The business
Sendcloud, often referred to as the “Adyen for shipping”, is an Eindhoven-based scale-up and one of the fastest growing companies in the Netherlands. Sendcloud helps retailers optimizing their shipping process with their shipping platform. Sendcloud grew quickly over the past 7 years and is determined to do so in the upcoming years. Machine Learning helps Sendcloud to optimize sales, marketing and customer success.
Their business challenge: grow even faster
As Sendcloud grows, they experience the struggle many scale-ups go through. Where the focus in the first years lie on pure growth, sales and marketing, focus shifts slowly to long-term customer value. Machine Learning holds a logical set of techniques to increase customer value by focusing on leads and customers which hold long-term value, and improve retention among those valuable customers. How do we make sure these models cooperate and increase the value of Sendcloud? Rather than having ‘competing’ models!
Their ML challenge: optimize for customer value
In order to make Machine Learning a success, it’s crucial to strive and optimize for the same goal. Let’s assume you don’t, and let’s say you got two models, where one model is optimizing for lead conversion and the other one is trying to reduce churn. Changes are high that sales focuses on the easy – but often rather useless – leads which converse easily, but also churn easily. On the other hand, customer support will have a difficult time making sure those same customers don’t churn in the first couple of months. The result: you end up with two models (and teams!) competing each other.
Our solution and approach
Therefore, it’s crucial the models (and teams) strive for one, unified target variable, in this case the net customer lifetime value (or in short: customer value), which comes down to the following formula:
Net CLTV = (MRR – MRC) * lifetime – CAC
- MRR = Monthly recurring revenue
- MRC = Monthly (recurring) costs
- Lifetime = lifetime of a customer
- CAC = cost of acquisition
So, how does steering on net CLTV affect the models and teams?
- Sales & marketing: focus on the most valuable leads, which represent long-term value for the customer and are unlikely to leave the company quickly. In the end: sales and marketing will prioritize their actions based on net CLTV, and so will the models do.
- Customer success: decrease MRC by automating what can be automated, and focus on the valuable customers first when dealing with issues and support.
- Product: focus on features which increase customer satisfaction, increase sales and reduce churn.
During the peak of Corona crisis, we started off with our (completely remote) ML audit, to draw the ML roadmap, ML blueprint and the business case behind each use case. The four most valuable use case are:
- Churn engine: how can we increase lifetime using the patterns in customer behaviour?
- Lead prediction engine: which lead will convert to a valuable customer?
- Upsell engine: which customers are ready for a higher plan, to increase MRR?
- Ticket automation engine: how can we reduce costs by automating what can be automated?
The infrastructure
The first step: setup the ML infrastructure based on our ML blueprint, which you can find below:
A crucial component in this blueprint is our ML deployment platform: Deeploy. Using this software, we can easily deploy (new versions) of the ML models, and monitor and explain the behaviour of our ML models.
After setting up the infrastructure, next step: build the first ML engines, which are the churn and lead prediction engines. Together with the specialists of Sendcloud, we develop and implement these models and iterate to more and more accurate versions, which supports the sales and customer support teams at Sendcloud.
The first few months were an interesting ride, which taste like more! Together with us, 27 new colleagues started at Sendcloud to boost growth even further, slowly taking over the shipping market in Europe. We’re happy and proud to be part of this adventure!