Earlier version published at Data-Driven Investor.
AI is now mature and widespread enough to encounter criticism. Since most AI systems today aren’t quite software, the difference in their business model renders them less attractive for venture capitalists. AI systems involve ongoing human support and material variable costs, so they don’t often scale as easily and they don’t follow the strong “build once/sell many times” software model.
We, an AI professional, and an AI investor wanted to tackle these arguments. The following quotes are mostly driven from an Andreessen Horowitz article discussing the idea of how the AI business is different from the software one.
“Deep learning costs a lot in computing resources, for marginal payoffs.”
It is important to note that the AI business model differs from a typical SaaS business model in the sense that there are higher cloud/computing resource costs associated with building models and execution. The AI business model is costly to maintain, while cloud costs can reach 25% of revenues, not an insignificant number. Another aspect of the AI model that differs from a traditional SaaS business model is that the “human factor” plays a larger role in the former than the latter. As a result, moving between “domains” can prove costly. Improving the AI business model generates marginal benefits in the sense that you can spend a lot and experience relatively small improvements.While all of the above is true, there are additional facts that are equally important to consider.
For starters, the objective of AI is to develop a single model that can be easily adapted to more models or fields. We can speak from experience here, as Bigabid developed its AI infrastructure so that it can serve many clients from totally different domains. This way, moving between domains is relatively easy and, in the worst-case scenario, only requires minor adjustments from our end. We serve clients like Uber with the same AI architecture that serves games like Candy Crush.
Furthermore, there are some ML techniques that can help reduce cost and effort. One is using transfer learning, which involves storing knowledge from one problem and applying it to a different problem.
Another option is making use of some amazing open-source, pre-trained models like BERT by Google. These are a great way to reuse powerful models that have been trained by others—and they do not cost anything since they are open source.
Regarding high server costs, we have found that they can be dramatically reduced by using the services of tech companies that were designed with to alleviate this pain point. A great example of this is Granulate (which cut our CPU costs by almost 60%!).
In the near future, we expect special AI chips to be developed that will dramatically reduce the costs of training deep neural networks. In addition to the anticipated rapid development and improvement of frameworks in the deep learning domain, we also predict that training a deep neural network will be much easier and thus relatively inexpensive.
“Machine learning startups generally have no economic moat or meaningful special sauce.”
In the AI world, your economic moat is your ability to access data others don’t have access to and generate models that outperform those of your competition.
In some domains, off-the-shelf deep learning (DL) architectures can get you close to state-of-the-art performance and black box models are kings. The moat in these fields is primarily the data sets models are trained on and, to a lesser degree, the DL architecture being used. In other areas, hand-crafted features and domain knowledge, as well as the unique data sets a company has access to, will be the game-changers. Sometimes, feeding a model the right signals is all a company needs to create the state-of-the-art. This is about knowing a domain and asking the right questions, as much as it’s about having the right model or the latest DL architecture in place.
“Machine learning startups are mostly services businesses, not software businesses.”
It needs to be clarified that there are opportunities in providing services alongside products. Customers need high-quality consulting surrounding their data, which is then enriched by models’ predictions and precomputed features. These services can accelerate the adoption rate of AI technology while yielding better results.
The more the same model is used for many companies, the more the service component become smaller and as a result, a company’s business model gets closer to being like the lucrative software business (SaaS).
“Machine learning will be most productive inside large organizations that have data and process inefficiencies.”
The way we see it, organizations of all sizes can benefit from the power of machine learning. While it may seem like larger companies have more data to work with—and that is usually the case—smaller businesses can be much more focused. For example, smaller businesses can gather data in a specific domain and become the market leaders within it.
Another point to consider when solving business problems using machine learning – moving fast, and trying new things is the way to go. This is the way smaller companies or startups work inherently, as it’s more difficult to move fast and try new things at larger companies that don’t have data science built into their DNA.
We feel that companies that adopt AI techniques will gain a competitive edge over ones that continue to rely on traditional approaches. Unlike in the early days of AI, costs can now be better managed, models can be transferred between domains and deep domain knowledge can be difficult for competitors to obtain. Hence, there are few reasons for business not to bring some aspect of AI into their businesses.