Turning subscriber events into strategic intelligence
Objective
Enable Fyxer.ai to understand customer value early enough to influence marketing spend decisions.
Obstacle
True LTV only materialises months after signup, when spend is already locked in and cohorts have churned.
Outcome
A production-ready LTV prediction system that scores new users on day zero, giving the growth team a reliable signal to optimise spend and forecast revenue with confidence.
Background
Fyxer.ai is one of the most exciting AI productivity platforms in its category, a company laser-focused on helping modern teams eliminate busywork and operate at a higher level. With a compelling product and rapid user acquisition, the business was scaling quickly. But growth at speed always surfaces the same hard question: are we acquiring the right users?
As acquisition spend increased, the leadership team needed a reliable way to distinguish high-value users from those who would churn early. Without that signal, every budget decision was essentially a bet placed in the dark; optimised for volume, not for the long-term revenue that actually makes a business sustainable.
Challenges
Behavioural data without a value framework: Fyxer.ai had a wealth of signals captured in the first hours and days of a user’s journey, including product interactions, activation steps, and feature engagement. But without a framework to connect those signals to long-term revenue, the data sat unused. Acquisition and activation were optimised for getting users in the door, not for identifying which users would actually stick around and spend.
A feedback loop too slow to act on: Actual LTV metrics arrived far too late to be operationally useful. Real revenue patterns only emerge over months, by which point marketing budgets have already been committed and the window to intervene has passed. For a team trying to make fast, confident growth decisions, retrospective data simply was not enough.
Solution
Understanding early behaviour: The most important question in any LTV model is not statistical, it’s human. What does a high-value user actually do differently in their first day? To answer that, 173tech analysed Fyxer.ai’s user data to identify the behavioural signals that separated engaged, high-intent users from those likely to disengage. This included patterns like which features users explored, how quickly they reached key activation milestones, the depth and frequency of their early interactions, and whether their usage suggested they were embedding the product into their workflow or just taking a look. These signals were then engineered into a structured set of inputs the model could learn from, translating messy, real-world behaviour into a clean, predictive picture of user intent.
Training the model: With those features in place, 173tech trained and tested multiple predictive models, evaluating each against actual revenue outcomes to find the approach that best reflected reality. The winning model was then built into a fully automated pipeline, meaning every new user is scored automatically, without any manual intervention.
Accuracy drift: The pipeline includes ongoing performance monitoring that tracks whether predictions are holding up against real outcomes over time. If the model starts to drift, the team is alerted before it becomes a problem. This means Fyxer.ai isn’t just getting a one-time solution, they are getting a system that stays reliable as they scale.
Impact
Acquisition spend that works harder: Before this project, every pound spent on acquisition was evaluated on volume: signups, installs, activations. Now Fyxer.ai can see the predicted long-term value of the users those campaigns are actually bringing in. That changes everything about how budget gets allocated. Channels and campaigns that look efficient on the surface but attract low-value users can be identified and deprioritised. Those that consistently bring in high-intent, high-value users get more resource behind them. The result is acquisition spend that compounds over time rather than leaking quietly into churn.
Financial forecasts the business can actually rely on: Perhaps the most significant shift is in how Fyxer.ai plans. Per-user LTV predictions now feed directly into revenue forecasts, replacing broad assumptions with signals grounded in real user behaviour. The result is a 12-18 month revenue outlook that leadership can present to investors, use to plan headcount, and build product roadmaps around, with genuine confidence in the numbers underlying it.
Creating Value For Fyxer...
Day zero behavioural signals engineered into a rich predictive feature set,
Trained and evaluated across multiple machine learning models,
Delivering cohort-level LTV predictions within days of signup.
Get In Touch
Our friendly team are always on hand to answer questions, troubleshoot problems and point you in the right direction.
