This article is part of our series that explores the business of artificial intelligence
In line with the current state of capital markets, funding for artificial intelligence continues to slow, according to the latest quarterly State of AI report published by CB Insights.
Total funding in AI startups is down 31 percent since last quarter and at its lowest since Q3 2020. Mega funding rounds ($100M+) are down 39 percent quarter over quarter and at a nine-quarter low.
While the stagnation in AI funding will slow down the field, it is also forcing investors to focus more on AI initiatives that are more likely to reach a sustainable business model. A better look at the companies that are getting funding gives a sense of where the industry might be headed in the next few months.
The AI business model
AI startup is a vague term and is often applied to all kinds of companies ranging from those that focus on providing AI tools (e.g., MLOps, predictive analytics tools, no-code/low-code model development) to those that use AI in their products (e.g., an insurtech company that uses machine learning to predict risk).
However, there are a few things that define the success of a business model shaped around machine learning. Some of them are the shared principles of all products:
1) Product/market fit: The product must either address an unsolved problem or provide enough added value over existing solutions to make it worthwhile for customers to make the switch.
2) Growth strategy: There must be scalable channels to allow the product to deliver its value to its target users (e.g., paid advertising, integration with existing applications). These channels must be defensible and make it difficult for competitors to cut into the product’s market share.
3) Addressable market: Investors want to achieve a return on their investment. There must be a sizeable market for the product to grow and reach its target valuation. If the product is too niche and very few customers desire it, investors will not be interested in funding it.
In addition to the above principles, products that use machine learning must also solve a few extra problems:
1) Training data: The product team will need to have enough quality data to train and test its models. In some cases, this data is easy to come by (e.g., public datasets, existing data in company databases). In others, it is harder to obtain (e.g., health data). And for some applications, the data might have nuances across different geographical regions and audiences, which will require their own data-gathering efforts. If the models are supervised, the team needs to have a strategy for scalable data annotation.
2) Continuous improvement: Machine learning models need to be constantly updated as the world changes. After deploying the ML models, the product team must have a strategy to continuously collect data to update and improve the models. This continuous improvement also strengthens the product’s defensibility against competitors.
With these principles in mind, I looked at CB Insights’ report and tried to see if there are any patterns in the startups that are attracting funding for their AI initiatives despite the downturn.
Early-stage funding for AI
The median deal size for early-stage funding has remained steady at around $3 million. In comparison, mid-stage and late-stage deals have seen a quarter-over-quarter drop of 15 and 53 percent respectively.
But the number of early-stage deals has shrunk, which means founders will have a harder time finding cash for their product ideas.
Among the top seed and angel deals mentioned in CB Insights’ report is Voyantis, an Israel-based company that secured $19 million in July to develop its “predictive growth platform.” Voyantis is addressing the problem marketers face as the advertising environment changes with stricter rules on user data and privacy. For example, Apple has recently added a feature to iOS that allows users to prevent advertisers from collecting their device IDs. Without granular data on users, previous rule-based advertising campaigns provide inferior results, which increases the cost per acquisition (CAC). Voyantis uses machine learning on zero- and first-party data points owned by the customer to predict user behavior and lifetime value. The predictions help make informed decisions and improve the ROI of marketing campaigns. One of the benefits of Voyantis is that its predictions can be directly fed as signals into ad networks such as Google and Facebook, which reduces the friction and the costs of adopting the solution.
Eleven Therapeutics, another Israel-based company that received $22 million in seed funding in August, is a biotechnology startup that focuses on RNA therapeutics, a field that has received much attention in recent years, especially during the covid pandemic. The company is developing a deep learning framework for “generating functional data” about the activity profile of siRNA molecules, a problem space that has a vast number of possible combinations. There isn’t much information about the company’s AI technology. But tackling big problem spaces is one of the areas where deep learning excels, especially when it is guided by human intuition. And since the company is working on data generation, it will not have to deal with many of the complications of handling health data. Its financial backers include the Bill & Melinda Gates Foundation.
Spice AI, a U.S.-based startup that secured $14 million in seed funding in September, is building the digital infrastructure for creating AI-driven Web3 applications. It is interesting that the company has managed to attract funding at a time when crypto startups are faring worse than other industries. However, three things are notable about this company. First, it is creating the data engineering infrastructure to index existing data on major blockchains, which means it doesn’t have any major barrier to obtaining data. Second, its founders are Microsoft Azure veterans; its cap table also includes Microsoft Azure CTO Mark Russinovich along with the former and current CEOs of GitHub (acquired by Microsoft in 2018). Having such high-profile figures on the board make it easier to attract funding, even in the hardest of times. And third, blockchain data engineering is a largely unsolved problem and one that will surely lie in wait for Web3 companies as the industry matures. Therefore, this can be considered one of the less-risky ventures of Web3.
Who is getting mega funding in AI?
Among the startups that obtained mega-funding in Q3 2022 is Afresh, a U.S.-based startup that secured $115 million in series B in August. Afresh uses machine learning to help grocery store operators reduce food waste. The company’s platform tracks fresh food sales and helps predict future customer demand. Supply chain teams can use the platform to optimize their purchases from suppliers to minimize food waste. Users can directly use the platform to place orders from suppliers, which makes it easier to integrate it into existing workflows. According to Afresh, stores that use its platform reduce food waste by up to 25 percent. The company already has several thousand customers in 40 states across the U.S. It will use its new funding for growth, expanding its market to other countries and adding new features to increase the value and market coverage of its product.
Another interesting company is Bending Spoons, an Italy-based mobile app developer that raised $340 million in September. Bending Spoons has become popular for its mobile video and photo editing applications, which use machine learning to perform complicated tasks such as background removal, automatic captions, and photo enhancement. The company’s apps use a freemium model, in which users have free access to basic functions and must pay for advanced features. The company has been around since 2013 and has over 500 million app downloads. It generates more than $100 million in annual revenue and has been bootstrapped for several years. Given that it has been around for so long, it also has a strong data backbone to train superior machine learning models. The company will use the new funding to develop new products and make acquisitions. Its vast user base and established market will provide it with strong channels to upsell its new products to its existing customers and to gather more data to further expand its lead over competitors.
What are the patterns in AI funding?
There is a lot more to find out if you dig into the companies that are receiving funding. But here are a few things that I noticed:
1) Stick to good product principles: No matter how good your AI is, you’ll need a product that solves a real problem, that is considerably better than the alternatives, that has the least friction to adopt. It also needs to have a sizeable market, room for expansion, and a clear vision for sustainable growth.
2) B2B AI is on top: While AI-powered applications offer convenience to consumers, they have much greater value for businesses, especially as the economy moves into recession. Well-implemented AI can reduce waste, optimize recommendations, and automate manual functions, all of which can affect a company’s expenses and bottom line.
3) Look for new AI markets in unsolved problems: Established markets are very hard to conquer in AI because the incumbents already have superior datasets to train their models. New markets are easier and less expensive to enter, especially if you can move fast enough to gather data to train your ML models before your competitors.
4) Reduce the costs of acquiring data: Look for AI ideas where the data already exists and is annotated (e.g., financial transactions, sales history, patient outcomes). Alternatively, look for solutions to generate the data needed for your models to reduce the need for data gathering. If your application requires a new pipeline to gather, clean, and annotate data, you’ll need more time, talent, and funding, which is hard to come by in the current circumstances.
5) High-profile founders are luckier than others: Founders who have previous experience in large tech companies are more likely to attract funding for risky AI ideas (e.g., data infrastructure for Web3 AI). If you have one on board you’re lucky.