Building AI in Europe
Insights from events: Deep-Learning startups in Paris
This is a guest post by Oliver Behr, Marketing Lead at Navee. He shares insights from a conference about building Deep Learning startups, organized by Entrepreneur First at STATION F.
Deep Learning
Artificial Intelligence (AI) seems to be everywhere in tech right now, powering everything from your car to medical diagnosis. Yet, the phrase “artificial intelligence”, is often mistakenly intoned by technologists, academicians, journalists and venture capitalists alike, and it became a prominent buzzword within the startup environment. Artificial General Intelligence (machines that compare to or surpass the human mind) belong in the distant future, while today, AI gradually approaches human levels when performing “simple” tasks, like language processing or computer vision.
Deep Learning, a technology to build pattern recognition systems, however, drives more and more AI-projects. Therefore, during a conference at STATION F, we explored what Deep Learning is, and what it takes to build a DL-startup in Europe. After describing the concepts of Artificial Intelligence, Machine Learning, and Deep Learning, this report will dive into the learnings from the conference where the experts provided valuable insights from their journeys.
Deep-Learning companies
Venture capital funding of AI and Deep Learning startups have hit a record of $9.3 billion last year. Today, AI applications receive large excitement around the industry and society; and not without reason. The powerful and flexible Deep Learning technology enables AI-projects to be more broadly applied than ever, creating opportunities for early adopters even in businesses and activities to which it previously seemed unsuited.
Tech-View: What are deep-learning companies?
Let’s get clear on the terms!
- Artificial Intelligence (AI) is the achievement of computer systems that manage to complete tasks that usually require human intelligence, like language processing or computer vision.
- Machine Learning (ML) is a science and approach to developing algorithms that automatically improve with experience. These algorithms “learn” from a dataset (=experience) to detect patterns and nuances.
- Deep Learning (DL) is a way of building algorithms that extract useful patterns from data. This approach is inspired by the multilayered structure of the neural network in our brains. Artificial Neural Networks become “deep” when complexity in their algorithmic structure increases (more layers of analysis). These Deep Networks learn from training data to solve all kind of different challenges — Deep Learning. They are characterized by their ability to deduct abstract patterns by self-enhancing their algorithmic structure.
AI, ML and DL describe different hierarchal levels of building complex computer systems. AI summarises overall achievements of computer systems, ML is a way to build a system that can learn and DL is a toolset to build algorithms that detect patterns and nuances. Companies and startups throughout the world, deploy the tools of Deep Learning to build AI systems that automate business processes, gain insight through data analysis, or engage with customers and employees.
The expert panel
To explore what it takes to launch a successful deep-learning startup, Entrepreneur First’s Allister Furey & Gwendolyn Regina, invited Wilder Lopes, CTO of Upstride and Matteo Amerio, CEO of Navee.
On stage at STATION F, they discuss the challenges and opportunities of the European startup ecosystem and share learnings from their approach to finding product-market fit.
The European startup environment
Wilder and Matteo, from the Paris-based startups Upstride & Navee, have both considered establishing in the United States. However, strategic reasons around talent access and market opportunities led to a decision in favor of the EU. The panel agreed that the American market offers easy access to capital and resources, but that access to talent can be hard for early-stage companies. Silicon Valley cumulates an impressive number of prime engineers. But, at the same time, the high quantity of capital-backed startups represents tough competitors in the war of talent. In times of Trump and trade-wars, however, Europe’s talent-hungry, policy-backed digital hubs have gained momentum. They’re now offering a desirable alternative for ambitious workers, generating talent-access for startup projects in European states.
Europe might have missed out on building breakthrough internet companies using consumer-data, social-media companies, or huge mobile-application companies. However, the founders on stage see substantial potential for Europe to deliver on AI and catch up against the most AI-ready countries, such as the United States and emerging leaders like China. Allister from EntrepreneurFirst elaborated, that Europe’s startups shouldn’t try to compete head to head in core industries of US-Internet giants, but rather on new frontiers where Europe has an edge. Business-to-Business [B2B], automotive, advanced robotics or industrial engineering and manufacturing are historically strong pillars in Europe. In this context, there emerge various new frontiers, where startups can find customers in the existing European industry and take a leading role by generating value from applying Deep Learning technology.
Build a deep learning startup
Navee and Upstride are startups that embody the European startup momentum, even though on differing frontiers. Upstride supports AI development teams directly, by focusing on the advancement of the Deep Learning technology itself, while reducing the amount of required training data. They started the company based on insights from the founder’s prior research and technology. After reaching out to many development teams in various spaces, they were able to verify the demand for training data reduction.
Unlike Upstride, that has technical roots, our panelists pointed out, that a majority of startups begin by identifying real-world problems and then choose a technology that helps to solve it. For instance, ineffective business processes have led to the application of Computer Vision, build on DL tools, to make the processes more efficient. Problem first, technology follows.
Matteo, the CEO of Navee, observed repeated apartment scam listings on real estate platforms when he searched a flat in San Francisco. He started to explore this kind of scam by talking to multiple industry representatives. While learning about user-generated content platforms, he identified the bottleneck of every fraudster — the images he needs to create a fake profile. Using advanced computer vision, his team developed tools that enable content moderation teams on digital platforms to track down unwanted content more effectively. To do so, Navee derives insights from images, such as the image footprints across the web and image traits (quality or nudity). Since then, Navee continued to build on its computer vision strength and has grown into related verticals like real estate or online dating.
The panel agreed, that the problem-first-approach to finding the product-market fit is essential for AI startups. To understand which technology might help solving a problem, here’s what our panels suggests: (1) Look for real-world problems, (2) deep-dive into a problem, (3) by getting into conversations with stakeholders, and (4) make sure they have diverse backgrounds (industries, levels within organization, demographics). This will provide guidance in the choice of technology that suits best to solve the problem. Technology-driven people are often excited and motivated by the science of machine learning and the challenge of coming up with creative new algorithms and methods, but solving the right problem is key. Deep Learning is a powerful tool to solve a problem, yet shouldn’t come first. The value of Deep Learning and AI is measured in the context of the problem that they solve and the products that they empower, not its technical sophistication.
Grasping a problem and translating it into solution hypotheses (how to solve the problem) is a difficult task. It helps when there is regular exchange with that potential client, which is most likely to pay money for a Minimum Viable Product (MVP = the smallest thing you can build that delivers customer value and as a bonus captures some of that value back). Potential clients help confirm and shape the initial hypotheses and ensure that the startup’s product development direction leads to monetization. Allister from EntrepreneurFirst summed it up with “Customer First, Talk to them”! Matteo from Navee also mentioned that startups need to be careful not to become too responsive to the client’s demands. It’s hard for a startup to say “no”, but it is also hard to solve every problem. With too many promises, the startup might lose focus and fail to deliver on the customer’s expectation. In many cases, a clear no is better than a weak yes.
Getting in contact with potential clients in the first place is easier said than done. Mapping hierarchies of corporations can help to find the stakeholders and most promising conversation partners. When approaching the potential client, the communication should be distinct and should always answer “What’s in it for me” (from the potential client’s perspective). The panel agreed that if a startup focuses on a (niche) aspect of a significant problem while adjusting its product development according to the potential client’s needs, the chances to find product-market fit increase significantly. When potential clients do not want to listen, there might be a problem in the value proposition and market fit. Matteo Amerio summarises:
“It’s not about failing companies, but failing products.”
Summary
- Europe’s potential around new applications of AI and DL technology is on new frontiers, that emerge around industries that lagged in the digital transformation, like industrial manufacturing, engineering & robotics.
- Finding a problem and shaping solution hypotheses should come first. Technology follows accordingly.
- Shaping a solution should be verified in constant exchange with potential clients (those, that are most likely to pay for an MVP). When it’s hard to get into a conversation with potential clients, the value proposition (hypotheses) might require adjustments.
See you at the next discussion!
For more events at STATION F, take a look at stationf.co/events!
