A Bar-Ilan NLP PhD went to Amazon Research, then Yahoo, then back to school, then built a startup that resolves your shipping complaint before you finish typing it.
The premise of Tymely, the company Ohad Rozen has been running out of New York since 2021, is that support tickets are not a chat problem. They are a reasoning problem. A retailer's inbox fills up with sentences like the second sneaker is the wrong size but only the left one and my delivery says complete but I am staring at my empty stoop. Chatbots have historically responded to these by offering to help you track your order. Rozen's pitch is that the actual job is understanding the sentence, checking the order history, deciding whether policy allows a swap, and closing the ticket. He calls it end-to-end resolution. Investors — Hetz Ventures, DESCOvery from D.E. Shaw, and 97212 Ventures — have put roughly $15.25 million behind the idea.
Rozen is unusual among AI founders in that he spent time doing the underlying research. Between 2018 and 2021 he worked toward a PhD at Bar-Ilan University in Israel on natural language processing and deep learning, publishing papers on how NLP models generalize (or don't), on question answering across products, and on targeted knowledge enhancement for textual inference. His most-cited paper, "Diversify Your Datasets," has 46 citations, which is not going to make anyone a Turing laureate, but does place him in the working set of researchers who have thought seriously about how language models fail on inputs that look almost, but not quite, like their training data. Customer service emails are exactly that.
He is also unusual in that he has done this before. Before Tymely, Rozen was CEO of Toonimo, a cloud-based platform that walked website visitors through their tasks with guided audio and animation. He raised about $7 million there. Toonimo taught him — you can see this in his subsequent product decisions — that guiding humans through a workflow is hard, and that most of the difficulty is not in the interface but in reading what the human actually wants. He then went back to finish the PhD, spent time as an NLP research scientist at Amazon and at Yahoo, and started Tymely with Roy Penn, a former VP of engineering at the Israeli NLP shop Amenity Analytics.
Customer service today is broken. Many businesses are losing a lot of money because of poor customer service. — Ohad Rozen
Tymely does not measure itself by how naturally its bot chats. It measures itself by how many tickets it closes without a human touching them. Refunds issued. Addresses changed. Exchanges processed. Everything else is theater.
Rozen has said publicly that Tymely employs experts who review AI outputs and correct them in real time. The tradeoff is deliberate. The AI gets to sound like a top agent because a top agent is somewhere in the loop.
The stated cost claim is 50–80% below an outsourced contact center. The SLA claim is minutes, not hours. This is the argument you make when you are pitching a VP of CX, not a CTO.
Source: Tymely funding disclosures and press coverage (CXM Today, Calcalist Tech, Tymely news page).
Rozen's path to founder-CEO is the sort of path that reads clean in a pitch deck and looks improbable in real life. He built a startup, raised money, then paused to complete a PhD. Most people do one or the other; doing both suggests either an unusual amount of stamina or an unwillingness to leave open loops. Reading the papers, the second explanation is at least partly true. His research is preoccupied with the ways language models look like they understand a sentence but do not, and with what happens when you give them slightly harder examples than the ones they were trained on. That is essentially the Tymely problem statement, restated in academic prose.
The company itself is small — 24 people at last count — and it sells into a category that is often cynical about AI, because the category has been sold AI many times before and has watched it fail. E-commerce customer service is dense with edge cases. The customer wants to return one shoe. The customer's package was marked delivered but wasn't. The customer bought during a sale, wants to price-match, and is emailing at 2 a.m. Tymely's response to this has been the opposite of most generative AI startups: rather than promise more autonomy, it promises more correctness, backed by human reviewers. This is the sort of decision a research scientist makes.
Rozen has also written, once, for Entrepreneur.com on why audio content works for engaging audiences. It is a curious sideline for someone whose product is entirely text-based, but it fits a pattern: he thinks about how humans decide, and he thinks about how to guide them through decisions. Toonimo did that with voice-over-website tours. Tymely does it by writing back to them like a competent employee. The medium has changed. The problem has not.
His co-founder, Roy Penn, comes out of Amenity Analytics, one of the more underrated Israeli NLP shops, which specialized in extracting structured signals from unstructured financial text. That team was, essentially, in the business of teaching machines to read the boring parts of the internet accurately. Tymely is a natural port of that skill set. The angry-customer email is arguably the least boring text on the internet, but it demands the same core capability: reading what a person actually meant.
The business claim — 50 to 80 percent cheaper than a contact center, SLAs measured in minutes — is a claim about unit economics, not about model capability. It is also, importantly, the claim that gets you a Series B in 2024, a year when AI startups selling agentic products had to explain, in numbers, why they were not just a wrapper. Tymely's answer has been that its edge is not the model. Its edge is the loop around the model. Whether that continues to hold as base models get better is the interesting question, and it is the one Rozen is presumably being asked in every board meeting.
He is, notably, not a public founder. There is no Twitter feed of hot takes. His LinkedIn is a résumé. He gives interviews when there is news and mostly stays quiet in between. This is either strategic or temperamental, and given his background, probably temperamental. Researchers tend to prefer publishing findings over publishing threads.
"Customer service today is broken. Many businesses are losing a lot of money because of poor customer service."
"Tymely employs experts that review each AI input and, if needed, correct it in real-time. This results in human-level accuracy that enables us to understand tiny and implicit nuances in a customer's text."
"I am very happy for the backing of Hetz Investments, D.E. Shaw Group and 97212 Ventures in helping with the expansion of our vision to improve customer support for everyone."
Co-founder and CEO of Tymely, an AI startup building an autonomous agent for e-commerce customer service. Previously CEO of Toonimo, and an NLP research scientist at Amazon and Yahoo.
It provides an AI agent that resolves complex customer service tickets — refunds, exchanges, address changes and more — end-to-end via email, chat and web forms, replacing outsourced contact centers for retailers.
PhD in Natural Language Processing from Bar-Ilan University (2018–2021), with published research on NLP generalization, textual inference and question answering.
Approximately $15.25M in total — a $7M seed in 2022 and an $8.25M Series B in September 2024. Investors include Hetz Ventures, DESCOvery (D.E. Shaw's venture studio) and 97212 Ventures.
Headquartered in New York with roots in Israel's NLP research community. Around 24 employees at the latest available count.