
- 2014Amelia takes her first calls
The cognitive agent, named for Amelia Earhart, debuts in enterprise service desks.
- 20171Desk launches
Later Amelia AIOps: conversational AI wired into end-to-end IT operations.
- 2020IPsoft becomes Amelia
The company renames itself after its product and ships the Digital Employee Builder.
- 2022–23Analyst sweep
Gartner Magic Quadrant Leader twice; Everest Group Leader three years running.
- 2023$175M investment
BuildGroup and Monroe Capital back the company; BNP Paribas launches NOA.
- 2024SoundHound acquires Amelia
~$80M in cash and stock; Amelia becomes SoundHound's enterprise agent platform.
- 2025Amelia 7.0 goes agentic
Voice-enabled autonomous agents on the Agentic+ framework, with Polaris ASR.