By Eddie Lee, General Partner, and Ted Vinnitchouk, Analyst
TLDR: White Star Capital co-led Sequen’s $16m Series A, joined by Threshold VC and returning investors Greycroft and Vinyl Capital, to support the continued development of enterprise-scale personalisation for enterprise consumer applications.
Direction will always beat prediction.
Think about all of your favorite consumer applications of the last decade, they would be pretty boring if they only showed you content that you already know and like.
Instead, they subtly push your taste forward, introducing content you may never have searched for or expected to care about. So how do they consistently know what to show next? And why is it so effective at keeping us engaged, even when we intend to log off?
It’s no secret that the best social media platforms don’t just show content based on what they think users already like. They focus on understanding user intent in the moment, and use it to shape what people see next, keeping them engaged longer. They are powerful platforms because of this reason.
But, how do they actually do this?
Every aspect of our in-app behavior (clicks, hovers, rewatches, pauses) is a datapoint, and every time we log-on, our behavior is recorded as a sequence of these events. With enough user data, neural networks can extrapolate patterns based on historical users, and use them to predict the next set of events in the sequence.
Feeding these models a handful of early signals is enough to predict whether a user intends to be deeply engaged for the remainder of the session or quickly drop off.
However, understanding intent is only the first step. Intent is not fixed; it is fluid, and can always be reshaped with the right content at the right moment.
Reinforcement learning models drive this process. These models push various pieces of content to users to learn how different content shapes downstream behavior. Over time, these models understand exactly which content strategies sustain engagement across different user states.
This is why our favorite platforms feel so deeply personalised. They aren’t just showing you what you like, they are reading your intent moment-by-moment and adapting its content strategy accordingly. At times, it likely knows more about what you want than you do.
The opportunity set for this state-of-the-art technology is vast, especially for businesses like online stores, travel booking platforms, and marketplaces where consumer intent is most directly mapped to revenue.
However, these models have historically only been viable for a small number of the world’s largest platforms, which serve billions of users who often engage 20+ times per day. For the other 99% of platforms, a lack of dense behavioral data has historically made it impossible to train these models with comparable predictive power or intent-shaping accuracy.
Sequen’s Large Event Models to super charge in-session personalisation
With its invention of Large Event Models (LEMs), Sequen was built to bring the class of models that made platforms like TikTok so effective into any consumer application..
LEMs ingest sequences of events to understand user intent and adapt user experience in real time. They continuously rank and re-rank outputs to maximise the likelihood of downstream business outcomes, drawing on patterns observed from users with similar intent profiles. Sequen delivers these capabilities through an ultra-low-latency API in under 20 milliseconds.
These models don’t optimise only for what happens in the moment. Sequen’s LEMs use delayed reward modeling to learn from events across sessions, allowing the system to optimise not just immediate outcomes like conversion and average order value, but also longer-term metrics such as lifetime value.
To solve the data constraints that have historically hindered in-session personalisation, Sequen aggregates and generalizes sequence and event data across all platforms within a vertical, creating a shared learning layer that overcomes the sparsity of any single site. Today, Sequen is already processing over 20 billion monthly requests.
Partnering with Sequen
Over the past few years, we’ve seen transformer-based foundation models emerge with unprecedented speed across formats such as natural language, speech, audio, and video.
Despite the state-of-the-art research and extensive resources put behind them, we’ve seen limited enterprise adoption of foundation models that today largely focuses on productivity gains rather than direct business growth.
Sequen is different.
From our very first meeting, we were blown away by Sequen’s uncompromising focus on commercialisation and the speed at which the team delivered meaningful ROI to enterprise customers.
Sequen’s LEMs deliver immediate, directly attributable revenue lift for its enterprise customers from day one. This creates clear, quantifiable value for the world’s largest enterprises and positions Sequen at the core of customers’ revenue infrastructure across a $29 trillion industry.
The impact is undeniable. Since commercialising this quarter, Sequen has already partnered with some of the world’s largest consumer companies and has helped Fortune 500 companies see material lift in conversion at scale.
The company is also actively in the implementation phase with multiple Fortune 10 organisations, underscoring both the scalability and enterprise readiness of the platform.
Humans today; agents tomorrow?
Today, Sequen’s LEMs learn from human behavior to optimise outcomes like conversion and revenue. But the underlying architecture does not depend on whether those events are generated by a person or an AI agent. The models simply learn from sequences of behavior and optimize for the outcomes those sequences produce.
That distinction matters because the way consumers interact with platforms is beginning to shift. As AI agents start browsing, consuming, and purchasing on behalf of users, many systems built around traditional human interaction patterns may struggle to adapt.
Sequen’s approach is fundamentally different. Because LEMs learn directly from sequences of events rather than assumptions about the user behind them, the system can adapt naturally as those patterns evolve. If agent-driven interactions become more common, the reinforcement learning layer will update in the same way it does for any other change in behavior: by learning which sequences lead to the best outcomes and optimising accordingly.
The N-of-1 team redefining personalisation
No other team combines the technical depth and execution velocity required to make this vision real. Translating frontier AI research into large-scale commercial impact is in the founding team’s DNA.
Sequen is led by CEO Zoë Weil, who previously led Etsy’s ML strategy as a Staff Applied Scientist, where she and her team drove over $1bn in GMV lift in 2023.
Zoë is joined by CTO Ethan Benjamin, who worked alongside her as a Staff Applied ML Scientist at Etsy, Chief Scientist Mo Afshar, formerly Head of Data Science and Engineering at Motion MSK, and Chief Scientist Alex Thom, previously a Staff Software Engineer at Dataminr, and CPO Raphael Louca, previously an Engineering Manager at Meta. The rest of the founding team is made up of leaders from Google Deepmind and Anthropic.
This technical depth is further complemented by CRO Tim Satterwhite, who helped scale Braze from $12m in revenue to an IPO.
We couldn’t be more excited to partner with such an exceptional team alongside Threshold VC, Greycroft, and Vinyl Capital.
For more information, and to have a look at what the team is building, check out Sequen’s website: https://www.sequen.ai/