With ChatGPT, there is virtually no limit to the amount of text a company can produce to document its policies, practices, processes, and project progress. Edouard Reinach and Gabriel Lesperance, founders of the Quebec-based startup Trampoline AI, have devised an algorithm to facilitate information retrieval within an organization. Here’s how it works.
Like many entrepreneurial ventures, the idea behind Trampoline AI stems from a “boots-on-the-ground” problem experienced by one of the founders:
“For 3 years, I advised numerous companies on innovation issues, as well as productivity and performance challenges in a digital context,” explains Edouard Reinach, CEO of Trampoline AI. “And I would say that, among all the problems encountered, half of the solution came from managing the company’s knowledge.”
Upon closer examination, it becomes clear that the problem is both global and widespread: according to a study by McKinsey in 2021, an employee spends on average between 13% and 34% of their time searching for information to do their job. And the situation will only worsen as documentation centers grow.
“According to certain models, it is predicted that by 2025, companies will double the size of their documentation center every 75 days,” announces the co-founder of Trampoline AI. “However, at present, the main strategy for retrieving information within a company is to create archiving rules and conventions. No one follows these rules correctly, and there is nothing else to help businesses.”
One Question, One Answer
Trampoline AI is a direct response to this issue. The platform connects to the company’s work and communication tools (e.g., Office suite, Slack, Teams, etc.) and then serves as an interface for searching for information within the organization.
“We don’t want to be a new software added to the existing ones. That’s why Trampoline integrates as an extension into existing software. The user simply poses their question in a search bar, and the artificial intelligence algorithms take care of locating the information within the company. If the information is in an email, the application will redirect the question to the reference expert within the organization, rather than to the email content, out of confidentiality concerns.”
In response, the algorithm presents the most relevant “artifacts”; that is, it submits the most relevant excerpts rather than a summary of the gathered information, as ChatGPT might do – reducing the risk of receiving generative “hallucinations” as a response.
Trampoline AI is currently being tested by about a dozen clients (including Moment Factory and Enviro Integration) who use the platform on a daily basis.
“We have obtained our first paying customers,” confirms the co-founder. “And we plan an official announcement in March.”
The initial results are promising:
“We free up at least 3 hours of your employees’ time per week and allow them to quickly grasp any project,” it is announced on the website.
A project worth following!
