Why artificial intelligence is not easy to live in Russia
Artificial intelligence technologies are becoming a venture mainstream — an integral element of startups in the field of information technology. Nevertheless, in Russia, AI startups face a number of problems
Five years ago, the domains in the region .ai was free, but now they are impossible to buy. If before a serious application looked like a large dashboard, now the ideal is a product with one button that makes a “miracle”. And under the miracle of hiding artificial intelligence, carrying out all complex processes. Even the input of commands can now be performed by voice, gestures — in a sense, through the analysis of human behavior.
AI, machine learning and life
What do AI projects do and how does it differ from common automation systems? Machine learning systems have long optimized simple and predictable actions. Take, for example, filling out a questionnaire for the purchase of a ticket and the subsequent processing of this application on the company’s servers. Such basic processes (in dry scientific language — processes with low entropy) have long been automated algorithmically. This can not be said about the problems with non-standard input or output data — “data with high entropy”. The systems of the new generation are designed to cope with them. For example, you need to process the request: “I’m tired, I want to go to sea.” Based on the identity of the user, the intelligent assistant will pick up tickets, a place of rest with an optimal climate, organize a transfer from the airport and will be able to pay everything at once.
The technology itself usually solves the following types of problems:
Recognition (voice, face, defects – the so-called Anomaly Detection).
Search (Matching). A typical example is a search engine (find 1 out of 100,000) or more exotic recommendation systems.
Prediction. Who will win the election, what will be the weather, what will be the revenue from the outlet?
Startups in AI
Successful AI projects, unsurprisingly, are often a combination of science, business and marketing. A typical company success story is often based on the fact that their solution is much better than others automates the process or processes data that no one has ever cared about before. For example, sensor data at the plant allowed us to predict hardware failures, or analysis of Twitter posts helped to identify valuable trading signals to trade stocks.
At the same time, projects in the field of AI have certain specifics that derive them from the standards of the venture industry. In our experience, the AI project is most often b2b with a long sales cycle. These projects, especially in Russia, have specific problems — the lack of financial reserves to operate during the transaction and the lack of links when entering the international market. This further complicates the creation of such a project.
Focus on b2b
There are a huge number of companies whose employees use their working time to work on typical tasks. The money for the AI project here lies in the field of optimization and cost reduction. At the same time, the economic effect of robotization of 20 workplaces differs from robotization of 2000. Therefore, large companies can afford the introduction of AI systems with super human abilities (“superhuman abilities”). Finally, the “oil” of this industry — the “big data” — accumulates in large companies. That is why the B2B direction becomes the main one for the majority of projects offering solutions in the field of AI.
Sales to large companies is not an easy task for an AI startup. The first problem is the sales cycle. It is difficult to sell to a large company: it is a process of 6-18 months long. And startups without sales, especially in Russia, do not live so long.
Another problem is scaling. In enterprise sales, you can’t just buy more traffic like in b2c, marketers will forgive us. Scaling sales requires separate efforts and time. In this area, there are strict and time-consuming system of selection of suppliers. It is worth noting that in recent years something is changing for the better due to the emerging departments of innovation. In B2B technology solutions there are specific ways of scaling, when the product becomes the industry standard not at the expense of money in marketing, but due to the best solution of a separate problem. In this case, it is automatically sold as a component of larger systems.
Problems of the Russian market
Projects in Russia face additional challenges. In Russia, innovation budgets on average lag far behind the US and Europe. Because of this, often all the interaction of corporations with startups ends after the hackathon — there is simply not enough money for implementation. There are cases when corporations see that a startup solution works, but buy its full-scale implementation from an integrator or abroad. It also happens that the idea of the solution is stolen and implemented on their own.
Of course, you need to strive to be the best to make it easier to buy you than copy. But in a small market the latter does not guarantee.
There is also a seasonal specificity associated with the same budget year in all organizations. At the end of each reporting period, heads of innovation departments massively strive to do something as quickly as possible, and in the spring and summer are prudent and unhurried.
The world market can be a solution, but there are certain problems. In our experience, the budget in this area starts at $2 million, and the output takes about two years. At the same time, you need to sell yourself: a quick hire of a good seller in this area is a rare event.
The solution to the problems would have to be venture capital funds. By analogy with the Internet, AI is now a mandatory technology — a large interesting market in which there are many tasks that require solutions. But for foundations, a startup that is going to sell to large companies is still very inconvenient.
The Fund earns by selling the purchased share to the following investors or during the IPO. The share can become more expensive only with the growth of the company and increasing its attractiveness, which implies scaling and entering new markets. If you roughly describe the approach of the venture capital industry over the past 15 years, it looks like this: the rapid creation of a prototype that consumers like, and the subsequent conclusion in plus those channels where you can quickly increase the number of customers. Based on this view, venture investors evaluate projects and allocate funding to them. This approach is good for b2c. However, AI solutions have more application points in the b2b segment, while the business will not show rapid growth in the basic metrics (number of leads, sales, etc.). It is more difficult for funds to track the growth of the project, which, in turn, complicates the process of attracting money. Analysts (and by the way, the founders of the project) do not understand what to look at and on what basis to draw conclusions.
In the field of b2b, project scaling is primarily a qualitative technological superiority over competitors and integration into business solutions of vendors. Quality technological excellence implies development costs. In this area, every next 5% of the solution accuracy is given by orders of magnitude more difficult and expensive than the previous ones. Underfinancing at this point is disastrous: the project will burn money for wages, but will not be able to develop the product enough. If we calculate the result, we are talking about the amounts of about $1-3 million at the sowing stage. To invest such amounts, even without full-fledged sales, in Russia there is no one in practice.
It can be assumed that for projects in the field of AI, a hybrid service and product model of development becomes optimal. The solution is sold and defined in the form of paid custom developments, during which the competence of the team grows and accuracy increases. At a time when the team already has completed orders and developed technology, it becomes much easier to get a round of financing for development.
Another possibility is cooperation with government research centers, which ultimately allows to reduce costs while improving the accuracy of the solution. Thus, the co-founder of Coursera Andrew Ng Is already working on this model. AI Fund, created on the early prototype of the American startup platform Betaworks, which raised more than $175. Also, such approaches are now being developed by the Accelerator physics and technology in cooperation with commercial companies.