CMSPI helps many of the world's largest retailers make smarter payment decisions, with about one in four Global 500 companies among its clients. As the company built more sophisticated optimization capabilities on Databricks, consultants needed more than analytical output. Testing scenarios, saving drafts and refining insights required an interactive workflow that a purely analytical architecture was never designed to support. Adding Lakebase as a native operational layer closed that gap, cutting payment optimization from 30-plus hours of manual analysis to about 30 minutes. This identified approximately $27 million in realized incremental savings for existing merchant clients during the first six months of rollout, now being implemented, and additional to the value already delivered through CMSPI’s established optimization work.
Payment workflows needed more than analytics
CMSPI's platform processes around 400 different schema structures from merchant payment systems worldwide, normalizing that data into trusted products that expose opportunities across cost, revenue, fraud and all-around transaction performance. With the analytical foundation running on the Databricks Data + AI Platform, the next step was to make those insights actionable in the flow of real payment decisions.
Analytics could not preserve working context
CMSPI had built a web application that could serve graphs and analytics. However, consultants and payments associates were not just reading charts. They needed to save draft scenarios, adjust routing assumptions, compare options and return to their work without losing context.
"Every time a consultant made an amendment, they had to start from scratch. The architecture was built for analytical queries, not for holding user context across a working session," said Jordan Pierce, VP of Data at CMSPI.
Without a transactional layer, consultants had to export data just to keep working. That handoff could consume 30 to 40 or more hours for a single optimization project and was limited to aggregated estimates rather than transaction-level precision.
A separate database would add operational complexity
Adding a separate database would have meant managing ELT processes entirely outside Databricks.
"We don't want to manage loads of ELT processes outside of Databricks," said Jordan Pierce, VP of Data at CMSPI. "If Databricks produces something that solves that, it's always going to be the better answer."
Lakebase adds an operational layer where payment decisions happen
Instead of bolting on a separate database, the team deployed Lakebase alongside its existing Databricks infrastructure, creating a hybrid architecture where analytical and operational workloads each handle what they do best.
"What sold us on Lakebase was the synchronization. Delta tables flow straight into PostgreSQL with no friction, no separate processes to manage," said Jordan.
A hybrid architecture connects configuration and analysis
That hybrid design matches how payment optimization actually works. In CMSPI's Priority Routing Optimization Model, consultants build out scenarios through a web application, adjusting rate assumptions, saving drafts and comparing options as they go. Those working configurations are stored and preserved in Lakebase, so nothing is lost between sessions.
When a consultant submits a scenario, it triggers the analytical layer, where machine learning models and constraint-solving engines evaluate millions of individual transactions to find the optimal routing outcome. Results flow back through Lakebase into the application, giving consultants a seamless path from configuration to recommendation even with billions of available routing permutations to contend with.
"Lakebase gives us a completely clean separation of responsibility. Analytics lives where it belongs, and everything about how users interact with and customize their work is owned by Lakebase," said Jordan.
Consultants can work in context without switching tools
For the people using it, that separation is invisible in the best possible way. Consultants stay in one application, work in context and compare scenarios without switching between tools or losing progress.
"Without Lakebase, none of what we've built would actually be usable. The product simply could not exist in its current form," said Jordan.
Six months of transaction-level optimization delivered $27M in incremental savings
Work that once took 30 to 40 hours or more to reach an estimated answer now takes about 30 minutes, with greater transaction-level accuracy. That speed gives teams room to test more scenarios, refine recommendations and help clients act while the opportunity is still there.
"Before, consultants would pull data out of Databricks and try to make sense of it in Excel. Now that same work gets done in about 30 minutes, at a level of accuracy Excel could never match," said Jordan.
During the first six months of the Lakebase-enabled PROM rollout, CMSPI delivered measurable additional value across speed, realized savings and revenue:
Approximately $27 million in realized incremental savings for existing clients through PROM, now being implemented and additional to the value already delivered through existing engagements
About 30 minutes to complete work that previously took 30 to 40 hours or more
Roughly 14% more revenue per migrated project, reflecting the deeper savings surfaced for retail clients
Lakebase supports a broader product strategy
Lakebase is now becoming the operational layer behind CMSPI's broader product strategy. The team has launched three Genie workspaces to serve analytics through natural language and is building agentic workflows with knowledge assistants and supervisor agents that surface opportunities and generate recommendations.
Lakebase's integration with Databricks Apps and Replit also allows the product team to prototype new user experiences without dedicated web development.
"It is the foundation of how we do analytics as an organization. Without Databricks and Lakebase, a team of 25 could never keep pace with an industry that changes as fast as payments does," said Jordan.
FAQ: CMSPI and Lakebase on Databricks
Lakebase is a managed transactional database built natively within the Databricks Data + AI Platform. It exposes lakehouse data as Postgres tables with built-in autoscaling, governed by Unity Catalog. Unlike external databases, Lakebase eliminates the need for a separate ETL layer to move data from Delta Lake into production applications.