BLOK: Predicting the Product Before Its Shipped

The best product teams don't learn from failure. They learn before it happens.

That is the insight Tom Charman and Olivia Higgs are building around. Both spent years developing and deploying predictive models in some of the highest-stakes environments imaginable: defense, healthcare, and cybersecurity. Environments where the cost of being wrong is measured in lives and missions, not conversion rates. Olivia brings the product instincts to translate that scientific rigor into something product teams can act on immediately. Together, they founded Blok on a premise that sounds simple yet isn't: human behavior is not random. It is predictably irrational. And if you understand the patterns, you can simulate what users will do before they do it.

Oddbird is proud to be partnering with Tom, Olivia, and the Blok team from day one.

The Experiment Tax

There was a moment when experimentation was the answer. A/B tests gave product teams a rigorous way to learn. Run the test, wait for significance, ship the winner. The methodology worked because the conditions held: stable traffic, controlled exposure, and time to wait.

Those conditions no longer exist.

Teams now run hundreds of concurrent experiments and wait weeks for results that arrive inconclusive. Even Meta, with billions of daily active users and the infrastructure it essentially invented to make statistical significance tractable, struggles to achieve it outside the newsfeed. When the gold standard cannot solve the problem, the methodology is broken, not the execution.

The environment has shifted in ways that compound the dysfunction. AI-generated interfaces, personalized flows, and agent-driven interactions mean the surface being tested no longer stays fixed. You cannot run controlled experiments on systems that behave differently for every user. The experiment itself becomes the variable.

And the people who owned validation are disappearing. Product designers, front-end specialists, dedicated researchers: these roles are collapsing. One person is now expected to cover what three did. AI is compressing the build cycle further, meaning teams are shipping more with fewer people and less time to validate. That shift does not shrink the market for validation. It expands it. Every product manager and engineer who absorbed that work is now a potential Blok user, and they are the same people making the build decisions.

Predicting the Predictably Irrational

The hardest thing about building Blok is the same thing that makes it valuable: predicting human behavior.

Humans are not purely logical. They are influenced by biases, emotions, and cognitive shortcuts in ways that are deeply consistent and therefore learnable. Behavioral economists have spent decades cataloging these patterns. Tom has spent his career operationalizing them, building models that anticipate how people act under uncertainty in domains where the margin for error is near zero. That is the founding insight on which Blok is built, and it is the reason this company could not have been started by anyone else.

Most tools that claim to model user behavior work top-down: simulate a population, surface sentiment, and get a directional read. A digital focus group that scales. Useful, but limited. It can tell you what a population prefers. It cannot tell you where a flow breaks, why, or what fixes it.

Blok works bottom-up. It models behavior at the resolution of a single interaction: the hesitation before a form field, the exit from a multi-step flow, the exploration pattern that predicts drop-off three steps later. Then it reconstructs the population-level signal from the ground up. The result is not a sentiment score. It is a prediction of exactly where each user cohort drops off, what drives conversion, and what breaks at scale, before a line of code is written.

The agents running those simulations reflect the behavioral fingerprint of real user segments, modeled based on how those users interact with the product. Backtested against live experiments, they achieve up to 87% fidelity to real-world outcomes, well above the 30 to 60 percent variance explained by most behavioral models, and at the threshold where prediction becomes operationally useful.

Building this in-house is not a question of ambition. It requires aggregating behavioral data across products and industries to train a unified model, while simultaneously conducting academic-grade research that takes years to compound. Blok recently brought on two behavioral/cognitive scientists with PhDs from MIT and Harvard, in HCI, to lead exactly that work. The moat widens with every customer, every simulation, and every data point added to the system.

The Bet

The product team's problem is not a shortage of ideas. It is a shortage of conviction about which ideas are right before the cost of being wrong becomes real.

Early Blok customers are primarily enterprise organizations using the platform to experiment at higher velocity, validate assumptions before anything touches production, and walk into product reviews with evidence rather than opinion. A VP of Product at an edtech unicorn described spending a month developing screens and scheduling user testing. With Blok, those same insights take minutes. A CPO at a Fortune 500 company framed the problem plainly: too many experiments to run, not enough users to test against. Blok solves that.

For B2B product teams, the operating posture shifts most visibly. These teams use simulation to raise the evidentiary standard required before a feature gets built, drawing a direct line from a product decision to a business outcome before engineering ever touches it. The question moves from "what should we try?" to "what do we know is worth building?" That is a different relationship with the product process entirely, and it is the one that drives the deepest adoption.

Blok occupies a position in the modern GTM stack that did not exist five years ago and cannot be unoccupied. Analytics explains the past. A/B testing measures the post-release future. Blok answers the question that sits between them: what should we build next, and how will users respond when we ship it? Product leaders from Meta, Airbnb, Google, Slack, Uber, and Spotify have validated the workflow. The early champions are not replacing their experimentation infrastructure. They are making it dramatically more efficient by running fewer, better experiments against ideas already validated in simulation.

Why ODDBIRD

At ODDBIRD, we believe the category-defining companies of tomorrow may look odd today.

The idea that you can model a user's behavioral personality from product data and simulate their decisions with scientific rigor sounds ambitious to the point of implausibility, until you understand that Tom and Olivia have spent their careers solving exactly this problem in contexts where the stakes were immeasurably higher. That background is not a credential. It is a structural advantage that shapes how Blok thinks about model architecture in ways competitors will spend years trying to reverse-engineer.

Four forces are converging simultaneously: experimentation methodologies built for conditions that no longer exist, AI-driven interfaces that make controlled testing structurally impossible, build cycles that have outpaced the feedback loops meant to govern them, and the redistribution of validation to the people who never had time for it. These forces do not operate independently. They compound. Together, they create an environment where the old answer to "how will humans behave?" does not just underperform. It no longer exists.

We are proud to be betting on Tom, Olivia, and the insight that human behavior, properly understood, is predictable. Blok is that infrastructure.