The Startup Teaching Machines to Shape What You Do Next
Sequen AI has built something the hyperscalers kept to themselves: a behavior engine that predicts and steers user actions in real time. Now it wants to sell it to everyone else.
There is a machine running quietly behind every major consumer platform you use. It watches not just what you click, but the order in which you click things, how long you pause, what you skip. It uses that sequence of events to predict what you are likely to do next — and then gently adjusts what you see to push the outcome in a direction the platform prefers. TikTok built its entire empire on this idea. So did Instagram. So did Netflix.
The uncomfortable part is that most companies have no access to anything like it. They rely on demographic profiles, purchase history, and cookie-based tracking — tools that feel, by comparison, like using a map printed in 2009. The gap between what hyperscalers can do and what everyone else can do has only widened. Until, possibly, now.
Sequen AI, a New York-based startup that came out of stealth in early 2025, is betting that it has cracked a version of this problem. Its platform — which it calls a Behavior Design Engine — uses a proprietary model architecture called Large Event Models, or LEMs, to predict and influence user behavior in real time on sparse data. You do not need a billion users. That is the pitch, and it has already caught the attention of Greycroft, one of New York's better-known venture firms, which led Sequen's seed round. The company says its first five customers saw a combined $162 million in incremental revenue lift within seven months of deployment.
The Problem With Personalization
Most enterprise personalization tools are, in Sequen's framing, backward-looking. They build a profile of who you are based on your past behavior and serve content they think matches that profile. It sounds sensible. The trouble is that it misses the most useful signal of all: what you are doing right now, in this session, on this device, at this moment.
TikTok does not serve you content because it knows your age and city. It serves you content because it has watched billions of other users make similar sequences of choices — linger on this, scroll past that, pause here — and learned to predict what comes next. The model updates continuously. It is less a profile engine and more a sequence engine. Sequen is trying to replicate that logic in an API that an enterprise retailer, a travel platform, or an ad network can bolt onto its existing stack.
"Gently shaping user behavior, nudging you toward decisions before you've even realized it."
— Sequen AI, from its launch post, 2025
The technical architecture behind this matters. Sequen's LEMs are trained on what the company describes as billions of user event sequences — not just purchases or clicks, but the full behavioral trail. When a new user arrives, the model does not need that user's history. It needs only their current session signals, matched against patterns from everyone else who has moved through an application in a similar way. The result, the company claims, runs at under 25 milliseconds per query — fast enough to influence what appears on screen during a live browsing session without the user noticing any lag.
The Team and the Backing
Sequen was co-founded by Zoë Weil, Mo Afshar, and Alex Thom. Weil, who serves as CEO, previously held a senior vice president role at Citi and has a background in machine learning engineering, including time at Etsy where she worked on large-scale search and discovery systems. Afshar and Thom round out the technical core, with experience across AI infrastructure at firms including Anthropic and DataMiner.
Greycroft's investment memo, published in May 2025, drew a pointed comparison. It described the trio as having "Collison-level brilliance in AI," a reference to Stripe's founders, and highlighted what it saw as Weil's particular edge: the ability to get large enterprise customers on the phone before the product even launched publicly. From the first customer introduction, Sequen reached $1 million in annual recurring revenue. The next target is $10 million.
That trajectory matters. Enterprise AI sales tend to be slow, politically complex, and dependent on proof-of-concept work that can drag on for months. Reaching seven-figure ARR early suggests either a genuinely compelling product, unusually strong distribution skills, or both. Sequen appears to be claiming both.
What the Platform Actually Does
Sequen offers three primary APIs: one for recommendations, one for search, and one for discovery. The underlying infrastructure is described as reinforcement-learning native rather than retrofitted — meaning the system is designed from the ground up to optimize for outcomes, not just to rank content based on historical relevance scores.
The federated learning aspect is worth flagging, partly because privacy concerns around behavioral data have become a real obstacle to enterprise adoption. Sequen claims its model learns from patterns across all its customers without ever mixing individual company data. Each customer's data stays siloed; what transfers is the learned structure of how users behave, not the raw data itself. Whether that framing will fully satisfy regulators in the EU or cautious enterprise legal teams remains to be seen. But it is a coherent answer to an obvious objection.
| Feature | Traditional Personalization | Sequen (LEM-based) |
|---|---|---|
| Data model | User profile (static attributes) | Event sequences (real-time behavioral trail) |
| Update cadence | Batch (daily or weekly) | Continuous, in-session (<25ms) |
| Data requirement | Large historical dataset per user | Sparse data; generalizes from network |
| Privacy model | User-level tracking, often cookie-dependent | Federated; no cross-customer data leakage |
| Core mechanism | Collaborative or content-based filtering | Reinforcement learning, goal-optimized |
| Target use cases | Email, retargeting, static recommendations | Live search, discovery, in-session ranking |
The Market Sequen Is Entering
The personalization and recommendation engine market is not short of competition. Salesforce Einstein, Adobe Target, Dynamic Yield (now part of Mastercard), and dozens of smaller vendors all occupy this space. Most have been around long enough to accumulate significant enterprise contracts and integration depth.
Sequen's argument against the incumbents is essentially the same one any LLM startup makes against legacy software: the old tools were built for a different set of assumptions. Dynamic Yield was designed in an era of cookies and CRM data. Sequen says the underlying math is different now, and that retrofitting reinforcement learning onto legacy infrastructure produces worse results than building natively for it. That may be true. It is also the kind of claim that is easy to make and hard to verify from the outside.
What gives Sequen some credibility here is the specificity of its early results. The $162 million revenue lift figure covers only five customers and seven months, which means it is not a sample size anyone should treat as definitive. But enterprise software vendors rarely publish numbers that specific unless they are confident they will hold up to scrutiny. That Greycroft chose to highlight the figure in its investment memo suggests at least some external validation.
The Broader Tension
There is something worth sitting with here. Behavior design is not a neutral term. The technology Sequen is commercializing — optimizing for conversions, nudging users toward decisions, shaping what someone does before they have fully decided — can be used to surface products people will genuinely find useful. It can also be used to push people toward higher-margin items, to extend session time beyond what a user intended, or to exploit moments of low attention.
Sequen is targeting ecommerce, travel, and ad tech — sectors with mixed track records on user-aligned design. The company's framing focuses on enterprise revenue metrics: conversion rates, order value, engagement. That is a reasonable commercial frame. It is also worth asking whose outcomes are being optimized, and whether the two are always the same thing. This is not a reason to dismiss the technology. It is a reasonable question to keep asking as behavior optimization spreads beyond hyperscalers into the general enterprise stack.
What Comes Next
Sequen's path from $1 million to $10 million ARR will likely depend on two things: expanding within existing accounts and landing larger logos. The first is often easier than it sounds in enterprise software, where a successful pilot in one business unit may still require a separate procurement cycle in another. The second is where Weil's background in financial services and large-scale enterprise sales may prove relevant.
The company came out of stealth in early 2025 and is actively expanding its team. Its seed round was led by Greycroft with participation from Vinyl VC and Correlation Ventures. No Series A has been announced publicly as of this writing.
The core bet Sequen is making — that the behavioral modeling techniques locked inside TikTok and Instagram can be productized and sold to the rest of the market — is a reasonable one. The timing is not obviously wrong. Enterprise appetite for AI that moves revenue metrics, rather than AI that generates text summaries of internal documents, appears to be growing. And the post-cookie, privacy-first shift in digital advertising has left a gap in targeting infrastructure that something like Sequen's model appears designed to fill.
Whether the technology lives up to its early numbers at scale is a different question. Most enterprise AI platforms that looked compelling at five customers looked very different at five hundred. That is the next test.
Sources
| # | Source | URL |
|---|---|---|
| 1 | Sequen AI — Official launch post, "Introducing Sequen AI" | sequen.ai/blog-post/introducing-sequen-ai |
| 2 | Greycroft — Investment announcement, "Unlocking The Wisdom Of The Crowds" | greycroft.com/perspectives/…sequen |
| 3 | Sequen AI — About page | sequen.ai/about |
| 4 | Sequen AI — Platform overview | sequen.ai |
| 5 | ZoomInfo — Sequen AI company profile | zoominfo.com/c/sequen-ai/… |
| 6 | Zoë Weil — LinkedIn profile (Sequen AI) | linkedin.com/in/zoefrancesweil |
| 7 | Rivalsense — Sequen competitive intelligence profile | rivalsense.co/intel/sequen |
