Predicting renewals and reducing churn at scale

QuantSpark built a daily-refreshed machine learning model that predicts which customers of a large cybersecurity software provider are at risk of churning, and shaped its outputs into an operating rhythm the customer success team could actually act on. The provider's churn had crept up to nearly 8%, some four points above the 4-5% benchmark typical of SaaS businesses. On a revenue base above $500m, a single point of gross retention is worth millions of dollars a year, so this drift was a material threat to the business's trajectory, not a rounding error. QuantSpark's approach is projected to lift gross retention by 2-3 points, worth up to $30m a year. That figure is the business's own modelled projection based on the improvement the new approach enables, not yet a reported financial outcome, and it is presented here strictly on that basis.

The problem the provider faced was as much organisational as statistical. Its 300-strong customer success team had no standard way to track which renewal actions were being taken, or to judge whether those actions were working, so effort was not reliably directed to the accounts most likely to churn. The business needed three things at once: a scalable way to identify genuinely at-risk customers before they left, a method for turning that risk signal into interventions the team could act on day to day, and an interface that let leadership monitor churn as a managed metric rather than a lagging surprise discovered only at renewal time.

QuantSpark's method combined careful feature engineering with a model built for how risk actually behaves. The team engineered features tailored to the client's own product usage and customer base, then analysed which of those features carried the strongest predictive signal. The standout finding was that a lack of customer engagement over three or more months was a top predictor of churn, a concrete, actionable early-warning sign rather than an abstract risk score. Several algorithms were tested before the team selected a Long Short-Term Memory (LSTM) neural network, an architecture suited to modelling how a customer's risk evolves over time rather than treating churn as a single static snapshot. QuantSpark then built the infrastructure to run that model daily, serving fresh risk scores directly to frontline customer success teams and turning churn prediction into a live operational signal rather than a periodic report. Throughout, the model was judged on precision and recall rather than raw accuracy, reflecting that missing a genuinely at-risk account and wasting customer success effort chasing a false alarm carry different costs in a renewals context.

Technology alone does not change frontline behaviour, so a working group of customer success managers reviewed the risk scores against their own account knowledge and helped shape how the dashboards presented that information. That validation loop was what built the trust and adoption needed for the model to actually change what people did each day, rather than sitting unused alongside the old, instinct-driven process. One concrete result of that shift: the analysis linked missed customer-success meetings to higher churn risk, and the team instituted monthly check-ins in response, a direct, delivered behavioural change traceable to the model's output.

The value case rests on moving customer success from reactive firefighting, responding to churn after a customer has already disengaged, to proactive, risk-led nurturing that flags disengagement three or more months out. QuantSpark's projection is that this shift lifts gross retention by 2-3 points, worth up to $30m a year on the provider's revenue base, while also deepening customer relationships and creating room to grow expansion revenue. That projection is built by applying the modelled retention improvement to a revenue base of over $500m; because each point of gross retention is worth millions at that scale, a modest percentage-point shift compounds into a large annual figure. It should be read as QuantSpark's own estimate of the model's potential impact, not as an audited saving already booked. What is delivered today is the model itself, the daily scoring pipeline, the validated dashboards, and at least one instituted intervention; what is projected is the retention and revenue upside that follows if the business sustains that new way of working.