Artificial Intelligence Applications - Three Themes That Drive Value
Speak with enough investors in the AI and emerging technology space, and you start to get a sense for areas where these firms are looking to invest, but perhaps more importantly why they’re making investments in particular companies.
We set out to determine: What are the functions or characteristics of companies and innovations driving the next big investments in AI and machine learning (ML)?
Below are three meta-trends that we found between our various investor interviews, along with supporting quotes and comments from the investors themselves.
The “Proprietary Data Plume” of Defensibility
No theme was more prevalent across our investor interviews than that of data.
This is an easy theme to misconstrue, because collecting data in-and-of itself doesn’t impress any investors. Neither does the collection of unique data. The kind of data that might differentiate and drive business value for a company or application must be:
- High business value: data around sales staff success metrics may be of much more value than that of, say, server logs.
- Perpetual, building with momentum: data collection that plays to a business’ advantage isn’t a one-off collection; rather, it’s a perpetual cycle. Facebook users, for example, continuously engage with the platform and provide Facebook with direct feedback on the impact of their features and marketing.
- Proprietary: no other companies have access to Google’s search data, and this allows Google to more easily maintain search dominance and optimize their marketing strategies without hinderance from competitors.
The term “proprietary data plume” is one that I picked up from Ben Narasin at Canvas Ventures (a man of many apt phrases), and it roughly lines up with the factors outlined above.
Narasin referred to data as “the new oil”: “What I need to understand is when you’re feeding that engine...are you putting high octane fuel in this thing or unrefined oil?...Data is now the universal source of truth, and I think people understand that data is the new oil.”
Here are a few examples of this “ data plume” idea in action:
1 – First, in our interview with Gary Swart of Polaris Partners, he mentioned an example with enterprise software InsideSales, one of Polaris’ investments. The company collects data about leads, follow-up emails, phone calls, and appointments, and uses this data (along with context about the businesses and users on the software) to better predict lead scores, sales projections, and best next-actions for sales people (i.e. when to call a particular client).
2 – Second, we’ve had a number of interviews with infosecurity companies aiming to scale the impact of “anomaly detection” across client companies. For example, PatternEx CEO Uday Veeramachaneni aims to detect features and patterns of cyber attacks across hundreds or thousands of clients, eventually building a kind of immune system for all PatternEx customers by allowing them to be warned of similar attacks, even if it’s the first such attack on that company.
These proprietary “plumes” of useful business data become a kind of defensive moat around a company, giving them a potentially unmatched ability to drive decisions and results for clients or users. In this way, the “winner takes all” dynamic of a marketplace (like eBay, AirBnB, Houzz) extends outward to the “marketplace of data”, giving some enterprises an advantage of critical mass that competitors just can’t match.
Having the right kind of data and cultivating it at scale appear to be on the minds of most every investor we interviewed. In most cases, investors would rather know about your reliable plans for collecting and analyzing data over the ins and outs of your innovative prototype or that new ML algorithm you’ve been engineering.
“(Quality of) data is the lens through which we really evaluate our early-stage opportunities...we’re going to invest in the data-driven software 10 times out of 10,” said Flomenberg.
“A company we recently invested in...they're using time series data to make (manufacturing) operators more efficient and more effective without having to wait for data scientists...and at scale, they have data assets that nobody else has; so it’s ‘you’re gonna pay for the right to give me your data’, as is everybody else; everybody else is going to benefit from that aggregate data; and then at scale that data becomes...a very valuable asset,” said Polaris Partners’ Gary Swart.
Whether that data is proprietary or gathered through research or open-source methods is another question, though at present it would seem that owning the data — like the biggest players do (i.e. Google, Facebook, etc.) — is the winners-take-all scenario.
“As we think about near-term investments, we’re looking at areas where you can actually control that data...areas like IT systems, there's much more consolidation there, there’s the fear then that if you develop a system on top of other people's systems of record, you have to own that system of record — that’s more of an investor domain ie. do you have to own the data or do you have to sit on top of lots of other disparate data sources to become a valuable player?” said Floodgate Capital’s Ann Miura-Ko.
The hooked-on-data trend is prevalent across industries, and healthcare is a veritable goldmine of potential opportunities, as voiced by Zhuang: “All of this data — that’s why we're very bullish on this sector, it’s such a great opportunity, such a great time that we can combine the data and software to really improve on diagnostics, improve on disease monitoring, improve on personalized medicine, improve on clinical workflows...some before 2030 and some after.”
New Solutions to Old Problems
It starts with common sense, tossed with innovation and vision. There are plenty of age-old problems across industries that have been more or less unsolvable until the recent boom of big data and quicker, cheaper computing power.
Comet Labs’ Saman Farid advised, “We’re looking at the most unsexy technologies or industries that exist and seeing how AI can relate to those, because things like self-driving cars, there’s a certain amount of flashiness….but there’s actually so many fields that are kind of in the background that enable us to live the lives that we do.”
Fields like construction, agriculture, health-care, and food and beverage are just a handful that Farid referenced. He emphasized solving problems at the granular level and macro levels but avoiding the too-broad and overly abstract.
For example, instead of thinking how can I build a robot that manages a farm, think instead about innovating specific processes that can’t be done by one or a group of humans at scale. Drones for surveying and analyzing images of areas of land for best crop yield, for example, or automating the entire process of records management in healthcare and changing the clinician workflow and efficiency of patient care.
The shrinking size and cost of sensors, the diminishing cost of computing power, and improvements in wireless communication have made it possible to re-imagine old industries that were once “data dead” to now be “data rich.” Companies like UpTake and Tule are interesting examples of modern machine learning applied to previously stodgy sectors.
It’s often less about who has the most amazing core AI technology and more about who’s applying the technology in a new or genuinely useful way — or to a problem thus far ignored. Ben Narasin puts it nicely: “In my forward-looking lens as an investor...it’s less likely that I’m going to find the next big monolithic AI engine of the future and much more likely I’m going to find the next great expression of it that has been turned into a product and a company and a highly valuable resource.”
Farid also talked about this idea in how AI companies are making use of data in historically stodgy industries. “There’s really a lot of opportunities to be gained from taking the insights that you gain from AI and applying them in more meaningful ways, and on the other hand rethinking your whole business model...we’ve seen that from agriculture to construction to logistics in warehousing,” said Farid.
Enhancing Humans – AI as Augmentation
While automation has been a buzzword for the past few years, not every investor is a keen as the next on the notion — both in terms of the broader idea and in how AI is used as a tool.
Li Jiang, an investor with GSV Capital, stated, “One area I think that’s exciting and maybe scary is really products that are trying to replace human staff, like in sales for example...I think that’s one area where it’s starting to emerge and there’s lots of promise.” Others, like Farid, hold a different opinion. “When people say that they want to automate a task or create some sort of easy fix to what they’re working on, they’re generally thinking about automating something, and that in my opinion is a big misuse of AI in general,” he said.
While automation has already taken hold of some industries — manufacturing, for example — what seems to be more in line with investor (and maybe more importantly consumer and industry sentiment) is the augmentation rather than the outright automation of human abilities. Perhaps nowhere is this more evident than in the healthcare industry, where many high-level professionals in high-risk environments are wary of the AI touch.
“In terms of where it’s going to be applied first, it’s very hard to tell extremely smart teams of doctors they're going to be replaced, and so where you actually have a deficiency of expertise, people are more willing...and it’s going to be hand holding and not a replacement,” said Steve Gullans of Excel Venture Management. “As an investor we have to pay very close attention to who’s actually going to benefit financially from this — is it the physician, the insurer, the hospital system — and that really is a big determinant of where it gets adopted first and how quickly,” he added.
Additional Fodder for AI Startups
What’s the most important thing when pitching your AI startup to an investor? Having a clear business plan gets the vote, based on takeaways from AI and machine learning (ML) investors and venture capitalists whom TechEmergence has interviewed over the last three months.
“If you think about a startup as this systematic process of reducing risk for whatever you’re going after, the most important thing that they could spend their time on right now is the business model, not necessarily once you have a prototype,” says Jake Flomenberg of Palo-Alto based Accel.
Building this solid foundation applies across industries and regardless of innovation, though admittedly some spaces — like health-care — require different and additional strategic analysis and planning.
Shelley Zhuang, an investor with Eleven Two Capital that is focused — among other things — on investing in data-driven healthcare innovations, says, “I think it’s absolutely critical that a company clearly understand where their existing solution fits into clinician workflow.” This clear understanding is a predicate to having a regulatory and reimbursement strategy, which usually involves establishing and publishing clinical validity, as well as thinking about how to address insurer claims once your innovation is brought to market.
While a solid business plan, unique data sets, and innovative problem-solving applications might occupy most of a founder’s cognitive space, the investors we interviewed touched on some other useful considerations for AI and ML startups.
One is a matter of geography — where you decide to launch your bold business idea. While many tech companies flock to Silicon Valley with the flash of an idea, Ben Levy of Bootstrap Labs believes it’s important that founders know they can start anywhere. In fact, if a founder is originally from a different state or country, it’s often to their advantage to leverage the resources with which they’re familiar.
“We only invest in companies in Silicon Valley, but founders can be from anywhere in the world. The company could have been started elsewhere, which means we’re completely in favor for you to maintain an R&D somewhere in the world where you come from, where you speak the language, where you know the universities...it’s all about speed of execution and you want to have the right talent,” said Levy. Of course, this also means considering long-term objectives such as best place to scale and best sources for investors — and Silicon Valley remains the holy grail on this front.
Place and cultural differences are also important when designing new technologies, which goes back to using having a deeper understanding of hometown, state, or country. For example, in the world of chat bots and natural language technologies, we’ve talked before with Baidu USA’s Adam Coates about the differences between mobile users in the U.S. and those in China. Tak Lo, an investor with Hong Kong-based Zeroth.AI, mirrored this idea: “People interact with messenger applications (in Asia) very different than in the West...a lot of people in, say, China actually speak...it’s a very different interface; chat bots can take off in Asia, but people trying to solve chatbots in Asia need to think differently,” he said.
Narasin added another point of reflection on bucking trends; it’s one we’ve all heard before but that sometimes gets forgotten in the excitement of jumping onto the AI and ML innovation bandwagon.
“What’s hot now — that’s the wrong question; whatever is hot right now is too late for you; you need to design what’s going to be hot three or four years from now,” said Narasin. Instead, he advises taking the mindset of designing three, four, five years into the future — or even further out. Some industries have more stamina than others in terms of the “hot” factor, simply because there is more ground that needs to be covered.
Healthcare, for example, is riper by the day for future innovation, although regulatory, privacy, and other obstacles need to be addressed simultaneously, whereas sales and marketing automation software in general are very quickly becoming mainstream.
Is AI Integration Worth the Time and Investment?
If there’s a single takeaway that we gleaned from interviewing AI investors, it’s that the right data — not the right algorithms — is the key to gaining a business advantage.
Business leaders should ask themselves: 1) "Are we able to create a self-feeding data ecosystem to train our algorithms?" and 2) "Do we have the ability to use those proprietary insights to improve our business outcomes?" If the answer to either one of these questions is “no” for a current project, then AI may not be the solution to that particular business challenge.
Plenty of corporations seek to "invest in AI" in an attempt to stay innovative, but that's not how any smart investor would think. There is no pride or genuine benefit in "toy" projects. "AI" by itself is not exciting, but financial results are always exciting. Whether you're starting an AI startup or investing in a large business project in your organization, think about the likelihood of return the same way an investor would — and choose wisely.