From prototype to adoption: Building gen AI products that stick (AIM115)

Key Takeaways

1. Understand User Needs: Spend time with the end-users to understand their pain points, challenges, and requirements before building the AI solution.
2. Prototype with Purpose: Build a real AI prototype using real data that can be battle-tested by subject matter experts.
3. Lower Barriers of Adoption: Continuously gather feedback from users to fine-tune the solution and make it more user-friendly.
4. Measure Success: Track both technical and product metrics to ensure the AI solution is delivering value and being used by the target audience.

Detailed Summary

1. Understand User Needs

• It's crucial to understand the user needs before building any AI solution.
• The speaker gives an example of a media company that initially thought about generating more content using AI, but found that their users were overwhelmed by the existing content. Instead, they needed a solution to curate and personalize the content.
• Similarly, for a sales team, the needs could be around lead generation, RFP completion, onboarding, or handling industry changes.
• Understanding the user needs helps create a well-informed hypothesis about the AI solution to build.

2. Prototype with Purpose

• Building a quick demo or toy data solution is not enough. Instead, you should build a real AI prototype using real data that can be battle-tested by subject matter experts.
• Bringing in the subject matter experts (e.g., lawyers, finance professionals, sales teams) early in the process allows you to iteratively refine the solution and make it production-ready.

3. Lower Barriers of Adoption

• After the initial prototyping and feedback gathering, the next step is to lower the barriers of adoption.
• This involves customizing the AI system to address the specific needs and pain points identified by the end-users.
• The speaker provides an example of how they used the Deep set platform to gather user feedback on the prototype through written responses, thumbs up/down, and comparing different options.

4. Measure Success

• It's important to measure both the technical and product metrics to ensure the AI solution is delivering value and being used by the target audience.
• Technical metrics include precision, recall, F1 score, and groundedness, which help assess the accuracy and reliability of the AI system.
• Product metrics include activation rate, task completion time, retention, and ROI, which help measure the overall impact and adoption of the solution.

Practical Example: Y Pulse's Customer-Facing AI Assistant

• The speaker shares the story of how Deep set worked with Y Pulse, a company that provides research and data on Gen Z and Millennial trends, to build a customer-facing AI assistant.
• Y Pulse followed the four key steps outlined earlier, starting with understanding their customers' needs, building a real prototype, iterating based on user feedback, and measuring the success of the solution.
• The result was a successful launch of the AI assistant within 120 days, with 75% user participation in the beta program and a 25% increase in overall engagement with Y Pulse's digital properties.
• The solution also provided a 30% higher perceived value to the customers, leading to a new monetization opportunity for Y Pulse.

Conclusion

The speaker emphasizes the importance of building AI solutions that people actually want to use and love, rather than focusing solely on the technical capabilities of the AI. By following the four key steps – understanding user needs, prototyping with purpose, lowering barriers of adoption, and measuring success – organizations can increase the chances of building AI solutions that are widely adopted and deliver real value.