Background – AI-Powered Website Assistant
The team at the AI Lab developed an AI-powered web assistant for Keepr, a first-of-its-kind financial incentive platform that specializes in profit sharing, stock options and employee bonus plans.
Keepr allows businesses of all sizes to customize, launch, and manage profit-sharing, making it easy to share success with the people that matter most. Keepr is a subsidiary of Keeper Solutions. It is thus the perfect playground for the team at the Keeper Solutions AI Lab to investigate AI opportunities and develop proof of concept solutions.
Challenge – Providing Clear and Accessible Information
An AI-powered web assistant was first suggested as a way of increasing user engagement and helping users to better understand the concept of profit-sharing and stock options; an area that is traditionally fraught with confusion and over-complication.
This confusion presented a real challenge for the team at Keepr. Due to the innovative and novel nature of the platform, there would need to be a step-by-step learning process. The team at Keepr first wanted to educate users on the ins and outs of stock option plans and profit-sharing. After that, they could highlight the many advantages that the Keepr platform has over traditional financial incentive platforms.
Given the complexities of profit-sharing and stock options, it was essential to provide clear and accessible information to users. Additionally, Keepr wanted the AI assistant to capture key information from users, such as their company name and contact details, and to encourage them to schedule a meeting with the sales team.
To make matters slightly more complicated, the AI web assistant was created at an earlier stage of Keepr’s journey. The platform was in beta and evolving constantly, with new features added, and new perspectives discovered, on a near-daily basis. The team wanted to provide the web users with clear, up-to-the-minute information about the platform. However, due to its ever-evolving nature, this was easier said than done.
To overcome this, the team at the Keeper Solutions AI Lab created a single ‘Source of Truth’ document. This was a live document which contained the most up-to-date information about the platform, written in a style that could be easily interpreted by an AI assistant.
Solution – An OpenAI Assistant
To build the AI assistant, the Keeper Solutions AI Lab utilized the Assistants API that is available within OpenAI. Assistants is a purpose-built AI that uses OpenAI’s API and allows an AI-powered assistant to access files, maintain persistent threads and call tools.
For this proof of concept project, the AI assistant was programmed to use the ‘Source of Truth’ document, as well as the Glossary of Terms, to answer user queries about the platform.
As well as providing answers to queries, the AI assistant was trained to request contact details from web visitors and schedule sales calls. To schedule meetings, the team at the AI Lab developed a Python script which allowed the AI assistant to connect Google Calendar and Streak APIs. This allowed for seamless scheduling of meetings and lead management. This integration ensured that potential clients could easily book online meetings with the Keepr sales team, no matter their location.
The tech stack used to build the Keepr AI Assistant included:
-
-
-
-
-
-
-
-
-
-
-
- Frontend: VanillaJS, HTML, CSS
- Middleware Backend: Python FastAPI
- AI: OpenAI Assistant API, Google Calendar API, Streak API
- Server: Google Cloud Platform/Cloud Run Service
-
-
-
-
-
-
-
-
-
-
Results and Key Learnings
In order to gather insight and learn from this Proof of Concept project, the team at the AI Lab adopted a framework that included rigorous user testing.
The initial results were incredibly positive. User testing revealed that the assistant was highly effective at answering queries and guiding users through the platform’s features. Testing also uncovered opportunities for improvement such as the need to refine instructions given to the AI and update the ‘Source of Truth’ document to enhance accuracy.
There were also a number of stumbling blocks. The team encountered some reliability issues, particularly related to the partial downtime and slow response times of the OpenAI servers. Despite these challenges, the use of OpenAI Assistants (rather than relying solely on the completion endpoint) proved to be a major advantage. This low-code approach allowed for easy updates to the AI persona, ensuring the assistant remained relevant as the platform evolved.
Overall, the AI assistant has been instrumental in enhancing user engagement. Visitors to the Keepr website have used the assistant to answer any questions they may have and help them determine whether the platform is right for them. This interaction has fostered greater trust in the platform’s capabilities, which has ultimately contributed to client confidence.
On top of this, the AI assistant’s ability to use Google Calendar and Streak APIs to schedule meetings has streamlined the entire sales process. By automating meeting scheduling and lead management, the AI assistant has helped coordinate sales efforts across different time zones, ensuring timely and efficient follow-up with potential clients.
Key Learnings
AI Behavior: An AI assistant’s responses are shaped by the data it is trained on and the prompts it receives. While an AI assistant can occasionally produce unexpected outputs, these are not due to wrong intentions or desires but rather due to the complexity of the data. To improve performance, we have noted that it often helps to reexamine the data and the prompts at the source and make adjustments from there.
AI Limitations: While an AI assistant excels at casual communication and routine tasks, it can struggle with complex, high-stakes applications that require deep understanding and nuanced judgment. For these tasks, human oversight is crucial to ensure accuracy and reliability.
Client Engagement: The project highlights the incredible opportunity that companies have to use AI assistants to enhance user engagement. An AI-powered assistant can provide around-the-clock support, educate visitors and provide a highly personalized user experience. This can help to build trust and, ultimately, turn web visitors into dedicated customers.
