Amidst all the noise and hype around Artificial Intelligence (AI), it is hardly surprising that buyers are massively confused as to what the impact of AI for running their operations will be. Many lack an understanding of the technology building blocks, the starting points, and most crucially, narratives that depict use cases and outcomes. All too often buyers are being evangelized on the merits of evolving technologies such as machine learning or computer vision. The most accessible use cases tend to be either consumer-centric scenarios or use cases that cannot easily be transferred to horizontal processes. Amazon’s warehouse drones are a good example of the latter.
In one of our first HfS Think Tank sessions, we discussed these issues with the leadership team of Infosys’ financial services business unit. HfS' ThinkTank approach is all about getting the industry collaborating again. ThinkTanks are where we invite customers, suppliers, and advisors to drive joint problem-solving through Design Thinking techniques. It is where the HfS team challenges you, and you challenge us. In the session around automation and AI, we discussed emerging use cases that we have gleaned from our inaugural Blueprint on Enterprise AI services. In this report, we summarize the gist of those discussions.
In Highly Regulated Markets, Unsurprisingly the Focus Is on Data
HfS has stated repeatedly that the discussions for AI have different, often diverging starting points. Therefore, we have chosen a set of use cases to highlight the trends that we are currently seeing. Exhibit 1 provides an overview of some of the more compelling emerging use cases. In view of the stupendous amount of hype around chatbot, a good starting point are conversational services. While chatbots make a lot of sense from a consumer perspective on devices like smartphones, from an enterprise perspective the noise in the market is at times difficult to comprehend. Chatbots are low-level AI where the dialogue is scripted around a narrow topic and often linked to FAQ knowledge repositories. We had multiple discussions with buyers who said that those dialogues were built without asking their clients what they would expect from such tools. Perhaps unsurprisingly, many customers want more than a narrow conversation about a single issue and would rather discuss multiple issues at the same time when they interact with an organization. A fair few projects sputtered to a halt after such realizations. From an enterprise perspective, conversational services are about augmenting or even supplanting agents. Yes, they are also about providing the agility to progress toward the OneOffice, therefore aiming to provide more instantaneous interactions. But, without addressing the broader implications in a more transparent way, projects will continue to falter. Against this background, the use case of a virtual agent integrated with a Hadoop cluster by a Big Four consultancy demonstrates the difference to low-level chatbots. Beyond the vast amount of data integrated, the ability to feed every customer channel as well as leverage sentiment analysis in real-time call out the differences to chatbots, and therefore we at HfS refer to it as virtual agents.
While this virtual agent example is more the exception than the norm, we came across many examples of leveraging data science and machine learning to come up with better ways to deal with issues such as anti-money-laundering (AML). The algorithms developed help to set thresholds to trigger an action as well as eliminate false positives. Crucially, however, is that those algorithms deal with vast amounts of unstructured data such as external databases or information available over the internet. At the same time, we see deployments accelerating the progress toward autonomous processes that are critical to enable the OneOffice. Examples of this are a deployment for automated billing with unstructured text mining by Big Four provider. The specific use case is to scan fee schedules and invoices. An accuracy level of 96% and a revenue leakage of only 3% to 4% demonstrates a high level of efficiency. Similarly, a global BPO leverages machine learning to automatically aggregate and normalize information from multiple sources. The specific use case is on-demand performance reporting. A global SI went one step further by utilizing machine learning and other technology building blocks so that clients can drop disparate information for general ledger requirements with a view to be processed automatically. The last example in Exhibit 1 takes us in a completely different direction by suggesting looking for use cases outside the vantage point of a specific vertical that could be easily expanded to other verticals. Take the project of a global SI that had helped an Australian company to automatically identify telephone posts using Google’s Tensorflow and StreetView solutions to replace physical inspections. It is easy to see similar approaches for a raft of insurance use cases.
Exhibit 1: Examples of emerging AI use cases in Financial Services
Source: HfS Research 2017
Just Like in the Early RPA Days, We Have to Get Back to Basics
At times, it feels condescending to talk to audiences about going back to basics and different starting points, with completely different implications and requirements for projects subsumed by the “AI” moniker. But I feel reminded of the early days of the RPA discussions where the market got distorted by oversimplified communications, such as RPA tools being turn-key solutions and often being elevated to “non-intrusive and error-free” silver bullets. What was lacking was a clear articulation of use cases and outcomes that were meant to be achieved. With the discussions around AI, we appear to be at a similar stage. While AI is undoubtedly much more complex than RPA, the industry needs to focus on playing back the practical experiences from the early deployments. This has to be done in a language that resonates with the decision maker rather than the R&D teams. It is here where we at HfS put the emphasis on translating technology innovations into a language resonant to executives leading the transformation of service delivery and further on playing back the learnings from early deployments.
Bottom Line: The Market Is Accelerating Beyond Low-Level Chatbots and Machine Learning
Even though the industry is veering back to RPA discussions rather than expanding the perspective to capabilities that get subsumed under the “AI” moniker, the value, depth, and scale of those early AI deployments is vastly outstripping any RPA deployments. Just take the example of the emerging Watson ecosystem capabilities we highlighted earlier in the year. At the same time, we have to resist focusing on the communication of approaches that appear easier to grasp, such as chatbots and machine learning, though as BluePrism’s market capitalization just surpassed £1 billion, that might be a tough ask. That is why we need insights on the early deployments for much more complex and vastly different approaches. The longer-term direction of AI is to replace enterprise architectures and applications, not to just remain at the periphery of the enterprise or be a bolt-on solution. Therefore, stay tuned for our inaugural Blueprint for Enterprise AI Services, which will assess all those issues in greater detail.