Nytro.ai uses Deepgram’s Speech-to-Text (STT) API to Optimize Sales Pitch Performance

Digital Learning Management Systems have many advantages over on-site training programs. They tend to be more flexible in their pacing, they are customizable per individual, they reduce costs and time investments for management, and they can provide regular assessments of a rep’s progress. But the traditional LMS solutions fall short in a few key areas.

First of all, while quizzes and surveys can show what reps have memorized, they cannot give a manager insight into how the lessons from training are being applied during actual sales conversations. Second, they do not include experiential learning, or learning by doing. Research tells us that we retain a bare 5% of the information when we listen to a lecture. Even with audio-visual methods and live demonstrations, retention never rises beyond 30%.

However, an impressive 75% of knowledge is retained when we learn by doing. This vast difference is because active participation increases the level of reflection and cognitive processing undertaken by the learner. Nytro.ai’s Pitch Intelligence platform augments passive training methods with real-world pitch practice scenarios across a variety of different use cases and personas. Then, the algorithm assesses, gives feedback, and quickly identifies which reps are ready to engage with prospects and which need more training. The platform gets reps ready to pitch to real clients faster than traditional methods, and without risking loss of potential business to inexperienced reps forced to learn on the job.

There are three components of a practice pitch which the algorithm needs to assess to track progress and give feedback: first there is the visual data, body language, facial expression, etc. Then there are vocal qualities, such as tone of voice, pace of speech, diction and clarity. Finally, there is the text of the pitch; the most critical element. In order to determine how things are being said, first the AI needs to understand what is being said. This means that any Pitch Intelligence solution has to incorporate incredibly accurate STT. If the STT’s transcript is off even by a little bit, the subsequent downstream models’ analyses will be thrown off by a massive factor, and ultimately the algorithm won’t be able to give accurate feedback. Chief Product Officer, Ravish Kamath, says, “At the end of the day, our product is only going to be perceived as good as the quality of our transcriptions.”

After testing many STT solutions, from several open source offerings, Deepgram proved the only STT provider that could offer what Nytro.ai needed to build its Pitch Intelligence platform. Where other STT solutions fell between 75% and 80% accuracy, Deepgram’s accuracy is consistently 90% to 92%. And while other solutions could occasionally achieve a WER that was sufficient for Nytro.ai’s purposes, they simply weren’t scalable, and their transcription speed was inconsistent. Deepgram’s reliable, consistent, quality STT has enabled Nytro.ai to focus on their downstream models for analyzing and providing quick insights to their customers. Ravish noted, “Deepgram is a core part of our offering, and it will remain so. We have found more than a STT solution but a partner in Deepgram to help us innovate and improve upon our offering.”