In our fast-changing economy, the biggest business opportunity is expected to be artificial intelligence (AI). According to a recent report by PwC, the revenue forecast due to AI by 2030 is $15.7 trillion, which equates to a growth of 14% of the global GPD. Out of the whole amount, $6.6 trillion would come from productivity gains and $9.1 trillion would come from a demand increase.
The productivity gain would come from process automation procedures such as robotics, autonomous vehicles or automated intelligent services. It will also come from capabilities augmentation of the current workforce, such as decision-making support or business scenario simulations. Demand increase would come from higher quality products that will be more intelligent and better adapted to the specific customer needs.
The AI revolution has been possible to a high extent due to the success of deep learning. A key reason that makes deep learning attractive is that, generally, the more data you feed to the network, the better it performs. As you can imagine, it is not as simple as that, but it is true that this behavior does not appear in other machine learning algorithms. Thus, the performance of an algorithm then could be seen as a function of the amount of data. In a market where 90% of the data has been created in the last two years and where we will create 40x more by 2020, we can expect a performance improvement for this technology.
Now, what are the customer needs that AI could cover? It depends on the sector, so let's set some examples and discuss how I would address the problem (this does not mean that it is the best way to do it, but I just want to give an example of how the problem can be solved).
Health care is one of the hottest sectors, where IBM and Google have made important investments. An interesting business case is data-driven diagnosis support that could lead to the early identification of diseases. One specific example could be support for lung cancer detection. Given scans of the lung like in this Kaggle competition, a system using transfer learning with Convolutional Neural Networks (CNN) could help a doctor to identify cancer. A different use case could be continuous AI-powered monitoring of a unique patient, trying to identify an anomalous state to prevent diseases. A solution to this problem could be using an autoencoder to learn the normal patient distribution status and then perform error reconstruction for detecting the anomaly.
The automotive and transportation industry is another area where AI can have a big impact. Apart from computer vision systems for autonomous driving that is already being deployed, an interesting use case can be demand forecasting for ride-sharing fleets like Car2go or even taxi-like business like Uber. A way to solve this problem is to use an LSTM given temporal information of users. A different use case is to help insurance companies automatically manage and tag accidents. In a crash, the drivers can take pictures of the car state that could be automatically classified using transfer learning with CNNs.
Financial services is a sector where AI could be transformative. Traditionally, this sector is not very innovative, but banks, investment firms, and insurance companies are starting to realize the huge business opportunity that AI poses and are investing heavily in Fintech. An interesting use case is automatic asset wealth management and prediction. This could be solved using Reinforcement Learning (RL) to make a network learn an optimal action sequence (or optimal policy, as it is called in the RL jargon). The other business case that is attracting a lot of attention lately is bots agents for customer management. We can use an LSTM together with a CNN for speech recognition and then a different CNN for recommendation.
Other sectors are retail where a CNN generate features that are indexed and then retrieved for product recommendation; manufacturing where an LSTM can be used for predictive maintenance or energy where an LSTM could be used for demand forecasting.
A big question mark for sales professionals is how AI can be monetized. Companies that sell infrastructure provide ready-to-use virtual machines with most machine learning and deep learning packages preinstalled. In contrast to Hadoop or Spark clusters that are traditionally up all time, virtual machines can be switch on and off, so the consumption is cyclical. A workflow can be like this: a user set up a machine, trains a model with one or several GPUs and then switch them off. An alternative to generating more consumption, and therefore revenue, can be set up a managed GPU cluster, for training, retraining and scoring (in this situation, Kubernetes is a great asset). In some cases, a customer doesn't only need training but also retaining. Retraining is key for avoiding a decrease in model performance.
An extended business model in both corporations and startups is software as a service (SaaS), which in many cases is provided as an API. If the target is B2C, the billing can be per query. For B2B, it is common to sell packages of maximum queries and other premium services like customer support. Alternative, SaaS can be sold as licenses, that way a company can give different levels of services to customers of different sizes.
Finally, the traditional model of consultancy that VCs don't like (because it is not scalable), can be really interesting for B2B. Something that I learned when I was building a startup is that a successful business doesn't get distracted trying to master many areas at the same time, they focus. That's why many times they look externally for specialists in other fields.
The impact of AI is global and will touch every industry. Those companies or products that are not able to adapt to the new economy, might be replaced by more efficient, optimized and intelligent ones. Be prepared, AI is coming!