Livingbridge AI series: In conversation with T-Pro – how to build a competitive edge using AI

In a world where there is huge buzz around generative AI, where anyone can easily use ChatGPT and new models are coming out every day, many business leaders are asking: ‘where does it make sense to utilise AI in my strategy, and how should we get started?’

To get some answers, we spoke to our portfolio company T-Pro. This Dublin-based healthtech is a brilliant example of a business that has already used AI to help it deliver better clinical outcomes, drive efficiencies, and really stand out in the market.

Founded in 2012 by Jonathan Larbey and Mark Gilmartin, T-Pro is a leading provider of medical dictation, speech recognition and clinical documentation software. T-Pro didn’t start out as a software company, however. It began as a services business with traditional typists creating medical transcripts for hospital clients. T-Pro then identified an opportunity to improve the service using AI and machine learning technology, which has since provided a significant competitive advantage in the market.

Our Investment Director Amy Hunter recently sat down with Jonathan, T-Pro’s CEO, and the company’s Head of Speech and Machine Learning, Conrad Bernath, to get their insights on working with AI, including how to get started, and some key lessons they’ve learned along the way.

Amy: Can you explain a bit about T-Pro, your history, and what inspired you to first start integrating AI?

Jonathan: We started the business about a decade ago now, and originally it was more of an outsourcing business. We were looking to produce transcription documents more efficiently and we very quickly realised that our model wasn’t very scalable, so we made a decision pretty early on, around 2014, to invest in software.

We started with process automation and machine learning, speech recognition and a bit of natural language processing (NLP). That developed over the years, and we reached a point where the tech became commercialisable, which created a new revenue stream and a real point of differentiation for us. Software is now the mainstay of our business – it accounts for the majority of our revenues.

Amy: Given that barriers to using AI have significantly reduced, how do you think companies can gain a competitive advantage?

Jonathan: We’ve found that building our own speech engine and proprietary models based on high-quality, niche data has helped to deliver an edge.

Many companies plug in off-the-shelf speech technology, so they have no control over the technology and it becomes a ‘black box’ in their workflow. Having our own speech engine allows us to spot-fix things and make amends, or create new specialised models. This means we can provide a real enterprise solution.

Conrad: Being independent [from large APIs like ChatGPT] has definitely helped us move very quickly and compete. For example, our engineers can spot a new AI development and incorporate it without waiting for an external software update. I would say having this ability helps companies gain an advantage.

Amy: Lots of businesses are unsure where to begin with implementing AI. Can you go into the practicalities of getting started?

Conrad: If you’re not sure where to start, I would suggest looking for a discrete real-world problem in your niche that needs a solution, and then seeing if AI can help.

With AI there is a lot of hype, and there are many people out there trying to solve problems that don’t exist. So try to find a small, real problem in your sector that you can actually solve. It’s important to have a big vision, but start very small.

In terms of approach, creating Large Language Models (LLM) takes a lot of resources, so I wouldn’t embark on this until you’ve proven that there is definitely a value-add provided by the AI. So, to work out if AI will be a worthwhile investment for your company, I would develop an initial proof-of-concept using a third-party service. This will save time and money.

The fact that the AI industry community is a very open one is helpful here. There is so much you can open source, and the barrier of entry is getting lower and lower. This is really helpful for companies looking to get started with AI. I always say: focus on the value that you want to create, and not on the tech itself.

Amy: Where does data come into the picture?

Conrad: Great question. The most important part of anything with AI is the data. To build a specialised model from scratch, you need good quality data and a lot of niche domain expertise.

Our main advantage at T-Pro is that we have access to a full ocean of in-domain data from our work, and we can use it to fine-tune our own models. We can adapt models to a specific hospital or speciality or even an individual. So if you’re a company that has very specialised data, it’s very valuable. It can be the core of your competitive moat. Try to work with that, and get a clean and quality dataset.

Amy: Can you share any insights into key pitfalls to avoid along the way?

Jonathan: It’s important to iterate quickly and have clear direction. Around 5% of what you do actually gets into production with machine learning and data science. This stuff is notorious for being like a swan: there’s a lovely model on the surface, but 100 iterations underneath that failed. It's just this iterative process. If you have the right people and set a clear direction, you will get there.

Conrad: One of the lessons you learn working in AI is that there are so many things happening every day, and the space moves so fast that it’s really difficult to keep up. I’ve learned that it’s helpful not to get stressed by this, and to remember that no one has the resources to test everything. You need to learn and discern what might work, but be selective, or it will be too much.

Amy: And were there any approaches or decisions that you’ve found essential to success with AI and machine learning?

Jonathan: There have been a few things that we’ve found incredibly important.

Firstly, we made sure to underwrite the technology with a human-in-the-loop aspect, which is essential in any sensitive setting like healthcare. We build trust with clients through offering hybrid human and AI assistance, then iteratively move further towards technology at clients’ own pace. That then unlocks additional savings, more value, and more efficiency down the line.

Hiring is also massively, massively important. And we’ve found that if an academic has a commercial head, they can be the best hires. Our original machine learning lead came from academia and led on building our engine for years. Taking this route can be good for the academics, your company, and your brand.

Amy: Where do you think AI is going now? And how are you planning for the future at T-Pro?

Jonathan: In AI, I think the next 18 months is all going to be about size. Transformers are going to reduce the size of the models and that will be huge.

At T-Pro, we’re about to roll out a new ambient speech assistant that will dramatically reduce the administrative burden on clinicians. It will record entire medical consultations, then produce a transcript and documentation for doctors to review and sign-off. The software will also be able to give suggested diagnoses during the consultation on the basis of statistical probabilities. It’s all exciting stuff.

Amy: We invested in T-Pro last year. Are you happy to share a bit about how that’s going?

Jonathan: Things are moving at a real pace. Our platform is now used by over 600 healthcare organisations across three continents, and we’re pursuing an active M&A strategy and further geographical and product expansion. We recently acquired two Australian startups and launched our Melbourne HQ.

Livingbridge has really accelerated our journey, and not just from a funding point of view. It’s made us a lot more strategic and it’s helping us scale properly. When we do an acquisition or investment in new tech, we’ve now got much better corporate governance and better advice. The relationship has really moved our business forward, and Mark and I see it as a massive benefit. In fact, we probably should have done it sooner!