It should come as no surprise to all serious professionals that the age of Artificial Intelligence, Machine Learning, and Data Analytics is here – this time, to stay for good.
The first time around, back in the early 80’s to mid-90’s – a period of roughly two decades – AI got a lot of exciting hype, especially around the topics of Expert Systems, Natural Language Processing, and Computer Vision. The premise underlying all of AI at that time was that the human capability of learning was really a set of as yet unknown algorithms that knew how to take data of various types and reason from it. However, the common perception was that the promise of AI had not been met.
The current resurgence in AI stems from one fundamental insight: intelligence (at least of the actionable kind) comes not from some esoteric algorithms but from the application of comparatively simple techniques to massive amounts of data that is not necessarily well-structured. This is the reason why AI, ML, and data analytics continue to be linked to Big Data.
Switching context ninety degrees, business process management predominantly deals with transactions running through structured business processes. These processes are pre-defined in excruciating detail and standardized to work across various use case scenarios.
The first level of intelligence in business processes is to make processes self-aware of their state, to detect and report of performance metrics that are not dependent on the underlying domain or the actual transactions. The BPM platform becomes a process operating system. The computer operating system (OS) is constantly monitoring the performance of the computer, cleaning up memory, and dynamically changing the allocation of CPU and memory resources, and a number of other tasks. The OS performs all this without really caring whether the applications it supports are helping design engines or supporting sales activities. Similarly, the first level of iBPM would provide native support for process metrics without really understanding what the processes are doing. Examples of such metrics would include cycle time, variance, throughput, exceptions, value-added/non-value added ratio, etc., as applied to the process steps. None of this should require any explicit programming or configuration.
The second level of intelligence in BPM is to monitor the performance of the overall process itself from end-to-end, or some large sub-sections of it. A sub-process deals with some well-defined capability, department, or group of stakeholders. The BPM operating system monitors and reports on the process metrics of the whole sub-process and of the entire end-to-end process itself.
The next level of sophistication in BPM is the ability of the BPM OS to understand the interactions between the process steps and between the sub-processes. When the outputs of processes A and B merge conjunctively into process C, the BPM OS should detect when either A or B are delayed, thus delaying the start of C. Obviously, the system should be able to notify the stakeholders and process owners, rather than depending on people to watch the monitors.
None of this is revolutionary. The underlying infrastructure to enable this in a BPM OS is quite simple. But each capability, or level of sophistication, makes the BPM system behave in a way that seems to converge on human-like intelligent behavior. All this is linear evolution in intelligence. Switching back to the first context of AI, ML, and data analytics, we see a convergence of BPM and Big Data, with the idea that when the business process deals with large amounts of data, process steps that analyze this data impart more ‘intelligence’ to the transactions. In this sense, the intelligence is not in the BPMS but in the algorithms that the process steps invoke.
But there is a sense in which the BPMS could itself have higher levels of intelligence. This is where it morphs into cognitive BPM, which is saying a lot more than just ‘intelligence’. Cognition implies an understanding of context, interaction, behavior, and the ability to respond to these cues.
For example, we wouldn’t apply the label ‘intelligent’ to an average person at a party in the same sense in which we would apply it to Einstein. But when someone is in distress at this party, our ‘average person’ can detect that, show empathy, and offer support, just as easily (or perhaps more easily) as an Einstein.
This begs the question: is cognitive ability higher than intelligence or lower? We will leave that to the philosophers and focus instead on how cognitive BPM would make for some very interesting use cases.
Assume, for example, a customer service process that is interacting with a customer, say John, who starts out agreeably voicing his way through the phone tree. Let us assume that John is not getting what he wants. His voice shows increasing frustration. The system detects the frustration and its increasing level, determines that the trajectory of frustration is not good, and with a soothing voice, informs Joe that human help is on its way.
There are a number of emotion APIs that detect sentiment in the human tone and enable exactly this type of capability. The “old AI” approach would have us build a catalog of ‘frustration words’ and apply some esoteric algorithm to detect frustration. The more recent Big Data approach would have us analyze the massive amounts of call records to detect patterns of call abandonment and make a prediction about the likelihood of a particular caller exhibiting frustration.
The cognitive approach requires neither the “old AI” approach of some sophisticated expert system type of capability or the ability to crunch massive amounts of dynamic data to predict in real-time. Instead, it relies on, ironically enough, prior analysis of massive amounts of data (in this case, voice tones) to learn (through supervision) to detect frustration and take corrective action. That’s what these companies offer – a supervised cognitive capability to detect a target behavior and indeed, continue to improve the performance of this capability.
Here’s another to whet your appetite: Imagine doing a presentation (with PowerPoint, what else?) to an audience. At some point, your presentation gently suggests, through a flashing presenter note on the screen that your audience needs a break because many people are looking dazed, or to ask for questions on this slide because the audience isn’t getting it, or to use another example similar to the one you did on slide 7 because that really went down well.
Fantasy? Not at all. There is software available that takes a video feed in real time (say, from your webcam facing the audience, for instance), isolates faces, reads their emotion at millisecond intervals, and analyzes it. Hook up this capability to a PowerPoint plugin that takes that feed and provides suggestions, and you’ve got it made!
Granted, the two examples above are not true-blue business processes in the traditional sense, but they are important processes, nevertheless. Yes, if you apply the classic Turing test, which basically says, if it walks like a duck and quacks like a duck, it is a duck.
The age of cognitive BPM is here. Get on board!
