The Challenge
As Sendcloud scaled, Machine Learning shifted from an experiment to a core growth lever. The challenge wasn’t whether ML could create value, but how to ensure it did so consistently and coherently across teams. Without a shared objective and a production-grade setup, individual model optimizations risked actively undermining overall business performance. Three challenges stood at the center:
- Competing optimization goals: Sales and marketing models were optimized for short-term lead conversion, while retention efforts focused on reducing churn. Without a unified target variable, these models — and the teams behind them — worked against each other, systematically prioritizing easily convertible but low-value customers at the expense of long-term revenue.
- Absence of a value-centric ML strategy: Customer lifetime value was not structurally embedded in decision-making across the organization. This made it difficult to balance acquisition costs, recurring revenue, and support effort in a way that would allow Sendcloud to scale sustainably rather than just quickly.
- From experimentation to production: Early ML initiatives lacked a standardized infrastructure for deployment, monitoring, and iteration. Without production-ready MLOps, models were difficult to scale, explain, and embed into the daily workflows of sales and customer success teams where they needed to have real impact.
Our Approach
To break this deadlock, we partnered closely with Sendcloud to design a unified, production-ready ML strategy centered around a single guiding metric: net Customer Lifetime Value (CLTV).
One Target Variable, One Direction
Rather than optimizing isolated KPIs, we aligned all models and teams around net CLTV — factoring in monthly recurring revenue, recurring costs, customer lifetime, and acquisition cost. This created a shared language and a common objective across sales, marketing, product, and customer success, ensuring every model pulled in the same direction.
Design Sprint to ML Blueprint
We kicked off with a fully remote design sprint, translating Sendcloud’s business goals into a concrete ML roadmap and prioritized use case backlog. Four high-value engines were identified: churn prediction, lead value scoring, upsell opportunity detection, and ticket automation. The blueprint clarified sequencing and dependencies, avoiding technical debt from the outset.
Production-Ready ML Infrastructure
We implemented a scalable ML infrastructure built around Deeploy as the central deployment platform — enabling versioned model releases, real-time monitoring, and model explainability. This ensured models could be safely iterated and embedded directly into day-to-day sales and customer support operations without disruption.
Iterative Model Development with the Internal Team
Working closely with Sendcloud’s own specialists, we developed and deployed the first core ML engines — churn prediction and lead value scoring — iterating continuously to improve accuracy and adoption. By building alongside the internal team rather than handing over a black box, we ensured the models were understood, trusted, and actively used where it mattered most.
