Skip to content

Measure what your decisions really achieve

DoubleML PySpark estimates the true effect of campaigns, prices and product changes from the data you already have. It runs in your own Azure Databricks workspace, scales with Spark and provides valid confidence intervals.

Install from the Microsoft Marketplace See the business example

DoubleML PySpark

Same data, opposite decisions

Customers who received a coupon spend more, but coupons usually go to customers who would have spent more anyway. Does the campaign pay for itself? It depends on how you measure it.

Click to show or hide:
Campaign pays for itself above 8.57 USD
True effect 6.17
Simple comparison
19.24
Keep the campaign+3.73 USD per coupon
Regression
10.28
Keep the campaign+0.60 USD per coupon95 % interval 10.17 to 10.39
DoubleML
6.19
Stop it, or target it-0.83 USD per coupon95 % interval 6.08 to 6.29
05101520

Additional revenue per customer in 30 days (USD)

DoubleML, zoomed in: 95 % interval 6.08 to 6.29 USD, far below the break-even of 8.57 USD.

True effect 6.17
6.0 6.2 6.4
-38,633 USDwith the current targeting rule
+81,885 USDwhen targeting with DoubleML

Simulated data, so the true effect is known. Coupon cost 3 USD, margin 35 %. See the full notebook.

Where it makes a difference

These are the questions driving major enterprises. Industry leaders use Double Machine Learning to answer them. Our package brings that same method to your workflow.

Marketing and promotions
Which campaigns and coupons pay for themselves, and for which customers?
Delivery platformPromotion targeting and discount depth in production, with about 50 % lower cost per incremental order reported, and comparable order lift at over 70 % lower cost.
Pricing and discounts
What does a price change or discount do to demand and revenue?
Fashion retailerDouble machine learning combined with transformer forecasting models to estimate price elasticity for pricing decisions.
Customer loyalty
What really brings customers back, and who would have come back anyway?
Travel platformHow buying an additional product affects repeat bookings, across more than 20 million trips. A sensitivity analysis shows how much hidden confounding it would take to overturn the result. Joint work with our team.
Customer behaviour and product changes
Which actions and features really make a difference when an experiment is not possible?
Online retailerThe causal impact of hundreds of customer actions across hundreds of millions of customers, with a causal machine learning pipeline distributed on Spark.

What you get

Results you can defend
Effects with valid confidence intervals, plus a sensitivity analysis for hidden influences.
Effects for each customer group
Not only one average, so you see where to act and where to save the budget.
Built for large data
Runs distributed on Apache Spark, on serverless compute and on clusters, with Spark ML models.
Your data stays with you
The software runs in your workspace and your data is not sent to us. See the privacy notice.
Simple to buy and start
Available through the Microsoft Marketplace and billed on your Azure invoice. A few lines of code in a Databricks notebook are enough to start.
Support from the specialists
Checked against the DoubleML Python package. Support in English and German, see Support.

Next steps

Getting started Install on Databricks All examples

For technical support, purchasing or a private offer write to cloud@economicai.com. For other projects and products of Economic AI, see economicai.com.