Case Study

Eliminating busywork and building customer trust in online meetings

Deepgram’s fast and accurate speech recognition technology is the basis for UpdateAI’s action item detection engine for Zoom.

The Landscape

Improving Customer Success Productivity

The pandemic has changed how we work. This is even more true for Customer Success (CS) departments that are almost exclusively on Zoom or other online meeting platforms for their work. At the end of the day, CS professionals could have 6 or more back-to-back meetings, where they have to summarize and organize action items for each of their clients per day. That’s a lot of administrative work.

There are note-taking apps out there to help but they don’t provide any intelligence for the user to highlight all the action items and promises. CS still has to filter through all the transcription notes, which are not very accurate, and find all the action items.

The Challenge

Build an Action Item Engine with the Most Accurate Audio Data

Content: UpdateAI’s challenge was how capture all the action items during the meeting in real-time so immediately after the meeting the customer success managers have their list of to-dos. They can then immediately provide this list to their internal teams and back to their customers to work on immediately, without wasting time.

As for the action item collection, Josh Schachter, CEO of UpdateAI, said, “We had four main areas important to building our action item decision engine: accuracy, real-time speed, action item NLP patterning and NLP anti-patterning.” First, UpdateAI had to find an ASR solution that could provide the speed and accuracy they needed for their AI. They tested six different providers and chose Deepgram for its high accuracy and transcription speed. Josh goes on to say, “We were really impressed by the service, responsiveness, and general rapport of the Deepgram team complemented with superlative results of the speech recognition testing; it put them over the top.”

After solving for speed and accuracy, UpdateAI built their own natural language processing (NLP) engine to find the positive patterns that indicate actions items and filter out anti-patterns that create false-positives for action items. This was accomplished with their AI and data experts who developed a solid process and created a product that can be easily utilized by CS professionals. Their journey is described here.

The Solution

Speed and Accuracy Provides the Foundation for Success

Using Deepgram’s AI Speech Recognition solution, UpdateAI is getting the most accurate transcription data at real-time speeds. This data foundation is where their innovative AI engine can pattern match, and then learn and improve upon summarizing action items for each and every Zoom meeting.

UpdateAI’s action item detection is more than just keyword searching and uses AI to find the action items hidden within the transcript in order to highlight them to the user. In addition, UpdateAI has a team of data scientists and annotation experts to continually correct the AI and improve detection accuracy.