Machine learning is a trending topic today and for good reason. It has enormous potential to transform entire markets and industries. But there’s also a lot of hype surrounding the technology. As an investor, I look for four key characteristics that I believe distinguish winners who are successfully leveraging machine learning:
Successful machine-learning startups will be the ones targeting vertical applications with a clear need for the technology. The consumer packaged goods industry is a good example. Machine learning can more accurately predict inventory levels to better manage the supply chain, reduce inventory costs, minimize excess capacity requirements, and eliminate stockouts. According to an Accenture study, machine learning can lead to a 4.25x improvement in delivery times and a 2.6x improvement in supply chain efficiency.
Significant manual intervention implies that there is a real opportunity to optimize with complex prediction algorithms. In the same supply chain example, today analysts estimate inventory needs based on some historical data but also a lot of intuition. By leveraging data like production times, sell through rates, and others, learning models could more accurately predict future needs.
Startups need access to significant amounts of data to train machine learning models effectively. Companies that can either partner with large, established corporations to leverage their data to learn, or that create a product that entices users to input their own data, will win.
Algorithms will continue to be open-sourced, which makes proprietary data mission-critical. Input and feedback to a system improves its accuracy and creates a moat. Therefore a product should incent humans to provide feedback on its predictions and recommendations. For instance, Facebook’s photo-tagging algorithm learns from people who either accept or reject suggestions about who is in their photos.
Here are some of the verticals where I believe machine learning can have the greatest potential:
Machine learning will improve outcomes and reduce costs across the healthcare value chain. The potential to improve diagnoses, reduce errors, and streamline the drug discovery process is exciting. Patient data can be used to detect diseases early and to personalize treatment plans. Pharmaceutical and biotech companies can use computational methods to quickly and efficiently discover new drugs that are more effective than current ones on the market.
Machine learning can improve several aspects of the supply chain, including demand forecasting, market trends, trade promotions, and new products. It’s hard for companies today to assess changing market patterns and fluctuations to inform business decisions and to accurately make predictions.
Industrial IoT is a ~$12 billion market and adoption is still early. According to a global Genpact survey of 173 senior executives, only 25 percent have an IoT strategy and only 24 percent of those are happy with the execution. These executives are looking for machine learning enabled solutions to improve yield rates as well as reduce inventory and finished goods levels, driving real cost savings and profit opportunities.
Compliance in financial institutions is a huge market. JPMorgan alone has paid $36 billion in settlements and fines since 2008 and has hired 8,000 compliance and control employees. Machine learning can improve processes across client and employee compliance for banks and other corporations that must abide by audit and compliance regulations.
The complexity of analyzing speech has put voice on the outskirts of the machine learning trend despite being a key component of business workflows. 2013 research from NewVoiceMedia reported that $41 billion is lost annually due to call center inefficiencies. There are 2.4 million inside sales reps engaging in many millions of hours of conversations every year. So it’s clear there is significant opportunity in automating processes across call centers, meetings, and sales and marketing.
Insurance is a large and wide-ranging category where machine learning can help insurers deliver more targeted products at lower costs. For example, auto insurers can use driving and other behavioral data to individually price premiums or use better fraud detection to lower their overall cost structure. Consulting firm KPMG describes machine learning as a “fundamental gamechanger” for the insurance industry.
New data and segmentation models are unlocking financial products previously unavailable or undesirable to millennials (e.g. credit products). Additionally, smart, automated systems are lowering the cost of personalized advice for consumers by tracking behavior and offering suggestions based on preferences and goals. Erin Shipley and TX Zhou wrote a great piece in Techcrunch about the impact artificial intelligence will have on finance, including driving financial health and literacy, through personalized recommendations based on a user’s behavior.
A major limitation of traditional education is that a teacher has to teach one standard curriculum to an entire class, despite varying levels of understanding and different learning styles among students. What if companies leveraged data to help parents and schools identify problem areas and personalize lessons and pacing for each student, providing a tailored plan based on their problem areas and styles? Not only could this revolutionize the state of education in America, but it presents a significant economic opportunity. As of 2013, the U.S. was spending $620 billion annually on public education, with almost 50 million students enrolled in public schools.
While I’m very excited about the opportunity areas outlined above, there are also many companies that have popped up that I am not excited about today.
AI and machine learning are never the end goal. It is about the use case and enabling it via these technologies.
Technology is simply not advanced enough yet to give us positive experiences via generalized chatbots.