Thanks to Matt Sanchez, CTO and co-founder of CognitiveScale, for sharing his thoughts on the current and future state of artificial intelligence (AI) as well as providing some great use cases. Prior to founding CognitiveScale, Matt was the leader of Watson Labs for IBM, and as such he is well versed on cognitive computing and the superconvergence of cloud computing, big data, and artificial intelligence, and how these technologies are disrupting every business process and industry.
Q: What are the keys to a successful AI strategy?
A: It’s a more complex lifecycle than clients may think. We begin by mapping how the AI lifecycle looks and how it fits within the SDLC. We discuss what kind of problems you can solve and how to understand the complexity of the problems we are solving. We use AI to solve two kinds of problems. The first we call “engage,” where we are using personalized insights with evidence contextualized to anticipate the needs of the business process or the end user. The second is “amplify,” where we are using AI to identify exceptions and detect anomalies.
Healthcare billing, claims, and procurement tend to be manual processes. We apply AI to automate and reduce costs in the healthcare and financial sector. We are able to identify batches of high-value claims that need to be paid first. We can ensure customer complaint emails are being handled in a compliant manner. We bring an industry perspective to dozens of use cases for financial services, healthcare, and retail.
Q: What are the inflection points you’ve observed with regards to AI?
A: The introduction of the Watson Technology on Jeopardy was one inflection point showing the power of cognitive computing. Since then, we’ve seen the convergence of data, natural language processing (NLP), cloud, and IoT with new ideas of how AI can be applied. Today, we have clients inviting us to C-level meetings because their boards have mandated them to develop their AI strategy. It’s no longer just a technology conversation. Now, we’re talking about how to get unique business value out of AI. This requires another layer of intellectual property to realize.
Q: What are some real-world problems you’ve helped clients solve with AI?
A: In banking, we’ve enabled a trading platform to engage more customers via a mobile trading app. We combined the trading history of 300,000 clients with the behavior of 30 years of market data to identify market triggers. When a new market trigger occurs, we’re able to reach out to each client with a personalized stream of market insights every day, providing personalized insight at scale. In addition, there’s a built-in feedback loop for business performance.
For the digital channel of a large retailer, we started a small pilot to learn the profiles and preferences of anonymous shoppers, resulting in a double-digit increase in conversions. By showing the client how the machine learning algorithm was learning with additional data and seeing the impact on conversions, this solution was rolled out across the entire digital commerce platform so the retailer is now making personalized, real-time recommendations based on each click.
For a world-leading healthcare organization, we’ve developed an AI solution that identifies exceptions at >90% accuracy and provides relevant contributing factors to the accounts payable processor. The solution reduced the overall cost of processing a case by ~20-30% compared to current baseline efforts.
Q: What are the most common issues you see preventing companies from realizing the benefits of AI?
A: One of the most common misperceptions about AI is believing that AI equals machine learning or data science. In fact, AI is a superset of data science, requiring orchestration, composition, governance via SDLS, and assurance that the algorithms are learning the right things over time. Also, many times, organizations have a lack of control over the AI output and outcome. Additionally, skills required to build AI and time-to-market pressures are all common issues that can keep companies from benefitting from AI.
Q: Where do you see the biggest opportunities for the implementation of AI?
A: Like we are seeing autonomous vehicles with adaptive cruise control, we will see the same thing for automated business processes in any industry. The system will optimize towards the best business performance while staying within ethical and legal boundaries.
Q: What are your biggest concerns regarding the state of AI today?
A: The biggest concern is that AI is thought of as a magic bullet technology. There are right ways and wrong ways to go about using machine learning, natural language processing, and other related technologies. Finding the right use cases to attack is half the battle, and having realistic expectations about the outcomes is key to success. Those who do not understand the differences between data science and software engineering will quickly become frustrated by AI.
Q: What skills do developers need to be proficient on AI projects?
A: In an enterprise context, the skills developers need are a basic understanding of data science, including probability theory and statistics. There are many good online courses that teach this. Additionally, a strong background in SDLC and enterprise application deployment and experience with service-oriented architectures, business process management, cloud technologies, and abstractions are also very helpful.
Q: How do you make sure you are building AI systems you can trust?
A: Explainability must be built into every AI system — explainability for the end user (why did the system do this or recommend that?) and explainability from an audit and compliance standpoint are imperative. You also need to plan for and design continuous learning feedback loops into the end applications that are powered by the AI. You also need governance processes that provide assurance/guardrails for AI-powered insights and recommendations. And finally, you need a reliable mechanism to align business performance measurement with machine learning model performance over time.