AI is everywhere right now, and so are strong feelings about it. For every person excited by what the technology can do, there's someone else tired of seeing it pushed into every product they use, and skeptical that any of it is being built with users in mind. We understand that frustration. All too often, AI is deployed without enough thought and care, rolled out haphazardly to cut costs or add features that don’t quite work rather than help the people on the other end.
At InnoCaption, we've been thinking carefully about the role of AI for a long time now. Human connection is at the heart of everything we do, and it's incredibly important to us that our users can always trust that we have their best interests at heart. I'm therefore proud to say that our approach to AI has stayed the same even as the technology has advanced: embrace what it does well, be honest about its limitations, and keep humans in control.
Here's a look inside three ways we're putting that approach into practice.
We're proud to be the only captioned phone service provider to offer the choice between AI and live human captioning by stenographers. Users can switch between the two at any time, even mid-call, depending on their needs. But it wasn't always this way.
When the InnoCaption app launched in 2016, every call was captioned by a stenographer. Within a few years, though, we’d begun to investigate the viability of captioning powered by automated speech recognition. At the time, people were skeptical that AI could work for this use case; the technology hadn’t performed well in real-world settings up to that point. But our users had real needs we believed it could meet, like the desire for consistently fast captions or the privacy of a call without a third party on the line. So we decided to move forward.
In 2019, we became the first FCC-certified captioned phone service provider to offer AI captions. Between roughly 2020 and 2022, speech recognition engines improved at a rapid pace, and adoption of our AI caption mode grew alongside it — allowing us to scale to meet demand without being limited by the number of high-quality human captioners we could train and onboard. And by empowering our users with the best of both human and AI captioning, ensuring that their needs are met on every call, we’ve built a service for which demand keeps on growing.
Interestingly, even with significant adoption of automated captions, we have more human stenographers on our platform today than before we introduced the AI captioning option. (There are important reasons to offer both.) When AI is deployed to serve users rather than replace people, everyone ends up busier and better served.
Captions solve half of a phone call: understanding the other person. For users who don't voice or have a speech disability, our text-to-speech (TTS) solution solves the other half. You type what you want to say, and a natural-sounding voice speaks for you.
The challenge with typing on a live call has always been speed, because conversations don't wait. So when we developed our AI Refine feature, the goal was to help users who type to communicate keep up with the pace of real conversations. The struggle was how to do that without letting the AI take over: we wanted something that would help people communicate faster without putting words in their mouth, where the user would always be the author.
Striking that balance took extensive experimentation across different language models. The output had to sound natural, stay true to the user's intent, and keep pace with a live call, all without adding ideas of its own. We’re very proud of the end result. When you type a partial response — even just one or two words — and tap the AI Refine button, the app instantly generates a full sentence that’s carefully tailored to the context of the conversation. You can review and edit that sentence before it's voiced, so what gets spoken is exactly what you want to say. It's a great example of how AI can unlock human potential and make everyday life easier.
Now, for some users, typing on a call feels completely different. Deaf users have told us they can type the way they would communicate in ASL, and AI Refine then turns their words into complete English sentences instantly.
Another place we've put AI to work is in customer support. When you reach out with an issue, an AI agent responds first, because it's available 24/7 and answers within seconds, even at 2am. (We're completely transparent about this: responses are very clearly marked as coming from AI rather than a human.) The AI agent resolves more than 80% of the tickets it's involved in, with a customer satisfaction rating above 90%.
But those numbers don't come from the AI being smart on its own. They come from what sits behind it: a help center of more than 140 articles and an internal knowledge base that our support team has built over years and still updates every week. The team reviews the AI agent's conversations and actively retrains it, creating new responses whenever they spot a question it could be answering better. AI isn't much good without the right knowledge and training behind it, and ours exists because humans did the work first, and never stopped.
Just as important: reaching a human agent is always easy. (Anyone who has been trapped in an automated support loop knows why that matters.) When a ticket does reach our support team, we treat it as a signal that the AI's answer wasn't good enough. So the team digs in, closes the gap, and the next user gets a better answer. An escalation to humans isn't a failure of the system; it's how the system gets better at serving our community.
At the same time, AI hasn't shrunk our support team's role — it’s elevated it. Freed from having to answer repetitive questions, our agents now work as advanced technical support, assisting users with complicated issues and working directly with our engineers to resolve them.
Across captioning, text to speech, and support, the pattern is the same: AI provides speed, availability, and consistency, while human beings provide indispensable judgment and the ability to solve hard problems.
Adding AI to a product is the easy part, but whether it improves the user experience depends on the human care and effort behind it. That's been true of every feature on this list, and it will be true of whatever we build next.
Want to hear more? I talked through all of this and much more with Brian Taylor on This Week in Hearing. You can watch the full interview here.
Paul Lee is the CEO of InnoCaption. He joined the company in 2020 after a career in finance spanning New York, San Francisco, Hong Kong, and London. The son of InnoCaption founder Joseph Lee, he left his finance career to join the family business and carry forward its mission of advancing accessible communication. Under his product leadership, InnoCaption rebuilt its flagship mobile app from scratch, launched advanced text-to-speech features, and introduced InnoCaption Video. Paul is deeply engaged with InnoCaption's user community and he writes and speaks frequently on accessible technology and responsible AI adoption.
InnoCaption provides real-time captioning technology making phone calls easy and accessible for the deaf and hard of hearing community. Offered at no cost to individuals with hearing loss because we are certified by the FCC. InnoCaption is the only mobile app that offers real-time captioning of phone calls through live stenographers and automated speech recognition software. The choice is yours.
