Former AngelList Exec Explains What Startups Could Teach Enterprises About AI

There are plenty of people betting big on how technologies like artificial intelligence will change consumers’ lives, but Ash Fontana is strictly putting his money on line-of-business applications.

Fontana, managing director at Silicon Valley VC firm Zetta Venture Partners, was in Toronto last week as part of the CDO Summit. The former head of fundraising projects at AngelList, he was key to the organization’s expansion into Europe, created the first startup “index fund” and developed its funds management division. That means he has considerable expertise in assessing where some of the most likely areas of growth in technology will emerge.

“A lot of what we see startups do can be done at a larger level,” he told the audience, citing line-of-business applications as a hotbed of potential opportunity. “They can create a virtuous loop where as you use (the application) more, it gets better. You can make a correction to a prediction, use a bit of data and deliver a result. You can retrain a model or tune a parameter.”

Examples of such applications in market today include Zetta-funded companies such as Clearbit, which focuses on “contact enrichment,” among other areas.

“If you have a marketing database with names and phone numbers but only e-mail for some of them, they’ll find the rest of the addresses. They fill in the gaps,” he explained.

Then there’s Lilt, which automates the translation of of business data through machine learning, or Tractable, which uses AI to study images of car crashes taken on site to improve how claims are handled and lower repair costs.

Fontana suggested that may enterprises could develop similar game-changing capabilities, but first they need to use data science to determine what kind of predictions they can make with the information they already have. If the predictions are “of commercial or industrial significance,” he said, they can be developed for LOB users.

“Software developers today are more like software assemblers. They’re not writing a lot of custom functionality,” he said. “It’s about taking third party data in flight as it moves out of a database to an application. You can find and add data to your applications that you don’t have. Things that may be useful for an algorithm in your organization.”

What if you don’t have the right data, though, or enough of it? Fontana discussed how startups routinely create “customer data networks,” whereby if customers agree to submit data, they will be able to use any AI-based applications the company creates free of charge. Customers can generate considerable data about how an application works, he added, which can be fed back to improve performance. Some organizations also offer consumer-facing apps that offer useful functionality, like weather information, but can also bring data in for enterprise purposes. Finally, he said large businesses should look at complementary data partnerships where they exchange information with trusted third parties.

For now, much of what AI is offering the enterprise is more iterative in nature, Fontana said, but as machine learning and other technologies are used to assist with medical diagnosis or issues with power supplies, that could change quickly.

“We’re moving into areas where it’s more crucial that the AI is right,” he said.

Earlier this month, Zetta announced a $100 million fund that will go towards some 20 early-stage AI companies.