How Real AI startups are different
The potential for AI has excited investors into pouring over $60B in AI startups. Startups have taken the hint and are liberally sprinkling AI/ML buzzwords over their pitch decks. Having evaluated over 350+ startups with AI/ML in their elevator pitch, I found a lot of hype. Many startups claiming to be AI startups just consume AI/ML APIs of the large cloud providers with a thin wrapper. But sifting through all the hay, I did find a few needles and invested in 9 AI startups either as an angel or through my fund, SenseAI Ventures. As I was reviewing the startups, I found 4 factors that were a common theme and I think differentiate real AI startups.
Real AI Problems
Only some problems need AI. For many use cases simple rule-based models would suffice. Real AI is both complex and expensive to develop and thus makes sense only for those use cases which have multiple types of input data and a variety of possible outcomes. For example, you could put AI into traffic lights to make them more attuned to traffic. But rule base red lights offer much the same benefit at much lower cost. On the other hand, autonomous driving with its phenomenal complexity can only be truly built using AI.
Cureskin has built an AI dermatologist app which handles 17 different skin conditions like acne and dark spots to Melasma. The app provides a complete experience that hides all the complexity like capturing enough detail using a phone camera to analyzing the condition and recommending the right treatment regime for the detected skin type. By solving a real AI use case, cureskin is empowering anyone anywhere to quickly resolve their skin problem sitting comfortably at home.
Data is the Moat
Building for real AI use cases requires sufficient quantities of labelled training data for each input stream. Such data sets are hard to come by. We see startups deploy two kinds of strategies to solve this challenge. Technology led strategies enable startups to work with sparse data sets or enrich them using active learning and GANs. The other path is to either partner with institutions that have the data or create a low-cost method to build a rich data set. A creative strategy to build up data is a real advantage for an AI startup.
Vernacular.ai, is working on building the voice engine for local Indian languages. Given the lack of vernacular data, many startups would hire teams of native speakers of different languages to build the required voice data. Vernacular.ai took the smarter path of building a gamified app which tapped into the free time of farmers, guards, drivers, servants… and motivated them to read out lines with small prizes.
Technology is the key driver of value
Using pre-built AI models from cloud providers may be a good way to test a hypothesis or build internal applications. Founders need to go much deeper to build a differentiated AI product. This is challenging as the AI frontier is rapidly changing. New frameworks like Reinforcement Learning & GANs offer potential to go well beyond just learning from and mimicking prior human work. We are now seeing AI create good quality content, develop novel ‘Go’ strategies and bluff in poker. AI startups that meet this challenge and create differentiated IP offer phenomenal investment multiples with easy exits.
Monet Networks provides enterprises, politicians or anyone interested in understanding consumers, a new way of analysing users, their emotions and potentially predict their actions. Their cutting-edge tools to fathom the emotions of the viewer based on their facial cues and psychographic profile, can turn any camera into a survey tool to predict behaviour. Thankfully, their practice of permission based, opt-in user panels ensures both the solution and data are legal and ethical.
Founders look different
We are at the first 10m of the AI marathon. AI shall disrupt multiple industries and change the way we live. Automation solutions are low hanging fruit that create value by reducing costs. Unicorn builders, on the other hand, are founders that have the passion and ability to innovate and create new markets. Sometimes an MBA can be a hinderance in imagining and building such a future. We find that the ideal founder is a visionary engineer with experience in big data /AI/ML and the ability to attract tech talent.
Such founders are rare and I have waded through hundreds of pitches to find the founding teams SenseAI and I have backed. It is well worth the effort both from a returns perspective as well as the emotional and intellectual reward of seeing a genius at work, shaping the future.