Hear it from Amway
[TBD: one-line intro to the quote]
[Quote from Robby, pending.]
Client Challenge
Two pipelines that worked, but were slow and costly to maintain
Amway is the world’s largest direct selling company ($7.4B in reported sales, 2024). It makes nutrition, beauty, personal care and home products and sells them through Independent Business Owners in 100 countries. Its data already runs on Google Cloud and BigQuery.
We looked at two of Amway’s data pipelines: its account data, held in the Master Data Management System (MDMS), and its LATAM sales orders. Both worked, but they were slow and costly to maintain. Sales order data refreshed only every 10 minutes. The whole account table was rebuilt from scratch four times a day, even when only a few accounts had changed. Adding a market or a field meant new code or manual updates across dozens of files. Nothing checked data quality automatically, and four separate systems each had their own rules for who could see what.
Amway didn’t want a rebuild. The new setup had to sit alongside its existing tools, match the real data across 63 million accounts, and prove it could do so for less money.
The Solution
Both pipelines rebuilt on Databricks
In five weeks, we rebuilt both pipelines on Databricks and tested them head-to-head against the GCP versions running today.
The same build runs live or on a schedule, so Amway pays for real-time only where it matters.
Adding a new market takes one line of setup instead of new code.
The system no longer rebuilds everything every time.
A single system controls who sees what, and teams can ask questions of their data in plain language.
Analysts keep using BigQuery to read the results, and no data is copied.
Easy to extend: adding a new market takes one line of setup instead of new code, and a new field is one more line, with no downtime.
Pay for real time only where it matters
The same build runs live or on a schedule. Cost depends on how a pipeline runs, not how it is built, so each one runs at the speed the business actually needs.
Run-mode savings are measured against streaming sales orders nonstop, which costs about $32 a day at 0.3 messages per second (a 22-hour test run used 82 DBUs for about 16K messages). Sales orders cost about the same as today’s landing-only step, but now also cover cleanup, quality checks and the final reporting tables.
The Pipeline at Work
A sales order used to reach analysts on the next 10-minute refresh. Now it is ready in 30–90 seconds, checked for quality on the way, and analysts still read it in BigQuery.
Ready for analysts in about a minute
Reliable from the first record, and ready for questions
Checked against Amway’s real data
The few remaining gaps, including the 473-account difference in the history load, are small, known and being tracked.
Key Benefits Achieved
Documented business results, before and after
All figures were measured during the five-week pilot, running the Databricks pipelines head-to-head against Amway’s current GCP pipelines on real Amway data. Across all of Amway’s data, we project about $913K in savings over three years; that figure is a projection, not a measured result. Both workloads are now live in production.
What Amway can do now
Both pipelines are now live in production. Amway can move the rest of its data work to Databricks one pipeline at a time, without disrupting analysts or reports. Teams can already ask questions about sales data in plain language. Across all of Amway’s data, we project about $913K in savings over three years.
The takeaway. If you’re running critical data on a legacy cloud warehouse and want proof before you commit, a five-week pilot on your own data answers the cost, speed and accuracy questions. Talk to SunnyData about scoping yours.
