March 28, 2022
Those who have been fans of Buzzle (formerly q&ai) from the very beginning (all 3 of them) know that we didn’t start with our current focus area - analytics and intelligence from recorded sales conversations. In fact, something that many can attest to, our YC partners chief among them, is that we’ve iterated from one idea to the next. Yet, the surprising thing is that while progressing on that journey, the core philosophy that motivates much of what we do today remains the same.
One of the most critical elements of our research in educational technology & NLP (Natural Language Processing) was the fact that we wanted to build an ML-centric solution that didn’t target completion but rather focused on evaluating understanding.
In an education system that’s basement, first floor and second floor are all built from bricks of grades and report card mortar, the natural evolution has turned to a completion oriented mindset. Information flows through students and the objective isn’t necessarily understanding but rather memorization for the sake of completing the next homework assignment.
In education, understanding can be viewed as a process by which the student creates a personal meaning or representation of what is being experienced. - Gary Jeffery, Memorial University of Newfoundland - Faculty of Education
While we didn’t see a realistic opportunity to shift this entire mentality, we did consider, how can we augment current practices to reinforce and encourage understanding?
So, Buzzle began as a tutee. We built a solution for students to tap into and for teachers to configure through which a student could explain concepts to our agent. By replacing or augmenting their existing completion centric assignments or studying habits with interacting with Buzzle, students were able to simultaneously able to complete work while reinforcing and understanding material. Teachers tuned our models with relevant subject matter, and students explained concepts to the agent. In turn, Buzzle would ask appropriate follow up questions, naturally designed to help the student fill in any knowledge gaps identified by the solution.
These efforts ultimately led to a paper published at NeurIPS2018.
In the world of product intel, customer voice and evaluating messaging, there is no such thing as completion. Every single piece of feedback and evaluation is centrally built upon a foundation of understanding: understanding how the customer feels, understanding how the customer responds and understanding what the customer wants.
The issue that persists is that the mechanisms by which gathering this understanding is difficult. Extracting and evaluating one’s understanding is a fundamentally human action, and in the world or revenue, marketing and product, this means biased, time consuming sync ups with the sales team.
So, as we took our trip through investor world and began the YC sprint, we started seeking out opportunities to apply our core solution in alternative application areas. The options were limited and not obvious at first. Until we heard about call recordings specifically from the perspective of a product manager:
“We have thousands of hours of recorded customer voice but don’t have the time to listen”
Imagine? Companies spend hundreds of thousands of dollars to capture VOC or Win/Loss data and feedback on marketing, messaging and positioning. Additionally, there was now this body of recorded customer voice with all of these answers - just locked up. What was the need? A bridge or conduit between these recorded conversations and the answers locked within.
The bridge? A solution that helps capture and evaluate knowledge in natural language data. Which sounded exactly like Buzzle.
Right now, no where. We’re abundantly aware of the fact that we’re just beginning to scratch the surface of what’s possible with this type of intelligence generation and remain blindly dedicated to improving our machine learning capability. We want to understand better, unlock deeper insights and empower faster.
However, beyond this door exists another, and another one after that. We’re just scratching the surface. The interesting thing about the idea of knowledge identification in this space is that it is just one channel through which knowledge appears. There are opportunities to contextualize and accentuate captured information with a number of datasource.
How does the pain described by a customer relate to their recorded user? To the information that’s found in outward facing marketing materials? To their support tickets? There’s an opportunity to further contextualize this knowledge, and we can’t wait till we get there!