Winning Strategies for Applied AI Companies
Key Success Factors after reviewing over 70 companies that have raised at least $7M
As primarily SaaS investors at Point Nine, we are very interested by the applications of AI in the SaaS space. The aim of this post is to disclose a framework we have built when we look at Applied AI companies. If you’re building an applied AI company, please get in touch, we’d love to have a chat! If you are not raising funds but would also want to present at #MLBerlin, the community of AI practitioners that we help get together quarterly in Berlin, please get in touch as well.
Context
To give you a little more context — and paraphrasing Alex’s post — we have entered the third wave of AI startups. The wave of applied AI companies. The first wave was purely research-driven companies, with companies like Deepmind and Nnaissence standing out. Most of them never really commercialized their product and were acquihired before generating revenues . A second wave followed and consisted of companies building machine learning infrastructures. These startups did build some commercial traction, but most of them were also acquired before reaching scale. Wit.ai, which developed an open-source NLP API, is a telling example of this second wave and was bought by Facebook to power M, its ML-powered assistant on Facebook Messenger.
We are now at the beginning of a third wave, one of applied AI solutions. Companies in this bucket are differentiating themselves by developing end-user applications that are industry or category specific, and not focusing merely on infrastructure.
Applied AI startups are attracting increasingly larger investments and now represent the lion’s share of early stage fund raising. In the UK alone, they represent 85% of AI companies. Interestingly, they also require us, VCs, to adapt our framework of analysis, and potentially our investment criteria. As this post from MMC Ventures shows, 50% of the UK’s post-seed AI startups haven’t started monetizing but managed to raise more cash than traditional SaaS startups at the Series A (20–60% larger).
Whilst we ought to believe that we know what it takes to raise money in SaaS in 2017, we’re still fine tunning that model for ML . For more details on this, Zetta VP produced a great first framework.
Methodology
The exercise I run here consists in reviewing the 70+ applied AI companies that have raised more than $7m+ in VC funding globally, considering the amount of fundraising as a first proxy of success. Then, by inductive reasoning, I create a framework to find homogenous clusters and derive the key success factors in each of these. The goal is to create a framework based on the early success of some of these AI companies a posteriori, instead of defining it a priori.
I’ve built this database crossing CBInsight’s list of top 100 AI companies and adding a few companies that hit the news lately from Crunchbase (e.g. Cruise Automation, Grammarly) and a few companies in our deal flow. I removed from CBinsight’s list a number of “Core AI”companies (from the second wave) as well as companies in the data preparation (Paxata, Datarobot) or the data analysis/science (Rapidminer, Dataiku) space. The list does not aim at being exhaustive, but it constitutes a good enough list for a first analysis. You can find it here.
What the data says
Geos
The US alone accounts for 69% of our dataset while 8% came from the UK and 7% from Germany.
Valuable and positively skewed teams
The average number of employees at a Third Wave company is 133, while the median falls at 75. Both are small number and showcases the high level of dollar raised per employee by these companies.
Funding
In total, these companies have raised $6.4bn. The average amount raised per company lies around $90M. However, the median amount of fundraising is only $30M. Both number show the field’s relative naissance
The industries or categories gathering the largest amount of Applied AI companies are Finance, Sales & Marketing, Healthcare, Transportation and Cybersecurity. When combined, these 5 categories represent 65% of the total number of applied AI companies in the dataset and 89% of the total amount of fund raised. This amount is heavily skewed by Applied AI companies in Finance, which represent alone 52% of this amount.
Definition of the framework
Looking at the companies in this list, it seems that we can create a framework with 4 categories.
- “Full Stack”: companies that control the entire value chain. They own the relationship with customers and suppliers and don’t sell software to existing incumbents. Babylonhealth, which is building the next generation clinic using AI is a great example of a full stack AI company. Chris Dixon tells more about full-stack companies here.
2. “AI Tech Enabler” companies are selling a piece of software which includes AI.
- Going deeper into the categorisation of “AI tech-enablers”, I differentiate between vertical and horizontal solutions. Vertical solutions cater to the needs of a specific industry whereas horizontal ones have no industry focus. For vertical software, think of a software for autonomous cars that General Motors would pay for to include it in any of their cars. A CRM software like Salesforce is probably the best example of an horizontal software, that many companies in many industries can use.
- In the bucket of Tech enablers building horizontal solutions, I finally differentiate between companies that aim at replacing existing category leaders (“Category challenger”) and those creating new categories (“Emerging Category Leaders”). Before being acquired by Salesforce, RelateIQ was a challenger for Salesforce. An example of an “Emerging Category Leader” is Chorus.ai, which is part of a new category that we could call Conversation Intelligence software.
To illustrate this, compare Oracle’s on-premise CRM vs. Salesforce vs. RelateIQ. Just like how Salesforce won market share by understanding and leveraging the cloud 15 years ago, RelateIQ went after the same opportunity competing on the same value curve but providing additional value on several axes (user onboarding, ease of data input, workflow intuitiveness) thanks to machine learning. Salesforce would go on to buy the company for $390M just three years after its foundation, and now offer it as SalesforceIQ.
Another interesting differentiation is to think about the type of innovation these companies bring to their respective market (innovation vs. disruption). Full stack companies and Emerging Category Leaders are Disruptors. Full stack companies aim at displacing established market leading firms, Emerging Category Leaders create new markets with a new product — which creates new challenges for VCs that we’ll discuss later. Vertical Tech Enabler and Horizontal Category Challenger are Innovators, they provide incremental value streamlining an existing process.
A few stats about each of these categories
- Two thirds of theses companies are tech enablers and the rest are full stack companies.
- Interestingly, full stack companies have raised ca. 5x more than tech enablers. This results mirrors the fact that full stack companies need to bear customer acquisition costs and hire people across the whole value chain. The data also shows that they have more employees than in any other categories.
- The three categories of tech enablers have raised very similar amounts so far. That being said, Vertical AI companies have by far the smallest staff size . Just like in SaaS, Vertical AI companies might be more engineering driven than Horizontal companies, requiring less sales and marketing workforce.
Risk and opportunities
The reason why this framework is relevant is that each of these categories have:
- different challenges (e.g. user or data acquisition strategy, amount of fundraising needed),
- different risks (e.g. competition or dependency with incumbents), but also,
- different opportunities (e.g. moving up the value chain or displacing existing solutions, exit options).
I try to outline and explain each of them in the table below:
Here are 6 risks and opportunities that we felt were worth mentioning into more details:
1. Bridging the full stack gap to increase market size.
Well positioned Vertical AI companies can leverage their clients’ data and lessons learned to transition to full stack. These companies can provide the entire value chain, and even compete with their previous clients. As a result, their TAM could grow 10x. A good example is Infermedica, which is a ML powered tool that helps doctors make better decisions. Infermedica could become full stack and compete with its existing clinics, just like Babylonhealth does today. The implication of this is that, as investors, we need to think about two TAMs: both the potential software spendings within a certain market, and that of the total size of the market once AI has gained larger acceptance.
2. Vertical AI companies have emerged to help incumbents compete with new full stack AI companies in the same field.
Cruise Automation was bought by GM so that they could work on autonomous cars. Zestfinance is helping banks integrate ML to assess credit scores, just like Kreditech, Affirm or Avant. This reinforces the idea that there might be new opportunities to build Vertical AI companies where full stack companies gain larger market acceptance.
3. Building a first solution without AI to generate valuable transactional data and further automate workflows.
Many of the emerging winners in our database started selling to their customers before integrating a ML component. Their first offering was a clever way to gain user generated data. They could then extract more value by adding it a later point in time. This also built defensibility thanks to data network effects. InsideSales is probably the best example of this category. They first gathered significant amount of data on sales efficiency gamifying sales processes. Only then, the company started selling a sales forecast platform based on AI. Important point to note here is that customers were already ready to pay for the software although it did not integrate any AI.
4. The risk for a Category Challenger to be displaced by innovative incumbents.
We often wonder about the defensibility of a business that would use data collected by large SaaS companies. An example of that is a company, which would process Zendesk data. Zendesk has a significant advantage of sitting on years of ticketing data. If defensibility stems from owning the data. If we assume that algorithms are becoming a commodity, what is the long-term chance of building a new winner which relies on third party data? We believe that understanding the product velocity and the data strategy of these large incumbents is often key to assessing this risk.
5. Collecting and processing new data streams to build defensibility.
One of the interesting ways to counterbalance this aforementioned risk (#4) is to collect new data streams that incumbents do not own. InsideSales collects data on sales productivity on top of Salesforce’s database. Chorus processes voice data, which Salesforce does not neither.
6. The risk for a Vertical AI Tech Enabler not to get access to valuable data over time.
We may wonder if there really is an opportunity to build a long lasting Vertical tech enabler. It’s unclear whether or not incumbents that operate in the same industry as a Vertical AI startup have long term interests in sharing their data with this single company. Using the solution developed by a Vertical Tech Enabler, they make AI agents better. In the meantime, they also share the competitive advantage that lies within their data with their competitors. Tom Tunguz builds a very interesting point on that, drawing a parallel with the Adtech world and DMPs like Bluekai in this post. In other words, industry-specific or vertical AI Tech Enablers may have to become full stack if they want to be sustainable over the long run (see #1).
Conclusion
Far from the AI winters of the 80’s, the field of vertical AI seems to be blossoming, as witnessed by the increasing number of startups, capital and media coverage. We could even be at a tipping point with regards to the scale AI companies could achieve in the coming years. From an operational and from an investment strategy standpoint, they offer many interesting fields to explore. What are the new moats for these companies? What are the winning strategies? What are the early proxies of long lasting success? Hopefully, this framework can bring a little transparency in the way that seed investors currently look at this market and assist early-stage founders as they tackle the risks associated with Vertical AI companies and embrace their opportunities!
If you’re building an Applied AI company, or have any feedback on this framework, please get in touch with us, we’d love to have a chat!
Join 25,000+ people who read the weekly 🤖Machine Learnings🤖
Very special thanks to my colleague Rodrigo, this framework is the result of several discussions (or debates) we’ve had in the past months. Thanks also to Alexandre, Bartosz, Robin and Clement for giving it a read when it was still a very rough draft.
