Where do top VCs invest in AI?
The best way to get the answer is to see where they have invested in.
Tealfeed Guest Blog
Driven by curiosity, I wanted to know where the top VCs invested into in the AI space. Over the past year, frankly speaking, our team has not seen a lot of exciting AI-related deals, and most of the stuff that we’ve seen was making some incremental improvement. Therefore, I was very eager to see how the top VC firms were thinking about what AI could do in the future. The best way to get the answer is to see where they have invested in.
I pulled out from PitchBook the list of companies that were founded after 2017 and raised early-stage funding (angel, pre-seed, seed, series-A) from the top VCs. The top VCs are defined by Preqin in 2017 according to capital raised in the last 10 years (2008–2017). See the full top VC list here.
I screened the list manually and took out the companies that were working on the things that we already knew such as autonomous driving, autonomous checkout, and object counting using computer vision. I got a shortlist of 86 companies, each of which is working on something new that we at least have not seen many times before.
By going through the shortlist, based on my overall sense of these companies, I categorized these companies by “vertical”, “the problems to be solved”, and “the value of AI”.
Categorization 1: by “Problems To Be Solved”
Overall, it is really interesting to see what the startups are working on. Based on the “problems to be solved”, I categorized the startups into four groups.
Group 1: AI for challenges derived from the main use case
As more and more companies join the AI competition, some new challenges have come up along the way. For example, people find it difficult to collect a great volume of data to train their models, people find a lot of data is unstructured and difficult to use, people find it difficult to build models without data scientists/ML experts, people find it difficult to monitor the performance of AI/ML models and update them in time, and people ask for explainability of ML models.
Overall, these challenges include “data collection/generation/simulation”, “data labelling”, “model building/training”, “model performance monitoring”, and “model explainability”. Therefore, many companies have developed solutions to address these new challenges.
From the shortlist, these companies include Determined AI, Labelbox, Fiddler Labs, Applied Intuition, Arthur AI, Basis AI, Graviti, Kaskada, Oneclick.ai, Superb AI, and Voicery. Keep in mind that this is not a full list by any means, and we have also seen a few similar startups before.
Beyond the startups listed above, we also met a lot of companies last year working on synthetic data. By generating synthetic data, these companies are helping their clients to lower the data collection costs, test AI/ML models in edge cases, protect privacy, and provide programmable/customizable data (image/video) in a scalable way. Some of the companies include HourOne, Mostly AI, Datagen, Edgecase AI, Synthesis AI, Datagrid, Statice, Hazy, and Yokai.
For companies working on/supporting explainable AI (XAI), we have also seen many over the past year. Many of them are not on the shortlist. These XAI companies include 6Estates, Beyond Limits, Manto AI, Stride AI, craft ai, Kyndi, Senfino, Digite, DarwinAI, Logical Glue (acquired by Temenos already), Optimizing Mind, Z Advanced Computing, DarkLight AI, Imandra, KenSci, Stratify, Tazi, Elemental Cognition, machines, Vian AI, Element AI, Fiddler Labs, BasisAI, H2O.ai, Ditto, Sisu AI, Calypso AI, Rolex AI, Determined AI, and Shapes AI.
Group 2: AI for new/emerging challenges
As the Internet and technologies evolve, our society encounters some new challenges. For example, people care more about their privacy as they have a bigger presence on the Internet, and enterprises care more about their cybersecurity as they use more external services and are exposed to more risks. For another example, as more companies and governments are using face recognition techniques, people are concerned about their identity and thinking about how to protect it. Besides, people are also concerned about online toxicity and fake news.
Therefore, some startups have been established to address these new emerging challenges. As the challenges are new, it might take some time for these startups to figure out a workable business model. However, we still need to solve these challenges.
Among the list, there are 7 companies from this category. They are Armorblox, Blue Hexagon, Canopy, D-ID, Mine, NIghtfall, and Preclusion. Again, this is not a full list, but I mention these companies to showcase how companies are using AI to address these challenges and where we can find similar ones.
Some other companies that I learned last year also fall into this category. L1ght is using AI to detect online toxicity. Fabula AI (acquired by Twitter), Cheq, and AdVerifai use AI to identify online fake news.
Group 3: AI for deep/indirect untapped challenges
As most of the direct or easy-to-see problems have been addressed by the first wave of AI companies, new companies began to tackle the deep and indirect challenges. For example, how to use AI to understand people's behaviour, and how to add a human layer on top of the existing AI.
Therefore, some companies have come to this space using AI to improve autonomous driving systems, to understand employee behaviours, and to coach business leaders.
Three companies on the shortlist are working in this space. They are Humu, Cultivate Technology, and Nestor. Some other emotional AI companies that are in the space but not on the list include Humanizing Autonomy and Affectiva.
Group 4: AI for use cases with smaller market size
This is easy to understand. When people have applied AI to the use cases with the largest market sizes, new companies start looking into the smaller markets.
61 out of the 86 companies on the shortlist fall into this category. I think this is also why we have not seen many great AI companies over the past year — the problems that these new companies are trying to solve are interesting but just too small.
The use cases vary from one industry to another. Some of these “small” use cases include:
AI for automated UI testing
AI for the noise cancelling and hearing strengthening
AI for cobalt (minerals) searching
AI (computer vision) for fish farming
AI (computer vision) for spotting weak trees
AI for gamer performance analytics
You can feel free to open the raw list and see all the 61 new use cases.
Categorization 2: by “Value of AI”
Another way that I found useful to categorize these startups is by the value brought by AI techniques. Across the 86 companies, I found that they could be categorized into two big categories.
Group 1: AI to help us understand better — enabling new opportunities and reducing risks
The core capability that AI adds to human beings is that we can understand complicated things much better than before. There are three ways that AI can help us understand things better.
(1) Understand the environment better. The environment can be the macro market, the physical environment, or the internal business environment. By helping us understand these environments better, AI enables us to find new trends, find new resources, and also find new risks. For example, KoBold Metals uses AI to help cobalt searching. For another example, Gaiascope helps us to predict energy prices and make electricity trading easier by gathering more market data.
(2) Understand consumers/customers better. Before, we use “user profile” to understand our customers or consumers. This might be biased, and it is difficult to find out the deep thoughts that customers may have. With the help of AI, we can now have many more insights about what our customers are thinking and what they will need in the future. For example, Woebot uses a chatbot to assess mental health. Curia uses AI to help the patient describe symptoms, prompt the answers a doctor needs to make an accurate diagnosis, and offer information to ensure the patient understands their condition and treatment options.
(3) Understand employees better. Similar to the AI for consumers, AI is also used to help us understand employees better. New companies are using AI to understand how our employees think, what they are good at and not good at, and how they can improve in the future. For example, Nestor uses AI to coach employees in both knowledge and leadership skills. Humu uses AI to help people become happier, more productive and stick around longer in real-time.
Group 2: AI to help us act faster — improving efficiency
It is not surprising to see a lot of companies are using AI to automate work and replace repetitive human labour work. These companies usually do not enable anything new but save customers a lot of labour costs and time.
Therefore, I am not going to elaborate on this group. Feel free to open the raw list and see all the 31 companies that fall into this group.
- Overall, startups have explored most of the use cases that have large market sizes. The AI competition is entering the second phase — competing by performance (efficiency and/or cost) and product (user experience).
- However, as new challenges coming up, we can still find new and big use cases for AI. For example, using AI to protect our privacy and cybersecurity.
- New challenges will keep coming up, as AI is being widely adopted. From high data collection costs to model performance instability, we can keep watching the new challenges and find opportunities from there.
This article was originally published at Medium and was written by Fan Wen.
Tealfeed Guest Blog