In the trajectory from invention to product, cognitive computing is following a familiar pathway, but with some additional twists in the road because of the complexity of the technologies involved. IBM’s Watson was a breakthrough in computing. As an invention, it heralded a new computing era. Since 2011, when Watson burst upon the scene, we have seen successful custom deployments, but to have a real market, we need robust products that can be quickly installed and that give repeatable and predictable results to customers. That’s because these applications may contain hundreds of technologies that must be highly integrated, and because the richness and ambiguity of language add to the difficulties of any language-based application.
In 2015, we are beginning to see real ready-for-market products. Two good examples of companies that have developed real cognitive computing products are CognitiveScale and CustomerMatrix. Both of these companies have developed knowledge bases for specific domains that contain the 80% of relationships, terms and entities that are common across companies in that industry. They have developed tools for data scientists to add the remaining 20% of internal terminology to these knowledge bases in order to customize the applications. But because one of the characteristics of cognitive computing is that it learns as it is used, it’s possible to deploy these products without that additional work and immediately get good results from the system.
Help desks are an obvious pain point for companies. They are expensive. It’s hard to ensure that customers get consistent answers, and service representatives usually consult 10-20 different sources to find the right answer. A cognitive digital assistant should be able to return faster, more accurate, more consistent answers. The diagram below shows the cognitive graph that CognitiveScale developed to support their Guided Service application for trouble tickets:
This graph can be reused throughout the industry without giving away internal secrets. But companies can also quickly customize the application to incorporate their internal knowledge bases and terminology. In this example, it took only two months from the initial planning stage through pilot stage to deployment at scale. That’s because it comes pre-configured and trained for this industry. That’s quite a change from the years that it took to create the initial Watson Jeopardy! application.
Similarly, CustomerMatrix has developed a cognitive system that is optimized for sales and customer engagement management (CEM). Its knowledge base contains the entities, processes, events and relationships that are germane to CRM. These are applicable across companies and industries, but within that sales and CEM realm. It uses this knowledge base to understand a customer’s data, to learn, to discover patterns, to infer relationships and predict actions, and to make recommendations for the best action to take at a specific time within a specific task. Like the CognitiveScale Guided Service applications, it is quick to deploy, and it learns from use. Tools for customizing it to a specific organization’s processes and terminology make it accessible to data scientists who can use a widget library to insert what CustomerMatrix calls ActionAlerts™ in the user workflow. It can also infer missing information—like who may be missing from an organization chart—by putting external sources and internal data together to make sure that they validate each other. With this information data scientists can push an ActionAlert™ to the user of the CRM or Relationship Manager portal that uncovers and explains previously undiscovered customer revenue possibilities as shown below.
This ability to put together information from multiple sources and to connect the dots is nothing new: people do it all the time. But they can’t manage the scale of information that a cognitive system can. And they also may not see unexpected connections that run counter to their biases. When we add cognitive assistants to humans, we get additional angles that neither might come up with separately.
Both of these applications are delivered in the Cloud, allowing the kind of scale and flexibility that large knowledge bases need. Customers need not recruit cognitive computing specialists with Cloud delivery, and the vendor can upgrade and improve the technology and tools easily. With the hurdles of generalizing applications, we have moved from custom deployments to products. It looks like we are on our way to a cognitive computing marketplace.Share