Challenges and Opportunities of Investing in Artificial Intelligence Startups
- It is difficult to understand and evaluate Artificial Intelligence investments
- There is a lot of hype and a lack of substance with many teams
- Funding for AI startups is tougher and the seed stage takes longer
- The market potential for applied AI is huge
- Teams often have a high level of education and research background but lack entrepreneurial experience
- Once AI works, it is hard to replace
As a business angel with Point Nine Capital, Team Europe and since 2014 with Asgard, I have been involved in roughly 50 startups. The earlier ones were in eCommerce, lead generation and marketplaces. I then dealt with Software as a Service (SaaS) and later with IoT (Internet of Things). Finally I have turned my attention to Artificial Intelligence.
I would like to share my experiences as an investor focussing on Artificial Intelligence compared to other early-stage investments.
Dealflow — Most is Not Artificial Intelligence
Dealflow is critical for successful Venture Capital firms (VCs). You receive your deals through your network, directly from founders, through accelerators, company builders, consultancies and other investors. You also find your deals reading blogs, news, screening databases or from recommendations.
However, 80 % of startups claiming to use AI in fact do not use it. Most of these companies use machine learning, but that does not make them an AI company.
For me Artificial Intelligence starts when a system is based on analytics and prediction, and is able to make its own decisions. For that you need to have a world model to reduce the complexity of our human environment. Furthermore, an AI needs to have a motivation and goal for constant self-improvement.
Most common technologies such as Natural Language Processing (NLP), Expert Systems or Deep Learning (CNNs, RNNs, GANs) therefore provide the building blocks for AI. Only the right combination, however, along with a learning model, gives you a form of intelligent machine — if all goes to plan.
This is the difficulty with dealflow for investors. While most founders with a business degree enjoy plastering their pitch decks with buzzwords, the research-oriented founders often lack the experience and network to approach VCs.
As a consequence, many VCs receive plenty of AI dealflow, but most teams are not able to fulfill their own vision. It’s easier to write a pitch deck than it is to build a working product.
Teams — Look Out for the Entrepreneurial Scientist
Every technology needs different teams. For eCommerce, lead generation and marketplace you often need a few business guys and former consultants. Execution is key and the product itself is less important.
For an IoT startup, you should have a hardware engineer and an industrial designer on your core team.
To start a SaaS company, there should be a least one front end and one back end developer.
Artificial Intelligence companies need more science-focused founders. People who enjoy complex models and lots of math. Most teams I have backed studied mathematics, physics, robotics, cognitive or computer science with a focus on machine learning.
I also think that AI founders already loved AI even before there was hype surrounding it. They enjoy solving problems and are less driven by money and business.
For me these are important skills, since it takes patience and endurance to build an AI company. Currently markets are still undeveloped and customer are uneducated.
Success Factors — Access to Data and Product Focus
I have already mentioned that for eCommerce you need execution skills. That means you want to grow fast and secure a market share. Once you can scale the business, your success depends on making fewer mistakes and securing more funding than the number 2 in the market.
For fintech and healthtech you need to build trust, seek cooperation with established players and solve the complexity of heavy regulations.
SaaS startups need to own their users with amazing usability, since they often replace excel or paper & pen.
It is more difficult to predict what the success factors for AI companies will be. Today I would say it is important to have a solid understanding of current research and be able to build a product with that knowledge.
Furthermore, it is critical to involve potential customers early enough in order to understand their problems and to get their data.
I have seen too many smart people with amazing ideas who could never build a real product out of it.
There are also very few role models for how to build an AI company. It is harder to transfer research results into software than it is to build a beautiful front end.
It can happen that you work for weeks on a learning model before realizing that everything it learned was incorrect (from a human perspective), and having to start all over again.
Furthermore, access to data is key. And once you have the data, it has to be cleaned, structured and labeled. Without data you cannot train your models. And if your data is not of sufficient quality, you need a lot of capital and time to fix it with human labor.
Key Performance Indicators — It Is Hard to Measure Artificial Intelligence
There are KPIs for eCommerce like customer acquisition costs, customer lifetime value, average basket size or conversation rate.
For SaaS companies you can check their MRR, churn und cohorts.
For AI companies, however, there is a lack of clear metrics. At the moment you cannot really measure how good or bad an AI company is.
Markets — Everyone Needs AI Sooner or Later
Software is eating the world and Artificial Intelligence is taking over software.
Even today there is plenty of AI hidden everywhere. But we only have 0.1% market penetration. There is so much potential for AI in our stupid world. Just think about accounting, traffic flow, logistics, robotics, fulfillment, autonomous moving systems, asset management, healthcare, advertising, sales and security. Pick a market and enrich it with AI. The world will thank you.
Nevertheless, I see over and over again that many teams do not know where to start. They go after every possible customer from large enterprises, governments and developers to end users. That is defocus and makes it harder to achieve a product market fit.
Business Models — There Are Not Many Options
You want to make money with your AI company? Then you have limited choices.
Most AI teams start with pilot customers. That makes sense since you get paid for it (at least you should – 30k to 150k USD is feasible). You have access to data and you can continue to build your product.
However, projects often solve very specific problems which are difficult to transfer to other companies and use cases.
It is therefore important to have a balance between doing projects for early customers and investing into a scalable product. Too many AI teams end up as engineering consultants. They make good revenue with clients, but do not really sell a product at the end. VCs do not like that. We want to see scalable products with limited implementation costs.
Sales & Marketing — It Takes Time
Currently most paying AI customers are enterprises. They often have the budget for trials and the strategic vision to play with AI. Projects can also be financed by governments.
All those customer have a long sales cycle in common. Allow between 9 and 18 months to have all necessary signatures.
It is also difficult to understand which enterprises are ready for AI. Many large corporations are not yet fully digitalized. How should they be able to use AI then?
If you are struggling to find the right companies and contact person, I recommend visiting an AI conference like Rise of AI.
Product Development — Patience and Timing
Developing Artificial Intelligence takes almost forever. It is normal to wait two years until you see your first revenue. The gap between theoretical idea and real product is huge.
You can use existing open source software, however most has to be developed internally. You therefore need specific experts on your team, which are expensive.
At the same time you have to speak with customers to understand their problems. These insights have to flow into your product to train your models.
Training AI models is particularly time and capital intensive. In my experience, it takes more than half a million Euros before you have the first working prototype.
It is a challenge to find the right balance between research and application — if you build your product too fast, it may not be technically strong enough to compete with more focused teams who have done more research.
On the other hand, if you wait too long to build a product, you can miss the right moment to enter the market.
I have seen teams build an AI over the weekend. Well, not a proper AI exactly. But I have also met teams who have been working on their AI for the past 20 years. They seem to have missed the right time to enter their market.
Competition — Technically Opaque
It happens that there is no competitor to your solution and you still have a huge market. And then the next day you meet teams competing with dozens of players for the same problem.
How is a customer or investor supposed to judge which AI solution is the right one for them? Most cannot understand the models or read the code. At the same time there are no indicators to judge the quality of the trained data (e.g. for biases).
It is very difficult to see differences between AI companies competing in the same market.
Funding — Investors Hesitate
Since assessing AI technologies is so challenging, many investors are hesitant to invest. Sure, the interest in AI is strong and everyone is looking at it. However, most VCs are risk averse and do not invest in something they don’t understand.
I have noticed that VCs with business degrees in particular struggle with AI dealflow. On the other hand, colleagues with a science degree seem to enjoy AI companies.
Furthermore, I have also noticed that seed rounds are more often extended and grow in size. The required product market fit for a Series-A is reached way later by AI companies compared to SaaS and eCommerce companies.
You could say that many AI companies receive plenty of attention but not enough funding. While everyone understands eCommerce, understanding cognitive systems is just more complex.
The positive effect is that startups can finally apply for grants and scholarships. There is plenty of (early-stage) funding for deep-tech in Europe.
Exits — Only for Teams
The M&A market for AI companies is too young. Most companies which have been purchased in the past three years did not have revenue or often no products. You may call them acquisition hires. There is a saying that for a great AI developer you have to pay 1 million USD and 10 million USD for the founders in the USA. Something we do not see in Europe (yet).
Challenges from VC Perspective
I have already mentioned it. Some AI teams continue to do projects for customers and do not achieve a scalable product. That is a risk most early-stage investors do not really like.
There are also not many experienced AI founders out there. AI was pretty unpopular until a while ago. As such, few people have started companies and survived until today.
It is difficult to secure follow-on funding, and seed phases take longer.
Product development takes years instead of months. You need patience as an investor.
Furthermore, since it is hard to evaluate the technology of AI companies, VCs need more insights than usual to make the right judgment.
Advantages of Investing in Artificial Intelligence
Artificial Intelligence is fascinating. You can observe systems which learn and improve. I mean, the more mature and well-trained a system is, the better the results. The learning curve can grow exponentially, which may lead to runaway advancement.
Just do the thinking for yourself. What do you trust more? An 18-year-old human driver with 1 month of driving experience, or a 1-month old car which has learned from billions of driven miles by its networked agent cars?
Furthermore, once AI is integrated into the value chain of a customer, it is hard to replace it. You can fire humans and force them to hand over their work to successors, but how are you going to remove the cumulative knowledge of an AI, which you have rented from a startup, and transfer it over?
I also really enjoy working with AI teams. Many of them, like myself, love SciFi (12 books to read to prepare for the singularity), and it is a great pleasure to talk with them about long-term AI development. There are many very pleasant people who are open to networking and interested in the business insights an investor can give.
And the last argument is that Artificial Intelligence will make this world a better place to live in. If you invest in AI, you are therefore also investing in yourself and future generations too.
Further links and reading: