What the $%&* is Applied AI?

When Costanoa opened its doors in late 2012, the data revolution was underway. Thanks to the intersection of the internet, mobility and cloud computing, the ensuing information explosion generated unique opportunities for entrepreneurs to deliver products that create and transform industries.

I have seen a lot of changes in the start up landscape in the six years since we opened our doors (or arguably our laptops), but two things have remained true:

1. Delivering extraordinary customer value requires deep understanding of the existing business process. In other words, “ #AI ” is not a substitute for excellent product management.

2. To build a uniquely valuable company driven by machine learning or artificial intelligence, the data strategy is every bit as important as the algorithm.

AI Should Solve Problems that Need Solving

As we sometimes do, let’s start at the end.

• Even if you’re an “AI person”, you need to get out of your bubble and see what businesses are doing in the real world. Focus on solving high-value business problems and be prepared to roll up your sleeves; finding real world problems to solve will take effort.
• They’re also not always the ones most immediately apparent to a technical audience. For example, Aquabyte, which brings machine learning and computer vision technologies to the aquaculture industry, didn’t come up with their idea by sitting in a room staring at the wall. Bryton Shang, the founder of Aquabyte, was curious and independent enough to go to fish farming conferences to explore solutions and build industry relationships. He met with companies to learn what challenges they were encountering and they talked about how they might collaborate to solve problems together.
• It’s easy to assume that the world will come knocking on your door simply because you have certain AI skills. But that won’t take you very far. Go out and think about other verticals and how you can bring your technology stack and data strategy to help solve a broader set of problems.
• Build software that captures users data and then create a closed loop system so the system gets better over time. That kind of test harness will not only improve the data but will improve the algorithms. For example, Demandbase uses their unique dataset to predict where companies should do their B2B advertising. There is no other company with a comparable dataset so they know their insights will be unique.
• And focus on building customer value. Think about how you can build a company of enduring value with an applied AI stack. If it doesn’t deliver better value than the alternatives, then why should customers care? They won’t. We’re already past the peak of the hype cycle around AI buzzwords to the point where the phrase is now almost meaningless.

Great Data > Great Algorithms

Even if a team develops the best-in-class AI algorithm in the world for a particular business purpose, the intense competition in AI-driven applications means that they’ll likely have a fleeting advantage without creating a unique and proprietary data set. AI algorithms are only as good as the data they process, so the two are tightly linked in creating solutions that create value. The world has too many smart people to scrape data from the internet or from public data sets and believe you’ll build a big business; it will only be a matter of time before someone else does the same with a better/smarter algorithm or UX. You’ll end up in a commodity race because everyone else will be able to scrape the same data from the same data sets as you. Public data sources can be a good starting point but innovators must plan to aggregate, curate and originate in order to retain a long term sustainable advantage.

It starts with your data.

In this world of applied AI, one way or another, you’ve got to find yourself in the data pipeline. That can be done by plugging into existing systems, providing intermediate applications that engages users, or partnering with pilot customers.

• Plugging into Existing Systems

Guardian Analytics prevents fraud for banking transactions by integrating into the core banking platform, such as Fiserv or Bottomline. The bank is in the system already so using available APIs to integrate directly and train as transactions let them prevent more fraud with fewer false positives than traditional rules-based system. This approach often takes time since older systems may not have well documented APIs, but Guardian has been able to plug into those systems for a while. Now that it is integrated into 400+ financial institutions and has evaluated billions of transactions, it has an enduring advantage that will be hard to replicate.

• Providing Intermediate Applications that Engages Users

Alation, the leading provider of Enterprise Data Catalogs (according to Gartner and Forrester) uses machine learning to help users find, understand and govern enterprise data. They built a SQL query tool that made it easier for them to see and operate on the data to answer questions like: what does this field mean, are HP and Hewlett Packard are the same thing in this query, who uses this field most frequently and in what Tableau reports is this field used.

Focal Systems uses computer vision and deep learning to improve customer experience and reduce costs in brick-and-mortar retail. Their goal is to eliminate “out of stocks” and deliver a frictionless self checkout solution. They began in local grocers deploying carts equipped with “Wayfinding Tablets.” As the tablets direct customers to specific items in a store, the devices take pictures in different light and angles and builds up the data in pursuit of delivering a reliable out of stock solution that meets grocer requirements. Full deployment of the out of stock solution similarly helps capture the variety and volume of images necessary to deliver self checkout as good or better as the Amazon Go store with much lower cost and available to everybody.

• Partnering with Pilot Customers

Aquabyte decided to partner closely with a handful of key pilot customers. The company built initial algorithms based on fish in test tanks and used that to reach pilot agreements with several Norwegian salmon farms. They are using these partnerships to get more advanced cameras into the water that capture the data required to deliver the first use case — detection of sea lice — in the real world. The basic pitch was, “Let’s get into production so we can solve an important problem today, and train our algorithms on data so we can solve even bigger problems over time.”

These are all companies using applied AI to solve real-world — often vertical — problems. But it is their aggregation of proprietary data that will ultimately allow them to derive useful and actionable insights that create a defensible moat of data assets that nobody else has.