Case Study: Embedding GenAI Features into SaaS

Technology Providers

Embedding GenAI Features into SaaS

The Challenge: Integrating Generative AI Securely and Effectively

A successful Software-as-a-Service (SaaS) company providing a Customer Relationship Management (CRM) platform recognized the transformative potential of generative AI to enhance its product offering and provide significant value to its users. They envisioned features like AI-assisted email drafting, automated meeting summaries, and intelligent contact suggestions directly within the CRM interface. However, the company lacked the in-house expertise in large language models (LLMs), prompt engineering, and the specific architectural patterns required to integrate these capabilities securely, efficiently, and cost-effectively into their existing multi-tenant SaaS platform. Key concerns included ensuring data privacy, managing API costs, maintaining performance, and selecting the right foundation models for their specific use cases while ensuring a seamless user experience.

Our Solution: Strategic AI Integration and Engineering on AWS

Lydatum partnered with the SaaS provider, offering end-to-end AI strategy and engineering services to embed generative AI features responsibly and effectively. Our approach focused on security, scalability, and user value:

• AI Strategy and Use Case Definition: We worked closely with the product team to refine the vision for AI features, prioritize use cases based on user impact and technical feasibility (starting with automated email drafting), and define clear success metrics.
• Foundation Model Selection (Amazon Bedrock): Lydatum guided the selection of appropriate foundation models available through Amazon Bedrock. This managed service provided secure access to various LLMs (e.g., models from Anthropic, AI21 Labs, Cohere, and Amazon Titan) without requiring direct management of the underlying infrastructure. The choice was based on factors like performance, cost, specific task suitability (e.g., summarization vs. generation), and data privacy considerations offered by Bedrock.
• Secure Integration Patterns: Designing secure integration was paramount. We architected patterns that ensured sensitive customer data from the CRM (potentially stored in Azure SQL, as indicated) was not directly exposed to the LLM APIs or used for model training. Techniques like prompt engineering to include relevant context without sending raw PII, and using intermediary backend services for data handling and API calls, were implemented.
• Scalable Backend Services on AWS: We designed and built scalable, serverless backend microservices on AWS (using services like AWS Lambda, API Gateway, and potentially ECS/EKS for more complex tasks) to handle communication between the CRM frontend, the CRM's data store (Azure SQL), and the Amazon Bedrock API endpoints. These services managed prompt construction, API calls, response handling, and logging, ensuring the integration was robust and performant.
• Iterative Development and Rollout: The AI features were developed and rolled out iteratively, starting with a beta program for the email drafting feature, allowing for user feedback collection and refinement before a wider release.

The Impact: Enhanced Engagement, Differentiation, and User Value

The successful integration of generative AI features delivered tangible benefits for the SaaS company and its users:

The AI-powered features, particularly the time-saving email drafting assistant, led to a measurable increase in user engagement within the CRM platform as users found new value in the tool. This enhancement provided a significant point of differentiation in the competitive CRM market, helping the company attract new customers and retain existing ones. Customer feedback during beta testing and post-launch was overwhelmingly positive, validating the utility of the embedded AI capabilities. Lydatum's strategic guidance and engineering expertise enabled the SaaS provider to successfully navigate the complexities of integrating generative AI, positioning their product for future innovation.

Technologies Used: Amazon Bedrock, AWS Lambda, AWS API Gateway, Azure SQL (as data source), AWS IAM (for security), Python