Cozmo AI Is Teaching Software to Answer the Phone
The San Francisco and Dubai startup builds AI "employees" that take the call, read the claim, and restructure the payment. Its founders once ran a Gen Z neobank. Now they want a machine on the other end of the line.
The most valuable thing a company owns is often the least glamorous: the phone line. Someone has to answer when a customer misses a payment, when a car is dented, when a policy is up for renewal. Cozmo AI is betting that "someone" no longer has to be a person. The startup builds what it calls AI employees - software that answers the call, reads the document, and finishes the task without handing it back to a human halfway through.
That framing - employee, not tool - is deliberate, and it runs through everything the company does. Cozmo AI does not sell a chatbot or a widget. It sells a role. You do not configure it so much as hire it. The distinction sounds like marketing until you sit with what it implies: an employee owns an outcome, not a task. It sees the job through from the first ring to the closed file.
01 / What it isSoftware that moves, not just talks
Cozmo AI, a Y Combinator Winter 2022 company, builds multimodal agents for regulated industries - mainly financial services and insurance. The platform is organized around four verbs the company uses like a spine: See reads documents, images, and video; Speak holds natural voice conversations on inbound and outbound calls; Think decides what to do next; and Do takes the action inside enterprise systems - restructuring a payment, filing a claim, updating a record.
The company's own description of its mission is unusually concrete for an AI startup. It wants to build "the system of action for the real economy" - software that answers the call, restructures the payment, approves the claim, and learns from every outcome. Note the absence of the word "insight." Cozmo AI is not selling a dashboard that tells you what happened. It is selling something that does the thing.
02 / The crewLiza, Alex, and Zara report for duty
When Cozmo AI launched its enterprise voice platform in December 2024, it did not unveil "modules." It introduced three colleagues, each with a name and a job title. The choice tells you how the company wants clients to think: you are not deploying software, you are staffing a desk.
Prospects, qualifies leads, negotiates, follows up, and keeps the CRM current - all through natural voice.
Runs collections, negotiates payment plans, and follows the regulatory rules a collections call demands.
Routes calls, books appointments, takes messages, and resolves the inquiries that never needed a human.
Above: three named AI employees, each shipped as a role rather than a feature. People manage people more easily than they configure software.
Underneath the friendly names sits the less friendly, more important part: the Agent Protocol Engine, or APE. It coordinates multiple agents working a process together and logs every step. In an unregulated business, that logging is a nicety. In collections or insurance, it is the product. A compliance officer will not let a bot near a customer unless every word and decision can be reviewed afterward. Cozmo AI built the audit trail into the foundation.
This is the part most AI demos skip. It is easy to make an agent that sounds convincing for ninety seconds on a stage. It is hard to make one an insurer will trust with a live collections call, where a wrong promise or a missed disclosure carries legal weight. The naming - Liza, Alex, Zara - is the warm front of a cold, careful machine underneath. Both halves matter: the human-shaped surface earns the client's comfort, and the logging earns the regulator's.
03 / The customerWho actually picks up
The company reports it is live with more than 10 enterprises, including Fortune Global 500 firms, and that its agents have handled over 1 million production calls and 10 million interactions. Those are the numbers that matter more than any funding headline, because each one means a real business let the AI answer a real phone. The named use cases are the unglamorous core of finance and insurance: customer support, collections and recoveries, first notice of loss on insurance claims, policy renewals, sales calls, and SME lending.
Company-reported figures. Numbers that mean someone trusted the AI to talk to their customer.
04 / The problemThe claim that took three weeks
Consider a property insurance claim. A customer calls to report damage. A human logs it, requests photos, waits, reviews them, drafts an estimate, routes it for approval. Weeks pass. Cozmo AI's pitch is that a multimodal agent can compress that chain: it answers the first notice of loss by voice, ingests the photos and video with its vision layer, produces a preliminary estimate, and moves the file forward - measured in minutes instead of weeks. To make that real, the platform plugs into the systems these companies already run on, including Guidewire in insurance, Temenos in banking, and Salesforce for CRM.
The four-verb stack. Perception on the left, action on the right, a decision in the middle.
The efficiency claims are large - the company cites 300 to 400 percent gains across client sectors and cost reductions of up to 70 percent. Those are self-reported and worth treating as directional rather than audited. But the underlying logic is hard to argue with: a voice workforce that runs around the clock, in dozens of languages, without breaks, bends the cost curve of frontline labor in a way a call center cannot match.
Illustrative view of Cozmo AI's stated focus areas, weighted by emphasis in company materials. The pattern: high-volume, rules-heavy, regulated.
05 / The foundersFrom teen banking to enterprise AI
Cozmo AI is run by Alok Kumar, its co-founder and CEO, and Nuha Hashem, its co-founder and CTO. Before this, the two built Zywa, a social banking app and card aimed at Gen Z users aged roughly 11 to 21 in the Middle East. Zywa went through Y Combinator's W22 batch and raised a $1 million pre-seed to expand across Egypt and the UAE. Then the founders walked away from teenage neobanking toward the opposite end of the market: regulated enterprise finance.
The pivot looks dramatic on paper, but the through-line is a specific skill. Winning the trust of a cautious customer over a screen or a phone - and doing it at scale, in the Middle East, across languages and dialects - is exactly the muscle an AI voice company needs. Kumar has been named to a Forbes 30 Under 30 list. The company kept its Y Combinator backing through the transition.
There is also a quieter lesson in the switch. Zywa taught the founders how a phone-shaped interaction can carry money and trust at once, in a region where trust is not assumed. A neobank for teenagers is, at bottom, a machine for having millions of small, high-stakes conversations. Cozmo AI is the same machine pointed at a different customer - the enterprise instead of the eleven-year-old - and stripped of the consumer branding. The product changed. The problem, talking to people at scale about money, did not.
06 / The mapArabic first, then the world
Most AI startups launch in San Francisco and treat other languages as a later port. Cozmo AI ran the other way. It positioned its platform as the first enterprise-ready voice AI employee product in the Middle East and Africa, built to handle multiple Arabic dialects natively, alongside English, Spanish, French, German, Hindi, and more than 70 other languages. Its go-to-market centers on MENA, Southeast Asia, and India, with operations reaching the GCC, Europe, and the US. Global headquarters sit in San Francisco; the EMEA base is in Dubai.
There is a strategic elegance to owning a room the incumbents overlook. And the marketing has a sense of humor about it - the company ran a campaign advertising its voice AI on the roofs of UAE taxis, meeting the market where it literally drives rather than waiting for a conference booth to find it.
07 / The businessHow it makes money, and against whom
Cozmo AI sells B2B software. Enterprises deploy its AI employees through a no-code platform that slots into their existing core systems, and the company frames the value as replacing both a stack of point tools and a set of repetitive human roles. The competitive field is crowded and getting more so - horizontal agent builders and voice-AI specialists like Sierra, Decagon, PolyAI, Parloa, and Cresta all circle the same enterprise budget, and legacy contact-center and automation vendors are not standing still.
Cozmo AI's answer to "why you" rests on three things that are hard to copy quickly: end-to-end ownership of a whole regulated process rather than a single step; a multimodal stack that reads a claim and speaks to a customer in the same workflow; and native strength in Arabic and adjacent markets. Backing the effort is a $3 million seed round from Y Combinator, the Dubai Future District Fund, Goodwater Capital, Rebel Fund, and others. It is not a war chest. It is enough to prove the calls get answered - which, for now, is the only proof that counts.
Where Cozmo AI fits in the market is still being written. The AI-employee category is loud, and much of it is demo-ware that folds the moment a real customer goes off-script. By choosing the hardest, most consequential calls first - collections, claims, lending - the company is either building the deepest moat in the category or the steepest hill to climb. The million calls already logged suggest, at minimum, that the phone is ringing and something is picking up.