Stop Adding Lazy AI to Your SaaS

Introduction to AI in SaaS [00:00]

• The talk aims to provide a framework for decision-making regarding AI implementation in SaaS companies.
• It breaks from a typical three-act structure into two main parts: the five rules of AI within a SaaS company and adding AI as a feature.
• The speaker emphasizes the confusion between using AI for internal operations versus implementing it as a product feature.

Five Rules of AI in a SaaS Company [01:16]

• AI can be the entire product, a feature, used for building the product, for growth (sales/marketing), or for internal operations.
• Examples of AI as the product include Fiscal.ai and Lead Truffle, where the product's existence relies heavily on AI capabilities.
• Building foundational AI models (like LLMs) is generally not recommended for bootstrappers due to the difficulty of competing with large players.
• AI as a feature involves adding AI capabilities to an existing product, like Notion's AI writing assistant or Zoom's meeting summaries.
• AI for growth and operations are also discussed, with examples like Apollo and Jasper for growth, and internal support and HR tools for operations.

Outcome-Driven AI vs. Lazy AI [06:48]

• A key mistake is implementing "lazy AI," often by simply adding a chat interface or "sparkle" to existing text boxes without providing significant value.
• The focus should be on "outcome-driven AI" that genuinely saves users time and solves problems.
• eBay is used as an example of "lazy AI" where a text generation feature offers minimal benefit compared to the potential for AI to auto-fill listing details.
• The goal is to leverage AI to solve real user problems, not just to add AI for its own sake.

Six Uses of AI in SaaS Products [12:22]

• The six categories are: Conversational Interface, Generation, Categorization, Ingestion, Analysis, and Agentic Interfaces.
• Conversational interfaces allow users to interact in plain English, simplifying complex tasks like filtering data (e.g., Find Email).
• Generation involves creating content like text or images, as seen in tools like ChatGPT or SessionLab for agenda creation.
• Categorization uses AI to sort and label data, useful for asset management (Anchor Point) or email filtering.
• Ingestion focuses on processing diverse and sometimes messy data into structured formats, such as importing agendas from screenshots (SessionLab) or extracting receipt details (Reimbi).
• Analysis involves AI interpreting data to provide insights or scores, like sales call analysis (Gong) or code review summaries (Front End Mentor).
• Agentic interfaces (MCP and CLI) are the most cutting-edge, allowing AI to perform tasks autonomously, with ongoing debate about the best approach (MCP vs. CLI).

Pricing and Implementation Considerations [37:00]

• Three pricing models for AI features are discussed: absorbing the cost, including it in tiers, or offering it as an add-on.
• Launching AI features with a small subset of customers is recommended to understand actual costs.
• The discussion touches on potential downsides like liability, compliance, and the need for deliberate, not rushed, implementation of AI.
• The concept of "headless SaaS" where AI drives the functionality without a traditional UI is considered, likely falling under agentic interfaces.
• Customer attitudes towards AI, including potential pushback and the need to tailor AI messaging, are also briefly discussed.

Future of AI in SaaS [48:41]

• The speaker believes that while agentic interfaces are the future, they may not be the present for all SaaS applications.
• The rapid evolution of AI means that current best practices may change quickly.
• The importance of considering customer needs and potential competitive advantages when deciding on AI implementation is highlighted.