‍What we say can often be less telling than how we say it. That’s the premise behind Helios, an AI company that generates insights about tone of voice from tens of thousands of hours of recorded earnings calls, analyst presentations and other publicly available corporate recordings. The company markets the data to asset managers and others who use the insights that text alone will not surface.

Sean Austin, CEO of Helios, talks with Bobsled about building data products in the age of AI, what data companies can borrow from software development, and how he thinks about improving user experience for data products.

Tell us about what you’re building at Helios?

We’ve pioneered the fusion of language and tone models to provide full intelligence on any conversation relevant to capital markets. Think about an executive on an earnings call or public communication: by understanding now only what leaders are saying -- but how they are saying it -- we can help inform risk assessments, volatility forecasts and much more.

You have built both software and data products. What’s similar -- and different -- in building a software versus data product?

I think there’s a lot that data and AI companies can borrow from software development. For instance, we’ve applied a lot of the core product management frameworks like Agile to the way we build our products and it’s been hugely productive. Rapid testing, rapid prototyping, working closely with the customer all remain critical to building great products.

Two things that are notably different with a data or AI product. First, there’s the team: we’re investing in roles around ML Ops and not in practices like user experience (UX) design. Second, there’s distribution: partnerships are incredibly important for bringing data to market, and it’s something that we’ve invested a lot in over the past year. 

How do you think about user experience with creating data and AI products without a clear visual interface?

When it comes to UX, I think a lot about how we can reduce friction for our customers as they use our data in their analysis. So we’ve spent a lot of time thinking about how to structure the analytics and output in a way that eliminates as much friction as possible while also not limiting what they can do with the dataset. Our customers might not want raw audio files, but they also do not want us telling them how to trade against our analysis. 

Delivery is another huge part of that equation.  We need to make sure that it’s both seamless for our customers to access our data and easy for our team to support it. Something like cloud delivery is critical for our customers, but we do not want to invest engineering hours on something where we’re never going to be experts. We want our team to focus on their actual expertise. That’s why we were excited to discover Bobsled. If we can get our customers testing and validating the value of our data faster with minimal extra work on both sides, that’s  very, very impactful for us.

Accelerating trials is a big challenge for many data providers. How have you thought about helping your customers vet and validate the value of your data faster?

One big area where we’ve seen improvement is on trial agreements. We learned quickly that the longer the agreement, the longer it will take for review. So we’ve made ours super simple. We’ve also invested in making our data more explicable. We want to make sure we can generate documentation so someone can understand it quickly. It’s very clear that the more time you save your customer, the faster they get to discovering the value of the product and eventually buy,

How has the explosion of large language models (LLMs) impacted the way you're building your business?

All of the hype over the past few months has educated the market about the power of AI and in doing so has eliminated a big hurdle for us when we meet potential customers. It’s also allowed us to start to build features that make our products easier to consume. For instance, we’re starting to use these LLMs to summarize the data we’re delivering. We generate a ton of analytics, and the ability to summarize it for customers automatically just removes another point of friction in our customer journey. We anticipate having something that could leverage Generative AI for summarization by the end of Q3.

‍What advice would you give to other entrepreneurs looking to build an AI or data business today? 

This applies generally, but is particularly important for data companies: get your data in front of customers as soon as possible. Customers use data in a lot of ways you would never expect, so it’s just hugely important to get that feedback as early on in your development process as possible. You’re not building a dashboard or an app, so you don’t get to control what others do with your data. That can be hugely powerful if you use it correctly.