A month ago now I have been at the AI Europe conference in London. Good event, well-organized and informative. The 2-days event covered AI theory and applications through a series of talks, specific tracks, and demos.
People coming to these type of events are usually quite optimistic about AI and the innovation AI is bringing. In fact, the general feeling of the conference was well-summarized by Robert Golladay from Cognitive Scale:
AI is moving from labs and pure science to practical applications and it represents an exponential business opportunity which will create new partnerships for mankind.
It also didn’t miss the standard conversation around deep learning techniques: people overall agreed that deep learning helps companies reaching quite easily reasonably good results and requires few feature engineering. Although, it comes with challenges, such as the interpretability of the results; the difficulty in debugging, fixing, and tuning the model; the high dimensionality of the problems and the long training time; and finally, the hardware costs.
It was a balanced event in which explanation of machine learning workflow within big companies was alternated by small degrees of speculation on when the inflection point for AI is going to occur.
I learned a few things there, which I would like to summarize with these three takeaways:
I. The AI French ecosystem is pretty robust.
There were several French companies and researchers, and it seemed obvious that France has a good understanding of AI overall. Actually, a deeper analysis confirms that: France is one of top ten countries investing and developing AI solutions, right after the ‘usual suspects’, i.e., US, China, UK, Canada, and India.
French are great probabilistic theorists and are well-known for their strong quant finance skills as well as for the fast-paced robotics development they encourage and support.
I guess that the union between those fields emerged recently under the common name of AI, but we should not be surprised if we consider that some of the best AI researchers worldwide (e.g., Yann LeCun and Antoine Blondeau between many) are French and big players are investing in the country (Facebook opened in 2016 an AI research center in Paris and an incubator for data startups).
Furthermore, Paris is clearly the epicenter of the AI revolution, and it seats in the top ten cities in the AI ecosystem.
This landscape is getting bigger fast and the reasons are manifold: pool of talents; public institutions fostering scientific research; great recognition worldwide from investors and clients.
This brings me to the final point: French (or French-leaded startups) are nowadays working prevalently on NLP, personal assistants, and trading applications. Few well-known examples are Wit.ai (acquired by Facebook and working on NLP); Snips (which aims to empower apps or IoT devices through AI brains); Walnut Algorithms (a trading startup that came out from Startupbootcamp Fintech).
II. The Investing Landscape is Changing
David Kelnar (MMC Ventures) presented some of his insights as a venture capitalist on what it means to invest in AI companies in the UK (read also his incredibly smart pieces here and here). I believe some of his conclusions can be generalized, and I would like to start from those and add my considerations on the AI market:
- AI startups are focusing more on business applications rather than technology per se. Developing a new technology is a long shot and it often does not pay off in the short term. It is mostly (although not always) carried on by ‘Academic Spin-offs’ (see this classification for more details).
- Many AI applications are not mature yet. I said this because many AI startups prefer to target other businesses (B2B) rather than final customers (B2C). David attributes this to the data obstacle small companies face which is undoubtedly correct, but there are also two other trends that are taking shape: startups are entering specific applications with i) either the goal of being acquired (hopefully within 3–5 years), or ii) to disrupt sectors where big tech companies are not present (e.g., finance, healthcare, insurance, etc.). Direct competition is no more in any founder’s long-term vision.
- AI is (really?) booming. There is no doubt AI is the next big thing and both the entrepreneurial activity and investments flows are increasing everyday.
If this is true though, it is also a sad reality that many entrepreneurs and companies are taking unfairly advantage of this revolution: it is enough to “sprinkle some A.I. into your pitch deck” to simply raise your amount and likelihood of funding.
The ‘unfair’ rebranding of companies as AI-driven is one of major threat I personally see for the technological development of new solutions for all of us.
David as well found that in the British AI companies get ‘typically 20% to 60% larger than average capital infusions’, which I don’t believe to be always justified.
- The journey to monetization is short-but-long. The reality is that there is not a simple unique monetization strategy for AI startups. Across any AI company and vertical, there is an extremely high cash burn rate and talent acquisition costs, so even the breakeven can be hard to achieve. I rather prefer to think about it as a twofold choice, depending on whether your company is developing a new AI technology or addressing a specific problem/function. If you are developing a specific AI solution, your monetization journey can be longer than you expect because, as David correctly stated, it is hard for AI companies to develop a robust Minimum Viable Product (MVP), the sales cycles are long in many industries for B2B businesses, and the deployment period for SaaS companies can be extremely long.
On the other side, though, if you are a research-driven company (or a startup developing a new technology) you probably have to burden more investments at first (talents, infrastructures, data) but your likelihood to capitalize even without generating revenues is higher. DeepMind is the classic example, but more recently Magic Pony and GeometricIntelligence may confirm this view.
III. It is not clear where AI is going (yet)
The market is huge and still too complicated to be addressed by a single player or business model. There are scattered clusters of knowledge that have been built in either geographical locations or within industry verticals, but this is not enough to understand what the landscape will look like in a decade.
AI is clearly a transversal technology which provide a better, faster, and cheaper service, but it is extremely hard to center your business around this concept and indeed 90% of companies do not reach their market potential (either they failed or get acquired).
The path to succeed with practical AI is arduous but the following pieces of advice may help you to do that:
- Drive business model and technical innovation together. I believe in a compromise between business needs and tech capabilities to drive innovation;
- Fail-fast-and-first. The trial-and-error feedback loop is the only way to train a neural network, and it is the only way as well to turn your organization into an AI-driven company too. Start with small data, supervised learning models, and use the cloud. Go for the low-hanging fruit.
- ‘Pay for Success’ VS ‘Pray for Success’. This sentence is an outcome of the conference, and it suggests to build incrementally and steadily rather than waiting for the big deployment all-in-once. Find the right project, experiment and deploy fast. Cognitive Scale uses the 10–10–10 rule, i.e., identify a use case in 10 hours; build reference app in 10 hours; and finally, go live in 10 weeks.
- It is an ecosystem matter. No matter how hard you try, none of us is so smart to crack human brain, consciousness, and all those matters by himself. Get yourself wisely as many partners as you can, and become part of the AI ecosystem.
Waiting for the next AI event…many more conferences coming soon, so stay tuned!