We've been talking a lot about predictive applications recently on the Blue Yonder blog, and for good reason. Predictive applications provide a way to automate mass decision-making. Predictive applications are massively transforming industries. And predictive applications are already here, right in your pocket.
When comparing predictive applications and predictive application platforms, I can see one common pattern emerging. Despite the many differences between consumer and enterprise predictive applications, between black-box predictive APIs and predictive application platforms, there is one thing all competitors can agree upon: predictive applications live in the cloud.
While cloud computing is not an entirely new approach, with more than a decade of adoption by enterprises big and small as well as by startups and industry disruptors, it is the first category of applications that wouldn't be possible without the cloud. When web applications became popular for enterprise use cases in the early 2000s, most clearly exemplified by Salesforce.com, many of the technology pieces that we take for granted now were not in place yet, so that enterprise software using web-based interfaces could be built and deployed on-premises as well as being a centralized SaaS (Software as a Service). This is no longer the case for predictive applications, and there are a couple of driving factors in this trend:
Predictive application adoption is driven by the business, not IT: as predictive applications automate existing business processes and are deeply integrated into the problem domain, it is no surprise that business leaders, not technology leaders, are pushing for them. Unconstrained with the need to justify capital investment in existing hardware, and under competitive pressure to realize the gains from better, automated decisions, these business leaders are looking for immediate gains without the constraints that existing transactional or analytical infrastructure has been imposing on them.
- Predictive applications have fast return on investment, but require extensive hardware: when going through a collection of recent Predictive Application rollouts based on Blue Yonder's platform and products, we could see that the average time for a customer to recoup initial investment in building and integrating a Predictive Application was three months, and not a single customer had to wait more than six months to get a positive return on investment. At the same time, the hardware that is involved in operating these Predictive Applications, ranging from high-memory database nodes to high-throughput computing nodes in a fine-tuned cluster, would have taken at least six months to provision and another six months to set up and install with a typical enterprise customer. This means, even in the best case time to market would have been increased two-fold and time to achieve ROI even more than that.
- Big Data and fluctuating load patterns require elastic scalability: One of the most striking properties of big data is not how big the data is per se, but how fast it is growing. As more data is one of the best ways to improve model (and decision) accuracy, just ignoring the growth of data is not an option, and a scalable and centralized option is required to harbor the data that is used for training decision models and making predictions. The ideal environment for these growing mountains of data is the cloud with its elastic scalability. Elastic scalability also helps with coping with the second aspect of Predictive Applications – load wether measured in the number of predictions or the amount of data needed for training varies vastly in the course of a day, week or month. Where an on-premises deployment would be massively oversized most of the time and massively undersized in the most crucial moments, scalable could infrastructure makes just the right amount of resources available at any time.
- Predictive applications are fluid and constantly updating: when it comes to predicting the future, there is no done and over. Just as businesses are changing, customers are changing and competitors are changing, and so are predictive applications. The constant cycle of updates to the data, models, algorithms and underlying framework would cause on-premises deployments to either stay stuck in the past, or to be in a constant uphill battle to stay up to date. In the cloud, automated and transparent deployment techniques keep your applications and the platform behind them as relevant as the decisions they are making.
- Predictive applications benefit massively from economies of scale: any data-driven business is dealing with two dominating trend curves: on the one hand, the exponential growth of data and computing power, driven by Moore's law. One the other hand, there is Metcalfe's law, promising quadratic returns due to network effects on a single platform. Predictive applications allow companies to benefit from both. By keeping historical data in the cloud, prediction quality and decision accuracy can be improved over time, without even changing the prediction model itself. Additionally, the network effects of a shared cloud platform allow all predictive applications to benefit from the model, algorithm and platform improvements immediately.
With these drivers in place, building predictive applications on-premises with slow hardware provisioning, long time to market, low flexibility and high cost quickly becomes an unattractive provision, for customers and vendors alike.