Yotpo Reduces Databricks & AWS Costs Costs by 39% and Boosts Performance with Zipher Real-Time Optimization

February 25, 2025

Yotpo is a leading eCommerce platform that helps eCommerce businesses drive more sales from repeat shoppers. Yotpo offers a suite of retention marketing products such as SMS & email marketing, subscriptions, reviews, and loyalty programs. 

• Customers: tens of thousands of global businesses.
• Databricks environment:
  • Cloud infra: AWS
  • Annual spend: $500K-$800K
  • Scale: 
~75,000 
monthly scheduled jobs.

Key results:

• 39% savings in annual Databricks and AWS costs.
• 21% reduction in job failure rate.
• 100% compliance with Yotpo’s SLA requirements.
• 0 engineering effort is required.

 

How can data processing operations be balanced with cost control?

Yotpo is using Databricks to manage scheduled jobs that serve customer-facing features and data analytics. Over the past few years, Yotpo has been growing rapidly, which has led to a growing challenge to maintain data performance and efficiency at scale. 

To support the growth in workloads, Yotpo had to increase its Databricks capacity to meet SLA requirements. While keeping performance at sufficient levels, this resulted in exploding costs:

“Managing thousands of variable jobs, our data team cannot chase every job and adjust the configurations. At some point, we tried the Databricks auto-scaler, but our costs kept going up”, Moshe Derri, Yotpo’s Data Platform Lead.

 

Dynamic & automated scaling in real-time with Zipher

To solve the problem, Yotpo engaged with Zipher in 2024 and set a goal of cutting its Databricks and correlated AWS costs by 20%. Zipher outperformed and enabled a 39% cost reduction in Yotpo’s annual Databricks + AWS costs. 

 

How did Zipher unlock the potential of Yotpo’s Databricks environment?

Autoscaling


Using a spark-aware ML-based model, Zipher actively learns and profiles the workloads. It then adjusts cluster resources at runtime. Zipher’s autoscaler demonstrates far better results than the default autoscaler that could be over-provisioned.

 

Dynamic cluster configuration

Instead of manual setup and static configurations, Zipher applies an automated & dynamic approach that selects the optimal settings per job, per run. The optimization engine adjusts different Databricks aspects such as worker node and driver node types, disk size allocation, while considering actual resources such as availability zones, spot instance availability and termination probabilities.

Seamless integration

Zipher easily integrated with Yotpo’s cloud data stack, to enable the optimization of jobs that are triggered either directly from Databricks, or through 3rd party solutions (e.g., Airflow).

 

Results

Zipher was gradually deployed on more and more jobs in Yotpo’s production environment, and demonstrated immediate and significant opportunities for savings (2 weeks from installation to actual savings).

By applying Zipher’s autoscaler and dynamic cluster configuration, Zipher cut Yotpo’s annual Databricks costs by 39%.

Zipher not only reduced Yotpo’s costs, but also enhanced job performance by reducing failure rate by 21%. Zipher’s dynamic configurations increased stability by ensuring that the optimal settings are selected per job per run, while considering market conditions such as actual spot terminations probabilities and availability of instances across different availability zones.

Want to learn more about our solution? Contact us to hear about our tailored optimizations!