Puppy Dogs and AI: Training Automation for Relevance
- Published: March 4, 2019
- Written by Peter Schooff
Peter Schooff: Hello, this is Peter Schooff, manager editor at BPM.com and today I have the great pleasure to be joined by Malcolm Ross, vice-president of product at Appian. Malcolm has been directly involved in the management and the implementation of Enterprise Software Solutions for over eighteen years and in this podcast we're going to drill down on one of the really hot topics of today that Appian has been right in the center of, which is intelligent automation. So, first of all Malcolm, thanks so much for joining me on this podcast.
Malcolm Ross: Thank you Peter, always a pleasure to join you.
Peter Schooff: Fantastic, now can you give me an overview of intelligent automation?
Malcolm Ross: Yes, I mean intelligent automation is obviously exactly what it sounds like. It's the application of automation technology to drive efficiency and better customer experience and intelligent interactions throughout your business enterprise. Now the question is really kind of what is the components of intelligent automation; how do I form an intelligent automation practice?
You know, Appian and a number of other people and myself have been involved in business process management, like yourself as well Peter, been involved in business process management, for a number of years, and of course we've seen a greater interest in automation technology over the past, about two to three years, which is great, but with that we've seen a bifurcation of interests in different technology sets where automation practices are building up around technologies called robotic process automation.
There's traditional BPM center for excellence, which has been around for a long time. There's applications of artificial intelligence to drive intelligent automation interactions across enterprises, and what we're seeing as a challenge is that many of these organizations are approaching these technologies separately, rather than forming a comprehensive practice focused on intelligent automation. Now that's what we're recommending, that everyone kind of take a step back and look at the overall conglomeration of automation technologies, bringing them together into a center of excellence or center of best practices around automation technologies and that should be a kind of foundation of an intelligent automation practice in any organization.
Peter Schooff: Interesting, you know what's really taken me by surprise is just how hot RPA still is but inevitably it's not that smart so, automation has been around for a while so how is intelligent automation different from the automation of the past?
Malcolm Ross: Automation of the past, you can go back, what two hundred years when it comes to automation right? From the foundation of the industrial revolution, automation has always been the key to building a successful enterprise, to be able to do consistent interactions with customers, build products consistently, improve quality; that's the foundation of any organization.
Now what changes is the technology, so for the past twenty years business process management has really risen to be a foundational component of overall process orchestration. BPM though has heavily aligned with process improvement best practices, such as back in the 90s there was a best practice called TQM, total quality management, then it evolved into six sigma in the 2000s and it kind of ebbs and flows with best practices for overall process analysis and process improvement.
With those self-improvement cycles and process improvement cycles tend to have very strong hype cycles where Six Sigma is really popular and then suddenly it really falls out of favor because its overly focused on just process efficiency and maybe not overall customer experience.
With that, we've seen automation start to rise in a number of more discreet areas, such as, as you mention RPA; robotic process automation. Although robotic process automation has also been around for over twenty years, at the core of it is desktop scripting, which has been around since the mid 90s. Robotic process automation is really identifying the key value point its able to deliver by the large amount of repetitive actions that have been growing inside enterprises, repetitive work and repetitive tasks. A better name to call robotic process automation, robotic task automation because it's really focused on those highly repetitive, core use of human capital tasks, to drive that work over to a digital workforce and robotic process actions to drive more efficiency and more data quality throughout the enterprise.
So robotic process automation has become really hot, but then again it's a tool in the toolbox, it's good for a specific thing, not for the overall thing. And on that same front, artificial intelligence has also gained a large amount of interest over the past years. Like other process automation technologies, AI is not new, it's been around for decades; the concept of applying predictive analytics and statistical analysis of past data to drive future decisions, that's been a core of statistical analysis actuarial mathematics for a long time, but what's different with AI is the evolution of the late 2000s in big data, and now also application of cloud technologies that makes it very accessible to have access to the computing resources needed for AI, it's only become mainstream.
So again, AI is being looked at in a discrete corner, where you have wholistic process orchestration and BPM, I have predictive rules, I can apply AI to predict outcomes based on past data, that's very important for things like triaging emails, for example, determining where it should go, it's very good for predicting insurance policy claims, very good for lots of different specific rule purposes, and then of course you've got robotic task automation, as I mentioned before robotic process automation for automating tasks.
Together, they all do automation, but they all do it discretely. What we need to do is think about how we combine these together in a more wholistic practice and that's really what's different, is taking these technologies together, binding it together in an overall automation practice.
Peter Schooff: Yeah I could see how that would make a big difference. Seeing as it's relatively new, intelligent automation, in which industries do you see it having the biggest impact right away?
Malcolm Ross: The biggest industries we see biggest impact on is primarily the digital industries, so those include things like financial services. If you think about bank operations, a large amount of what goes on in the bank is a digital operation; processing forms, moving money, balancing accounts, it all works in the digital realm.
Another area is where there's lots of labor-intensive activity, such as customer contact centers, where I need to have customer interactions and aim to automate those interactions in a more intelligent way, through applications of robotic process automation, eliminating tasks, overall process orchestration, understanding customer journey, AI to predict customer interactions, any contact center solution across any specific industry can definitely benefit from this.
So those are the areas that we see, I would say, primary adoption of these technologies of intelligent automation. The general contact centers and financial services but broadly though, these do have application to every single industry, and as you said before, every single business, every single organization should really have a foundation on strong process automation capabilities. This is how I generate repetitive products, repetitive customer success, that should be core to your business, so every single organization should be looking at these technologies. Some you'll just see adopt it much more quickly because they have the much more digital focus interaction.
Peter Schooff: I agree. Digital is such a big leap but the companies that have done it are so far ahead. Now this next question I could have probably asked you twenty years ago about automation, and I'm kind of hoping this sort of boils out the reason it's intelligent, is how do you make sure intelligent automation, in the automation side, you're not just automating mistakes?
Malcolm Ross: Yeah, I mean that goes back into good old BPM process best practices improvement so, how do I make sure I'm not just automating bad processes, but actually improving processes over time? So, while automation technology can be highly beneficial to drive efficiencies and better customer interactions and such, again if it's a bad customer experience you don't want to take that on and automate.
Let me give you an example. So, let's imagine I'm trying to improve the checkout counter line at a retail store. We can imagine a checkout counter line is like a process. I have an number of things queued up, those things are customers waiting to purchase their products, and a number of actions I could possibly improve, such as I could add additional checkout counters to the checkout process, I could also take a look at the time I schedule product releases, so maybe I stagger my product releases so I don't have a crowd of people on a single day. Lots of different things we can look to improve that process, but at the end of the day, we should ask ourselves, "did any customer like waiting in line, or waiting in a queue?".
It's not really desired by anyone, so we need to always think about the goal, the outcomes of what we want to achieve. Don't just look at the process, don't just say "hey I could remove these things, move these things", let's look at the customer experience, let's look at the asset, the products, and what we call this in Appian is taking a record centric approach. It is looking at the business object I want to identify to improve, and identifying the correlate processes that it interacts with, and in this case a customer, such as, I want to improve the customer experience, Waiting in line is one of those things, that I want to improve. But, again, don't just look at the processes, look at the overall objects I wanna improve.
Same thing applies if I was an airport. Let's not just look at the turnover process for an airplane; how quickly they can get it to the gate, swap out the plane with new customers and then get it back into the air. Let's look at the plane itself; what are the goals for the plane, what are the interactions I have with it? Taking a more wholistic approach to improving core elements of my business, not just processes, if that makes sense?
Peter Schooff: That makes a lot of sense. Now you've already given a significant number of examples, but how about just let's just take one true use case, of where AI has truly impacted a company today.
Malcolm Ross: I guess the first thing to understand is you're not gonna build a sentient being out of the gate with your AI practice, so keep expectations in line and look for quick, easy wins in artificial intelligence, don't try to build the sentient being per se.
So a great way to start adopting AI services right away, is to just start adopting package cognitive services or partially trained cognitive services. These are relatively plentiful; if you go to the cloud services from companies like Google and Microsoft, they have plentiful amounts of things like natural language processing, sentiment analysis, intent analysis. These are usually common fully trained cognitive services, or what's called partially trained cognitive services, the partially being that you need to add a little bit of information such as your unique data set for your organization, and how that maybe applies that decision.
So by looking at these low hanging fruit of cognitive services, and I take a look at how I can drop that in and improve, such as, wouldn't it be great if I receive inbound email communication from my customers, if I could automatically apply things like sentiment analysis? I can very quickly get an estimation of whether it's an angry customer or a happy customer. Very easy to apply and get a quick win.
Another thing that's even better is, what if I do intent analysis? Things like adopting the Microsoft Azure LUIS service, language understanding intent services. And what that's gonna do, is again it's partially trained and it's gonna allow you to say "well analyze the intent of the email" and say "well is this a product return, is it a new change request, is it a new order?", and then automatically route that, eliminating that level one triage in my customer contact center. So I can get that efficiency, apply AI to get recommendations, and as I use AI more and more it gets smarter and smarter.
Now as you think about evolving the AI then, the next step is really staring to build your own unique predictive rules, around raw machine learning and artificial intelligence services. So this means that you're building up your own unique data set of customer interactions, or specific interactions, I'm feeding that data set into a machine learning service to get a prediction outcome back.
Now that's even greater value to your organization because it’s unique to your organization but also it’s a lot higher cost, and you always need to look at AI as say, you’re adopting a child, or a puppy. It's not a one time thing where I simply adopt that child and, great, he's my child! No, you need to raise this thing, you need to train it, you need to have a continuous education process to make sure that AI service stays relevant to its predictions.
So, keep that in mind if you take on that higher level step. And then the highest level step is looking at then, master process orchestration tools, like BPM tools, trying to stitch these things together into wholistic processes and also stitching multiple AI services together.
For example, as I said before, combine sentiment analysis with intent analysis. Those are two different AI services. How do I orchestrate these? Through BPM I can orchestrate those services.
Now this gets you closer and closer to say, that sentient being, because if you think about the human brain itself, what is it? It's really not just a single AI, right? The human brain is actually a number of AI services. I have vision recognition, I have natural language crossing, I have sentiment, emotional analysis. I have all these capabilities, and the neurons in my brain are all stitched together so, again that's your overall highest level goal is starting to combine these things together into overall unique process orchestration, that define your unique organization.
Peter Schooff: Very cool. Excellent examples as well. So, this feels like we're at a fundamental shift, so what is the one key takeaway you want listeners to take away from this podcast about intelligent automation?
Malcolm Ross: Well, let me first take a step back and ask the customers to take a step back and look at all the different areas being invested in your organization around intelligent automation. Start to consolidate those. You see a lot of people adopting a single tool, like a BPM or RPA, or AI, and when you have a hammer everything looks like a nail. You don't want to take that approach, you want to have an overall intelligent automation practice, with a strong grounding of four key technologies. As we said before, BPM and rules, artificial intelligence, robotic process automation, and the fourth one that's very important is integration. How do I integrate data services together into these overall automation practices?
Those four technologies should be the foundation of a wholistic, intelligent automation practice so, stop the siloed automation applications and the siloed projects. Consolidate into intelligent automation center of excellence and make smarter decisions of where you can intelligently apply the right tool to the right problem.
Outside of that, don't abandon agile principles. Especially when thinking about artificial intelligence and continuous evolution of your automation practices. By agile principles, shoot for the MVP. The minimally viable product. Don't try to build a sentient being, don't go into a waterfall project and try to have a three year timeline for building out a comprehensive automation. Look for the quick wins, adopt cognitive services, adopt little process wins, with BPM and RPA technology, and then grow from there and expand and expand and expand and continuously be evolving this.
As I said, understand that if you take on artificial intelligence off your data sets and building your own AI rules, that is never a project, that is a continuous activity that you're constantly retraining the system, retraining its predictions to make sure the relevance it has as your business changes.
Peter Schooff: Fantastic! I could definitely see how intelligent automation would be a competitive differential, almost in any industry. This is Peter Schooff, speaking with Malcolm Ross of Appian.