Global Unified Loyalty & AI Agent Program

The Brief

Client working on a retail industry wanted a complete IT landscape transformation from older, scattered platforms into a single Salesforce platform. 17 countries was part of the scope with initially requested big-bang approach.

Constraints

Maximum of 5 seconds latency with a real-time first integration concept, and reliable backend to handle 1+ Mio daily events. Data transformation and migration for over 500 million records from various tables, and most storage optimized implementation to minimize the future extension investment. Europe-wide GDPR compliancy rules. Finally, agentic internal process handling framework and training.

Options Considered

1. Having a middleware with message queue to be sure that the integrations will scale and error-caused data failures are minimized
2. Direct integration to reduce the latency and to stick with real-time requirements
3. Hybrid-option to use a middleware for the non-realtime required flows, while keeping the integration direct for the true real-time handling required use cases + reduce the overhead for fast-evolving interfaces
4. System of records being Salesforce Core/CRM vs Salesforce Data Cloud

Decision

In the end, hybrid-option has been decided for the solution to cover the project requirements. Main drivers for the choice was the loyalty program being demanding about the true real-time processes and required flexibility over the certain interfaces managing core loyalty functionalities. For the system of records side, it has been decided to use Salesforce Core with the reason of Data Cloud not being an MDM platform and lacking the functionality of generating a persistent Id. Single golden id has been generated in Salesforce Core and distributed to all the systems internally and externally to support data governance framework designed.

To validate the platform scalability, load and stress testing has been done using tailored golang test suite of my creation.

Outcome

Scalable platform carrying to load of the 17 countries including the peak seasons. By using the data platform power of Data Cloud, data shared across the different Salesforce clouds and powered Agentforce for internal users and marketing purpose. On the busiest day, over 6 Mio daily events has been successfully processed by hitting the latency threshold.

What I'd do differently?

I would press harder on the middleware approach to be used wider than the current state to scale the platform better and avoid minimal impact on the first year where the implementation tested with full scale production load. In addition to that, I would increase the volume on the integration testing to get better prepared for the unexpected peak season load.