I sat down to talk AI with CallRail CPO Ryan Johnson and AssemblyAI Founder and CEO Dylan Fox. We covered the nature of the two companies’ partnership, CallRail’s history of AI innovation, and a sneak peek into what’s next for AI-powered Conversation Intelligence®.
What do you two make of the current AI boom? Will it make a lasting impact, or is it more like an NFT or cryptocurrency moment?
Dylan: The boom we're seeing in AI is real because it drives real outcomes and value to users, customers, and businesses.
Ryan: With ChatGPT, you have a hundred million people that used it in a very short period of time. It's taken that barrier down of people saying, we have to figure out how to use this. Unlike NFTs and cryptocurrency, more people get it because they’re actually using it.
Dylan: CallRail is a great example of a software platform leveraging AI to deliver incredible amounts of value to customers and users in thoughtful ways that were not possible, or that users and customers were not ready for six months ago.
Ryan: The most exciting part is we weren’t caught on our heels. We've been using AI for years. You know, CallRail launched Conversation Intelligence in 2016. So we weren’t in a rush to get something to market — just bolt this thing on to say we have AI.
Ryan, why did CallRail decide to partner with AssemblyAI?
Ryan: At CallRail, we're focused on what's happening during a call — there's so much intelligence there. Speech-to-text is so vital to Conversation Intelligence, so partnering with a company that is leaning into that aspect was critical: one focused on voice intelligence and deployed it to customers in a production environment, not in a science laboratory.
Dylan: Correct. We’re an applied AI company. We're focused on the ways real-world customers need AI for speech recognition. And they need it to operate with very high reliability and accuracy.
Ryan: That was the foundation of our partnership: knowing AssemblyAI is focused on ASR and speech-to-text because that’s the foundation of anything else we do, right? So, once you have superhuman transcription, all the other intelligence you can run on is much better than if you started with a bad transcript. Otherwise, it's bad in, bad out.
Dylan, tell us more about the speech recognition model that powers Conversation Intelligence. What makes it so accurate?
Dylan: Our current model, Conformer-1, was trained on over 1 million hours of audio data. It approaches human-level performance for speech recognition — excellent results. Soon, we’ll release an update to that model that'll be trained on three to four times that amount of data — over a million hours of audio data, around 200 terabytes. To put that into perspective, a couple of years ago, state-of-the-art systems were trained on less than 50,000 hours of audio data.
Ryan: Accuracy. The most challenging thing with speech-to-text is accuracy. And yet for many, the bar is set at human-level accuracy. That’s funny to me because we’re all human, and all humans are error-prone, right? That’s why talking about transcription and analysis with superhuman accuracy and reliability gets me excited.
Dylan: Me too. And within the next eight to 12 months, more powerful models will bring us even closer to that goal of superhuman accuracy and reliability. The technology to accomplish that goal exists today.
AI’s capabilities also seem to be getting more and more advanced faster than any of us could have imagined. Is that fair to say?
Ryan: Absolutely. I've never seen acceleration like this — products and features moving from concept to market in a matter of months. I mean, we just released features that wouldn’t have been possible six months ago. It’s rewarding to release new products and features so quickly, making them more accessible to our customers in a much shorter time.
Dylan: And I would say it's almost a disadvantage to those larger companies that have tried to do a lot of this on their own.
Ryan: Yes, it's been tough for many companies to keep up. For us, it's been an advantage because we’ve had a head start and a strong foundation in Conversation Intelligence that we’ve been able to iterate and improve on. The features we recently launched got me really excited.
You’re talking about call sentiments and call summaries?
Ryan: Yes. call sentiments and summaries, both features are part of Premium Conversation Intelligence, made possible by AssemblyAI and Conformer-1’s advanced capabilities.
Call summaries are 3-5 sentence takeaways of phone conversations, automatically created with near-human accuracy. Call sentiments let you know at a glance how conversations went — AI categorizes them as negative, positive, or neutral.
Some of the technology out there today cannot handle the summarization of a 30-minute phone call. But with Premium CallRail Conversation Intelligence, it doesn't matter how long the phone call is.
Whether it’s a five-minute or a 45-minute phone call, being able to know in a few sentences, this is exactly what Ryan and Dylan were talking about, is really magical. And that wasn’t possible in an accurate, reliable way six months ago.
So, what does the future hold?
Dylan: We're just at the beginning of what will be a really exciting couple of years in the AI space.
Ryan: Yep, I never could’ve imagined we’d be where we are right now. It’s just the tip of the iceberg.
Dylan: Being able to unlock voice data and leverage that with powerful large language models — that's when you arrive at this huge impact. It's really exciting to see the impact that the AI systems we're building are having in the real world.
Put AI-powered Premium Conversation Intelligence to work for your business. Try it free for 14 days!
