From the millions of Amazon Alexas to the self-driving car, new products are coming to market infused with machine learning. The innovation offered by machine learning techniques are real, and they will changed the SaaS world. But how? How can startups use machine learning to their advantage?
There are four broad applications of machine learning:
- Optimize - this morning, fastest way to travel from Sand Hill Road to South Park in San Francisco is highway 101. The job requisition for an account executive on our website uses too many clichés. To close more business, speak slower, talk about pricing later in the call, and use this case study.
- Identify objects - the photograph you just took with your smartphone contains a cat. Find all red plaid woolen shorts in an ecommerce store. The CT scan shows high likelihood of Parkinson's Disease.
- Detect anomalies - your credit card shows a $10,000 charge for a piano from a store in Nairobi. Your server cluster is operating at historically high CPU usage. Customers are responding to this morning's lead generation email at 25% greater rates than last week's campaign.
- Segment data - customers who come to our product through the mobile app store show 15% higher engagement.
These applications alone make for tremendous advances. But, combinations of these applications lead to incredible things. Object identification + anomaly detection + robotics = self driving cars. Or brick-laying robots that erect walls three times faster than humans.
I've written before about the monstrous acceleration in machine learning innovation. The monocloud vendors (Amazon, Google and Microsoft) are rapidly innovating, publishing breakthrough results and offering APIs that leverage this new research for pennies and nickels. Consequently, every startup can use these technologies for just the cost of a few ramen boxes.
But just plugging those APIs, buying the .ai domain name and inserting the words machine learning/artificial intelligence to your sales pitch isn't sufficient to succeed. Rather than trump machine learning, make the technology disappear into the product. Create that magical moment with the user.
The best sales and fundraising pitches describe a startup's value proposition without mentioning machine learning. Instead, they focus on how the product increases revenue, decreases cost, or wins the buyer a promotion.
We've invested in more than 20 companies leveraging machine learning to create enduring and category-defining companies, from Stripe to RelateIQ, Chorus to Caspida. As we look to invest in others, we've identified five attributes of companies to invest in:
- Proprietary access to data - the algorithms are off the shelf and available to everyone. Creating proprietary data through product usage, perhaps as an event-driven SaaS product, or through key partnerships is essential to creating sustainable competitive advantage.
- End to end applications, not platforms - the monoclouds are likely to win the ML-as-a-Service business. They have more researchers, lower costs of infrastructure, and far more marketing dollars than any startup. End-to-end applications offering revenue increases and/or cost reduction are much easier and a better path for a startup to go-to-market.
- Strong GTM enabled by ML - ML has the potential to be a technology innovation leading to a go-to-market advantage. By changing the way a buyer evaluates software and potentially reducing the cost of customer acquisition, ML based products can disrupt. But a technology innovation alone is not enough.
- Experts in the field - Yes, you can use the monocloud APIs out of the box. But those systems are tuned to be as broadly applicable as possible and generate pretty good results. To deliver an exceptional experience, a startup will need experts in speech recognition, natural language processing, or whatever their core discipline might be.
- Algorithmic advances - every once in a while, we may invest in a company with a fundamental algorithmic advance unlikely to be replicated elsewhere.
Like the database and the graphical user interface before it, machine learning is a new enabling technology that will change the way we build and sell software broadly. And though the terms might be the zeitgeist of the day, rapidly becoming corporate cliché, the impacts of the technology are just starting to be understood and leveraged.
Adapted from my presentation at Saastr.