Artificial intelligence has captured public imagination, dominated media coverage, and driven furious volumes of investment and acquisition activity. In the midst of this hype cycle, spotting the difference between phony wannabes and true investments can be a challenge.
We interviewed seasoned VCs from top firms like CRV, IA Ventures, Two Sigma, and more to find how these successful investors evaluate artificial intelligence startups. If you’re a founder thinking of starting an artificial intelligence company, be sure to have solid answers for all of these key questions.
How Top VCs Evaluate AI Startups
1. Is artificial intelligence core to your company’s value proposition?
“Many companies who can’t raise money try to shoehorn themselves as AI companies,” warns Varun Jain of Qualcomm Ventures. Jain has seen pitches ranging from AI-powered wifi routers to AI-powered juicers and everything in between.
In many of these cases, AI is an add-on feature and not core to the company’s value proposition. “Traditional wifi routers can use artificial intelligence methods to detect anomalies in network data and label those errors, but this feature does not materially change the value-add,” explains Jain.
By contrast, Qualcomm invested in Clarifai and Cruise Automation (acquired by GM). Cruise provides intelligence to power self-driving cars while Clarifai leverages advanced deep learning and computer vision techniques to identify objects within images and video with high accuracy.
2. How credible is your technical team and pedigree?
According to Max Gazor of CRV, “companies with significant and novel AI technology will have pedigrees from strong research labs in academia or come from reputable industry groups like Google Brain or Facebook’s AI teams.”
CRV’s commitment to technical pedigree is reflected in the extraordinary experience of the founders they invest in. Rod Brooks of Rethink Robotics was the founding director of MIT’s artificial intelligence lab and as well as a founder of successful robotics company iRobot (NASDAQ: IRBT). Cynthia Breazeal of Jibo previously founded the Personal Robotics Group at the MIT Media Lab and is a world-renowned expert on social robotics. Oren Jacob of Pullstring was previously CTO of Pixar and worked alongside Steve Jobs since the early days of the company.
David Cheng of DCM Ventures adds, “At this point in the industry’s lifecycle, there are a limited number of AI experts available who have the requisite experience from large companies or top universities to build truly innovative solutions. This scarcity allows us to be highly skeptical if a team purports to use AI in their product without a team that matches.”
3. Do you solve real business problems that customers will pay for?
“One rule I’ve found is: the more the CEO talks about AI, and not about their customer’s problems, the less interested I get,” quips Michael Dolbec of GE Ventures. “We fund valuable outcomes, not science projects.”
Every single investor we spoke to agreed.
“If I had to pick one, domain expertise trumps machine learning expertise,” added Brad Gillespie of IA Ventures. IA Ventures invested in Vectra Networks, a cybersecurity company headed by experienced domain experts who focused heavily on solving top customer problems and maximizing usability for security analysts.
A competitor to Vectra emphasized their sophisticated machine intelligence to customers, but the buyer feedback was: “These guys are smart but they don’t understand my business. Their product has lots of bells & whistles, but I can’t understand what it does.”
Solving business problems effectively requires a team to think beyond narrow technical approaches, but also to focus on and own a specific business domain and function. Colin Beirne of Two Sigma Ventures points out that “using an ensemble of different techniques is required to solve most hard problems today, but targeting A.I. on a narrow domain space reduces the complexity of what it has to learn to understand.”
4. Do you have a relevant, proprietary, and scalable data source?
Jain of Qualcomm Ventures always asks potential investments: “How do you source data? Are you relying on big companies to give you data or do you have an independent manner of sourcing your own?” Both methods can be viable, but independence is strongly preferred.
Self-driving cars are traditionally tested in suburbs, parking lots, and contained environments that do not reflect the reality of driving. Qualcomm’s portfolio company Cruise Automation captured critically missing data by operating test vehicles, monitored by human drivers, in urban environments. Similarly, their other AI investment Clarifai started with a popular consumer app that enabled them to capture unique crowdsourced data before scaling further to work with business-specific data.
In addition to data sources being unique and defensible, they must also be relevant to the challenge being solved. According to Dharmesh Thakker of Battery Ventures, “next generation artificial intelligence depends on the complexity of the data you are mining. Unstructured images, video, and audio data is far more difficult to mine than text.” Thakker also considers whether a company works with fast-moving data or static data. Algorithms for fast-moving data, such as the real-time images processed by a self-driving car, are often much more complex.
Finally, a team must demonstrate that they are continually improving their performance based on their unique data. Qualcomm’s Jain checks in periodically to observe whether a team can “showcase the ability to quickly process training data and optimize efficiently so that systems are more robust.”
5. Have you built unique technology or do you mostly leverage open-source frameworks?
“The extent of which companies are leveraging open source frameworks versus development of their own proprietary technology tends to be a giveaway,” observes Suresh Madhavan of Verizon Ventures. “Leveraging open source will let you analyze some superficial relationships, but it’s unlikely to be at the level that is required to solve hard business problems.”
Cheng of DCM Ventures agrees. The investment team at DCM relies on a robust network of industry advisors and technical experts that “help vet technology stacks, data architecture, and identify whether or not the team is properly approaching data collection, storing, parsing, or annotation. They also help sniff out the phonies.”
6. Do you have a sticky product and robust revenue?
Sumant Mandal is a Partner at March Capital and also co-founder of The Hive, an early-stage incubator focused on AI-driven startups. “If you do not bring at least a 5-10x improvement in efficiency, then it’s very hard for a new company to break in and to bring value to investors,” emphasizes Mandal, who advises that startups think in terms of their customers’ revenue. For example, if you want to apply AI to recruiting processes, Mandal suggests you ask yourself “If I provide a 5x improvement in efficiency, will this result in a 100x improvement in revenue from the people my customers hire?”
Additionally, he cautions that value improvement must be delivered to customers “in a way they can consume, such as in the form of a dashboard or actionable insights.” While cybersecurity is ripe for disruption by AI due to volume of data and dearth of human talent, Mandal warns that “security analysts do not need more alerts.”
Even if you have a desirable product, getting customers to commit to a single pilot program does not make for a viable business. Kartik Gada of Woodside Capital looks for diversified revenue and diversified customers: “Is your revenue robust and recurring? Do your customers want the same or more of your solution?”
7. Do you have a diverse team that includes business experts and sales & marketing professionals?
Last, but not least, investors look for diverse teams that can address all the challenges of starting and scaling an artificial intelligence business. Kiersten Stead of Monsanto Growth Ventures explains that “companies who have been successful hired diversely” to include domain experts, business leaders, and salespeople, not just engineering teams.
In contrast, Stead observes that homogenous startup teams, especially when comprised solely of AI researchers with no industry-specific experience, tend to fail more often. This is particularly important for outsiders seeking to address the niche applications that Monsanto invests in, such as agricultural operations and genetic breeding.
“Technical AI teams can’t relate to sales people very well, and vice versa,” she emphasizes. “We look either for an experienced AI founder with an established career who can kick it with a wide range of people, or a stacked team.” Sales and marketing are often overlooked in tech-heavy AI startups, yet sorely needed for success.
“The biggest mistake AI startups make is to skimp on marketing,” warns Gada of Woodside Capital. “Most customers don’t know they need their product.”