The business
Our client was an East Coast data and analytics company with hundreds of millions of dollars in annual revenue, later acquired by a large global information company. It turned communications and identity data into products for banks, lenders and insurers. Its mission was to help them verify customers and manage risk with accurate, current data.
The challenge
Turning raw data into products buyers trust
Petabyte-scale data
Six petabytes of production data had to stay fast, accurate and available.
Two environments
Data and tools were split across on-premises data centers and the cloud.
Machine learning on-premises
Models were built and run on-premises, which made them slow to scale.
Buyers expect fresh data
Banks and lenders paid for accurate, current identity and risk data, delivered on time.
The solution
What we delivered
- Data strategyFirst
- Data products at scaleOngoing
- Machine learning moved to AWSDone
- A data strategy
- Part of the client's plan to grow revenue from its data.
- 6 petabytes in production
- Data platforms run and supported across on-premises and cloud.
- Data product pipelines
- Pipelines that collect, clean and package data for each buyer, on schedule.
- Identity and risk products
- Data and models behind identity verification and fraud risk products, kept accurate and current.
- MLOps
- One standard way to train, test, release and monitor models.
- Machine learning moved to AWS
- On-premises models and pipelines moved to AWS, one at a time.
Results
The outcomes
Run in production
Across on-premises and cloud, with data products kept current.
Faster model releases
One MLOps path took models from test to production sooner.
Lower infrastructure cost
Machine learning scaled on AWS, paid for as it was used.
Faster product launches
New data products reached buyers sooner.
What it made possible
A data business built to grow
New data products sooner
Teams built and released new products for financial services buyers faster.
Scale on demand
Model training grew with demand, without new hardware.
Fresh identity and risk data
Products refreshed on schedule, with every change tracked.
Ready for AI
A modern platform for newer AI work on the same data.