In our “Engineering Energizers” Q&A series, we highlight the journeys of engineering innovators making significant strides in their fields. Today, we feature Sneha Singla, Director of Software Engineering at Salesforce, who leads the delivery of end-to-end, real-time personalization across web and mobile platforms using both rules-based and AI-driven recommendations.
Dive into how Sneha’s AI Personalization team tackled the challenges of launching AI web personalization capabilities on an aggressive timeline, seamlessly integrating AI-driven recommendations across multiple channels, and balancing rapid deployment with rigorous trust and security standards to redefine customer engagement.
What is your team’s mission?
Our team specializes in delivering end-to-end, real-time personalization across various channels, including web and mobile. We craft tailored experiences by combining both rules-based and AI-driven recommendations, which dynamically adapt to customer behavior on a large scale. These recommendations enable marketers to engage customers with greater relevance and precision.
Rules-based recommendations are straightforward and effective, using predefined criteria to suggest complementary products after a purchase, for example. AI-driven recommendations, however, take personalization to the next level by leveraging machine learning to analyze complex behaviors such as browsing history and geographic data. This advanced approach ensures that our tools provide actionable insights, facilitating data-driven decision-making.
At the heart of our personalization strategy is the use of AI-driven recommendations, which empower marketers to deliver smarter, more adaptive experiences. This focus aligns with Salesforce’s broader vision of using AI to revolutionize customer engagement.
Sneha explores Salesforce Engineering’s culture.
What was the most significant technical challenge your team faced recently?
Launching AI web personalization capabilities within an aggressive four-month timeline was a significant challenge for our team. With only six members, we had to develop, test, and demo a fully functional personalization tool for Dreamforce. Initially, marketers needed to write code within Data Cloud, which created usability issues.
To address this, we iteratively developed a point-and-click interface called the Web Personalization Manager (WPM). This interface enables marketers to configure personalized content without requiring technical expertise. The WPM serves as the gateway for delivering AI-driven recommendations, allowing marketers to access advanced machine learning capabilities in an intuitive, user-friendly format.
WPM testing parties were instrumental in identifying and resolving bugs collaboratively. Regular stakeholder demos provided valuable feedback, which helped us refine the tool. These efforts ensured that the WPM was reliable, intuitive, and ready for a successful debut at Dreamforce.
How does Data Cloud play a role in enabling real-time AI-based recommendations?
Data Cloud is the backbone of our team’s efforts to deliver real-time, cross-channel personalization powered by AI. Acting as a central hub, it consolidates customer interactions, providing a unified view that enables dynamic recommendations.
One of the key roles of Data Cloud is data integration. It consolidates data from Salesforce platforms like Sales Cloud and Service Cloud, creating a single source of truth. This ensures that all customer interactions are captured and analyzed in one place, enhancing the accuracy and relevance of our recommendations.
Data Cloud also captures and updates customer behaviors in real time, such as purchases or browsing history. This real-time insight is crucial for ensuring that AI models can adapt dynamically to changing customer preferences and actions.
Another important aspect is cross-channel consistency. Data Cloud supports personalization across web, mobile, and email, ensuring that customers have cohesive and seamless experiences regardless of the platform they use.
Additionally, Data Cloud is designed for scalability. It processes countless data points, enabling personalization for both small businesses and enterprise customers. This scalability ensures that our solutions can handle the vast amounts of data generated by different customer bases.
By serving as the foundation for AI-driven recommendations, Data Cloud transforms raw data into actionable insights. Its ability to power scalable, adaptive AI solutions directly supports our mission of redefining customer engagement.
Sneha shares what keeps her at Salesforce.
What specific challenges arose from integrating AI-based recommendations across multiple channels?
Integrating AI-based recommendations across web, mobile, and email channels presented several challenges, primarily in data synchronization and maintaining consistency. Each channel relied on unique data flows, making it difficult to create a unified customer view. To address this, we utilized Salesforce’s Data Cloud to centralize data, ensuring that interactions from one channel informed recommendations on others.
Latency posed another significant obstacle. By refining and scaling our recommendations and decision pipeline, we were able to ensure that AI-driven recommendations were not only timely but also optimized for diverse use cases across channels.
Additionally, aligning AI models to different channels required iterative tuning. For example, mobile recommendations need to be visually concise and easy to navigate, while web recommendations could leverage richer and more detailed data. These tailored adjustments demonstrated the versatility of AI-driven recommendations in delivering consistent, scalable experiences across multiple platforms.
How does your team balance the need for rapid deployment with maintaining high standards of trust and security?
Balancing speed and security is a core aspect of our development process. Salesforce’s robust Core environment provides a solid foundation, ensuring that features undergo rigorous testing before deployment. Functional integration testing (FIT) helps validate both the quality and security of our features, significantly reducing risks.
To streamline development, we broke the work into incremental vertical slices, focusing on shipping functional components while adhering to trust guardrails. Testing parties enabled developers to address bugs and edge cases efficiently, ensuring that each component was reliable and secure before moving on to the next.
For AI-driven recommendations, safeguarding customer data is paramount. We implemented strict validation processes and adhered to compliance standards to ensure that the recommendations generated by our models are both reliable and unbiased. This balance between security and innovation allows us to deliver impactful AI personalization without compromising trust.
Sneha spotlights what drives Salesforce Engineering innovation.
What does the test and evaluation process look like for ensuring the reliability of AI-based recommendations?
The test and evaluation process is designed to ensure that AI-driven recommendations are both reliable and scalable. It begins with functional tests to validate individual features within Salesforce’s core platform environment. Integration tests then simulate real-world workflows to assess seamless cross-channel functionality. Here are the key steps:
- Functional Testing: This step validates that individual components, such as AI-driven personalization algorithms, operate as intended. The focus is on identifying and resolving isolated issues within standalone features, ensuring that core functionalities—like real-time recommendation generation—are accurate and efficient before broader integration.
- Integration Testing: These tests simulate full workflows that mimic how various channels (web, mobile, email) interact in real-world scenarios. The goal is to ensure that AI recommendations are consistent across platforms and to identify any potential conflicts or synchronization issues that could disrupt the user experience.
- Testing Parties: Collaborative sessions where developers work together to identify edge cases, such as unexpected user behaviors or simultaneous interactions across multiple devices. These sessions help address rare but impactful scenarios that standard testing might overlook.
- Stakeholder Demos: Regular demos are conducted with internal stakeholders, including product managers and marketers, to gather feedback on how AI recommendations align with business goals. These sessions refine both the functionality and usability of the recommendations, ensuring they meet user expectations.
- Customer Success Validation: This involves testing AI-driven recommendations in real-world customer environments, such as retail or healthcare scenarios. The focus is on ensuring that the recommendations deliver measurable business value and adapt to the specific needs of different industries.
These rigorous processes are designed to optimize the performance of AI-driven recommendations, ensuring they provide consistent, high-quality personalization across a variety of customer interactions.
Learn more
- Stay connected — join our Talent Community!
- Check out our Technology and Product teams to learn how you can get involved.
