top of page
  • Writer's pictureMarty Pavlik

Unleashing Generative AI with ServiceNow: A Process Mining Approach

Unlock the power of process mining and generative AI with ServiceNow!


In our latest webinar, we revealed how to leverage ServiceNow's process mining capabilities to identify and prioritize AI use cases for optimizing ITSM, customer service, and HR processes.


Dive in for insights into Doculabs' proven methodology for continuous process improvement.


Here's the replay or keep on scrolling to read the very lightly edited transcript.



Tucker: Good afternoon everyone. Thank you for joining us on this glitchy Thursday afternoon. We appreciate you taking some time and learning about process mining and ServiceNow.


We're going to spend about 30-35 minutes talking through some use cases and some great opportunities for you to learn how to leverage ServiceNow's process mining capabilities from the Doculabs team.


As I start, we're going to introduce you to our two speakers for today. First is going to be Marty Pavlick, who's our EVP and practice leader and leads our process mining and ServiceNow initiatives, as well as Richard Medina, who is one of our founders and is the leader in our ServiceNow practice. We're looking forward to hearing the great stories and insights that Marty and Richard have to share. Marty, Rich, on your way.


Pavlik: Doculabs is a 30-year-old process optimization firm, and over the last few years, five years, we really focused on process mining. So in today's presentation, we want to teach using some of the things we've learned from using process mining, specifically how you could start using ServiceNow's new process mining capabilities to start identifying generative AI use cases. More importantly, the second objective is to leave you with a methodology on how to prioritize those use cases, those process improvement initiatives around mainly ITSM, customer service, and human resource delivery.


With that said, we'll jump to the next slide. So throughout the last almost decade, consultants have really been providing their clients with the advice of "focus on the future to be." You know that is where you want to get to; however, the design path is ultimately not what is really going on in an organization.



It looks more like this reality picture here, where you have a lot of various process variants, you have longer cycle times for this process, and by taking into that like design path of the happy path To-Be path ignore some of the fundamental reasons why there are these process variants. For example, you may have a large customer like Wal-Mart or Home Depot that is requiring you to deviate from your process, whether that's processing an invoice, shipping a good, right, which is a justifiable variant, which is okay, right?


But when you're looking at it from the To-Be, you don't take that into account. So process mining gives you that visibility into those tier 2 and tier 3 process variants and helps you identify whether or not you need to address that issue or if it's a good issue or I shouldn't say even an "issue" issue at all.



So why did we choose process mining and focus specifically on ServiceNow? This methodology could be adopted with any process mining tool. We're focusing on ServiceNow for a few different reasons. One is that if you're a ServiceNow customer, you may already own this capability, and you're aware of it. If not, it's easily able to be secured because you already have a ServiceNow license, so it's a longer or shorter procurement time. And then more importantly, if you're going to run process mining on ServiceNow solutions, the go-to value is much quicker, mainly because it's a native data model. It's ServiceNow process mining built on ServiceNow processes, and there's a lot less time in data modeling, data reconciliation, data validation, right?


So what process mining is able to do with inside ServiceNow solutions is, I kind of compare it to like a football game, right? If you go to ESPN at the end of the football game, there's great reporting, right? So there's great native reporting already in ServiceNow, but what process mining will give you is more the play-by-play. It will walk you through what plays are working and what plays aren't working, and it's able to do this mainly by taking the event logs from the workflow, identifying three common denominators, which is a unique identifier, an activity code, and a timestamp, and from there, it will be able to give you a digital X-ray of your process. And on top of that, you'll start being able to identify key KPIs like cycle times, number of variants, but then in addition to that, we'll be able to add additional data elements onto that to give you more data analytics and insight into the root causes of some of your processing inefficiencies.


You might be thinking, does it require that I use ServiceNow process mining?

 

Well, that really depends, right? So if you already have a process mining tool, let's say you know, you have it running in order to cash, you're looking to implement or have some type of ServiceNow workflow. It may not make sense to use ServiceNow right in that particular circumstance. But now, if you have a ServiceNow ITSM workflow, and you're looking to process mine that, and you already have a third party that you're not using for it, you really need to do a cost-benefit analysis on that to determine whether the new license cost versus the speed to information, I guess, if you will, is worth it. So it really depends, but that's a great question, Richard.

 


So as I mentioned earlier, Doculabs has been working in process mining for quite some time, and we've worked with all the various tools and done over 30 implementations. This is our process mining optimization cycle. I want to walk you through this because it's important. Some people view process mining as almost a one moment in time that I'm going to do it, I'm going to find the root causes, and I'm going to go on. But really, to get the full value out of process mining, you really have to look at it from a 12-month window, if you will, right?


In the first 30 to 60 days, that's when you're going to start visualizing your process, start diagnosing some of those root cause problems, identifying where your process improvement focus should be, and then also developing a list of prioritization of process improvement initiatives.


And from there, in the next two to six months, depending on that process improvement initiative, that's where you're going to start focusing on improving the process, right? That could be done in technology, go-to fashion, process reengineering, as well as human capital management. Once you have those improvements done, that's when you start monitoring it. And that's the key, right? You want to make sure those investments you made in that improvement cycle period are paying off, right, and that they're giving you the results you want.


So this is a whole new way of really monitoring your business case from a technology investment. I think in the past, you build out that business case; you don't want to go back to it to see if this, if this stuff works or not, right? Now you could actually monitor it on an ongoing basis by keeping process mining live, and then most importantly, the last phase, which is really coming at the end of the 12-month mark, is evolve.


Evolve the process around key business impacts, and that could be a couple of different things. That could be an acquisition you've done, that could also be a new product launch, a whole new restructuring of the company, right? Continuously use process mining to evolve the process and not just use it for that snapshot moment in time improvement opportunities.

Now, Richard, you want to jump in here and kind of walk us through some of the generative AI capabilities of ServiceNow?

Medina: Sure, so what I'm going to do now is, Marty talked about process mining, so we'll talk about how process mining works with AI, specifically ServiceNow's AI, ServiceNow Assist, and I'll give you some examples.


For example, you use process mining for Human Resources onboarding delays, right? You focus in process mining on that, and it will highlight stages in the HR onboarding process where new hires experience unnecessary delays, such as waiting for IT equipment provisioning or trying to get assigned access rights and so on. So process mining identifies that, and then ServiceNow Assist, the AI, can then automate these steps, ensuring that new hires have everything they need starting from day one, streamlining that onboarding process and experience.


The second one I want to go into is in customer service. So you point process mining at customer service and you find bottlenecks. Through the analysis of the event logs, process mining can uncover bottlenecks in the customer service workflow, such as repeated information requests from customers due to unclear communication. And then ServiceNow Assist can then be deployed to generate more accurate responses to common inquiries, reducing resolution times and improving satisfaction. And I'll go into more much more detail on this in just a few minutes.


The third type of use case we want to talk about is ITSM, the candidate here is ticket routing. So you point process mining at ITSM, and you can reveal inefficiencies in how tickets are categorized and routed, leading to delays in resolution. And then Assist, the AI, can better understand the nature of the tickets from natural language descriptions, ensuring that they're immediately sent to the correct team or individual for faster resolution. So that talked about process mining and AI, and now I want to talk about AI and automation.



We talked about process mining and AI. Now let's talk about AI and automation. So now, actually executing the solutions, and they can be applied across ServiceNow. I'm going to use the use cases that we've been describing, which is HR onboarding and customer service and ITSM. And a good way to think about AI and automation is in terms of five buckets.

The first bucket is, I'm using the graphic here, uses slightly different words, but it's basically improving task handling. And so ServiceNow uses AI to manage and automate complex tasks, so the tasks are the activities or the work in the flow.

 

Secondly, they simplify complex operations. So this is the stringing together of activities into a complex process, and I'll give examples, but they simplify that dramatically.


The third is speeding up responses. So for example, by using AI for common inquiries, think of, you know, a new employee or a customer service customer or an employee with ITSM, so ServiceNow ensures faster and more accurate responses.


The fourth is the ability to leverage lots of data. So that's data-driven decisions, so the ServiceNow leverages AI for analyzing vast amounts of data, enabling the business to predict the needs of the customer or the onboarded employee or the employee wants ITSM.


And then finally, much better user experience. So it, ServiceNow's AI enriches the user interface and experience by personalizing experiences.


So let's go into some examples for the task handling for example for the onboarding process. ServiceNow can automate documented exchanges with the onboarded employee and provide automated learning modules. For customer service for task automation, the AI categorizes and routes customer inquiries, speeding up resolution. For ITSM, it does the same thing, classifies and prioritizes IT issues, reducing manual effort.


And there are examples for each one. I'll just give, I'll give probably the most popular one, which is speeding up responses. And that is, for onboarding, you, ServiceNow AI-powered chatbots for common new hire queries, speeding up the information delivery process, analogous in customer service, where you can get immediate answers to customer inquiries. And in ITSM, you can get, the AI can automatically suggest solutions to common technical problems. And so across the board in each of these five areas, ServiceNow can provide AI-based automation solutions.


Tucker: So Marty, there's a question here from Guy Thompson: "Can a non-process mining native person use ServiceNow Assist to successfully use these processes, or do you see the newer generation of this to be able to do that? They can be many of the inefficiencies in HR, customer service, and ITSM hit you in the face or do so with a spreadsheet, and it's more the more subtle ones that require process mining and so on?"


Medina: Absolutely. The process mining itself is the diagnostic tool and the monitoring tool once you fix it, and generative AI is probably the most powerful, coupled with automation, is the most powerful tool to do the remediation, to do the fixing. Marty, do you want to rebut what I just said?



Pavlik: No, I totally agree. A lot of it depends on the complexity of the process variance and how you want to automate them, right, in the level of detail you need to go into to automate them. So going into the kind of the first use case, and I'll even give you an example of how to think about that question in this HR Town onboarding example.


I spend a lot of my time working in HR with clients from professional services to retail, and the number one problem we see in HR is you have this cookie-cutter template of HR processes for demands that are very personalized. And when I say personalized, think about when you’re onboarding somebody, you're talking about health benefits, you're talking about income, specifically also tax questions, all these things that are very personal.


Where the real hiccup comes in HR is usually around a time. In retail, for example, a lot of hiring is being done during the holiday season. For professional services firms, depending on what type of sector they serve, it could be after the holidays that's when their big hiring pushes, or it could be hiring new graduates out of college in fall, right?


So the real stress in HR comes from two things. One is the fact that it has to be personalized. Secondly, there's usually a big volume constraint on HR departments and processes when those timings are coming in. So how do you leverage process mining to identify where those key areas are that you want to kind of tackle on for automation?


Now I know ServiceNow has a couple of different HR solutions, but for this example, I'm going to focus on case management. And when you're looking at the case management and ServiceNow, you really have to start segmenting your data elements into what is going on in those workflows. So for example, you know, taking the basic process mining functionality, you're going to immediately get the process flow, which is great. That's going to give you two key important elements. It's going to give you cycle time as well as the number of variants, and from there, you could start building out what categories you want to go after to really start identifying the root cause.


So you may be looking at it from a role position; who's the case manager, how they're handling these cases. You may look at it from a departmental standpoint. You may also want to look at it from a geographical standpoint? So all this kind of further data analytics will help you start really pinning down where that root cause is, and from there, you could start identifying what the categories are, right?


There could be a couple of categories that you're looking at that are causing process variations. It could be a payroll discrepancy, for example, "My withholding tax is wrong on my paycheck. I'm not really sure why.” All that, all that right there could be a case.

You may also have benefit questions. Like if I get married this year can I add somebody in mid-year, those types of question. I'll give you another good example, time submissions. I'm a new employee. I'm wondering if I get paid for overtime, what is the appropriate time submission that I need to kind of, what's the policy for, is it every 15 minutes, every 30 minutes?


All these things could cause a case in which could cause a process variant to understand, you know, what is the inefficiency of getting this employee the right answer at the right time? So if you looked at this slide, this is really the "what" right, like what do you need to automate, how do you need to automate? So from there, you'll start getting kind of some of your root cause analysis, but now the important part of this, and this is the Doculabs blueprint, and really how to understand where the value is in your process and efficiency, is what we do here.


We do this in a couple of different ways. We build out a dashboard within the process mining tool, and we'll also build out some other deliverables, but really what it revolves around is taking that process flow out of the process mining tool, mapping it to the specific business objective it meets, but more importantly, we map it to what is the KPI you're looking to measure.


So I'm going to pause there. That's a very important point because you may have process improvement initiatives that aren't really impacting your cap, and with that process improvement initiative, right? So for example, that onboarding, you know, if you have a lot of new employees coming to you as far as saying, "My payroll is screwed up, my paycheck is screwed up," which trust me, if you want to have a low employee satisfaction score, there could be a number one reason, right? You mess up with somebody's pay, they're not going to be a happy employee, right?


But then it could also be, you know, a tax question? In whether that tax question is in Portugal versus the US, how does it get responded, and then also, you know, benefit questions, which are more low-hanging fruit. So from there, you'd start looking at kind of what those process variants are, start mapping it to the key KPIs, and start thinking through which one of these improvement opportunities could really impact that KPI. So as you could tell, out of these projects, you're going to have a laundry list of improvement opportunities, right?


Which now you're seeing yourself, "Well, that's great, but I have to get funding for it." So I could see how I have to impact a KPI, but then how do I prioritize that? And that's kind of our next slide here is our Doculabs impact effort analysis methodology. And this is, if you had to think about this slide, is along the Y-axis is the impact, right?


Going back to that KPI, this is where you're going to impact either from a quantifiable value perspective, from a resource perspective. So think about that customer, customer not customer score, but the employee satisfaction score, right? That's where we will chart this kind of impact.


Then along the X-axis, this is the effort. And now this is where we start measuring out the technical complexity of the solution, as well as the impact it would take to make that change with inside the process, and then any other kind of human management element to it, which would increase the time.


Ultimately, how long is this process improvement initiative going to take? I'm not going to go through all these because I don't want to steal from Richard's time, but I do want to focus back to kind of that case management example I gave, right? I said three different categories. I said a payroll discrepancy, a benefit question, as well as a time submission question.


So let's say we home in, and we say, you know what, we think a chatbot, generative AI chatbot could really remove this process variant, increase the efficiency of the process. How do I determine whether or not that is the right solution? So going back to my example, that payroll discrepancy, if the main variances are, and let's say, you know, 15% of their variances are tax questions versus 30% or 50% of their variances are more basic HR benefit questions, the HR benefit question may be the right solution, right?


Because why?


Because it's quick to kind of implement, it's a high impact, and it's a low effort. Versus if you're going to tackle the tax variances and the tax inefficiencies, those are complex problems, right? You have to have current up-to-date information around the tax law in your LLM model. More importantly, that's a very specific advisory type question that you might not want to put into the chatbot. So being able to get into those details of what the root causes are of the process issues and the process variances is key to having a successful generative AI solution. Richard, I said a lot, why don't you take over from me? I need a sip of water.



Medina: Sure, flip it over, and I want to address, let's see here, Ranna's question, which was, "Hey, can you provide an example of enterprise-level scenarios or specific domain type applications like healthcare?" I think we are with, it's not with HR and ITSM, though it is enterprise-wide, certainly with customer service. That's a general model, so the front door to the outside world and also actually internal customers as well. But these can be generalized, the methodology that Marty talked about that I'll talk about with two use cases can be generalized. Now, anytime you have like the FedEx model, a customer requests something, so you assess it, you process it, you deliver it, and then you confirm, anything like that very easily falls into this.


And so, we're going to talk now about customer service. As Marty said, this is one of the main parts of our methodology. So we're going to go through it. We're going to go through at a business level, assessing where, what you might do, and then using process mining, and then using, looking at the candidates of automation and gen AI with ServiceNow Assist, and then actually identifying and prioritizing, getting budget for the projects you want to do, and then go to the races.


So you start here with the methodology, right? The process is on the left-hand side, and you all have, everyone has a model, everyone's doing customer service, whether it's good or bad, designed or not designed, but it's something like that with those four steps. And then, if you're managing with the end in mind, you've got some business objectives for customer service to be successful.


So it might be something like resolution cost reduction, customer satisfaction, risk mitigation, usually it's a weighted mix of those three and maybe some others. And you manage those business objectives by using KPIs or key metrics that let you assess how you're doing and measure how you're improving and so on.


So for resolution cost reduction, it's average resolution time. For customer sat, it's NPS and a bunch of others. For risk mitigation, it's compliance adherence rate or error rate or something like that.


Then you bring in process mining, and you want to visualize and then diagnose your processes and see where you're failing short in those metrics, and you want to particularly focus on the activities where good or bad execution of the processes with those metrics means high impact, high good or bad impact in terms of the important metrics for your important business objectives.


If our organization here wants to focus on customer satisfaction, then we're looking at net promoter score, customer sat score, customer effort score, and the activities that impact those the most. And so where do you find problems in this fictional example? You're going to find exam problems in the first two steps, receive and classify and resolve, and the problems you're likely to find are, for example, repeated information requests, wrong classification and routing, routine inquiries from the customer that take the same long time every darn time to resolve, bottlenecks during peak times, and poor service after hours and off days.


There you have it, you've done the assessment and the diagnosis, those are the things you've got to fix. So then you look at the methods to improve them, and if you look on the right-hand side, I'm just going to point out a few of these. So for resolution cost reduction, you know, it's no surprise, it's AI-powered self-service with chatbots, and it's predictive analytics, so you can predict what the customer will have a problem with before he has the problem with it.


With customer sat, it's the same sort of thing, it's AI-personalized interactions and even sentiment analysis to see the customer getting angrier and angrier. Then down with risk mitigation, you can see it's things like fraud protection during interactions. Okay, so then the question is, that's great, we got all this. So it looks like we have three plus four plus four, 11 opportunities. So how do you prioritize them, what should we do first?



So let's go to the next slide, Marty, and this is the impact versus effort analysis that he was talking about. And you can see it's an easy way to categorize your opportunities by what good impact will they have to the enterprise if you fix them versus how much pain you're going to suffer trying to fix them. What you typically find is this sort of sorting. It's very important. However, it is sensitive to business objectives. That is, right now, this graphic, I'm focusing on customer, I'm prioritizing customer satisfaction. But if I were going to focus on risk mitigation, it'd be completely different.


For example, fraud detection very indirectly improves customer satisfaction but is the number one thing for fraud detection, you know, I mean, for risk mitigation. But what you see here is that the quick wins, so typically the projects you go after first are AI-powered self-service using chat, using Assist chat, and sentiment analysis can often be useful.


Then you go down, and you see that there are projects that are very high value, high impact, but they take more effort, and the two I have here are predictive customer support and AI-enhanced feedback. And the reason they cost is it's obvious to see why they cost, why they're effortful, and that they require, you know, significant data analysis, model training on the part of the AI, complex integration with your business systems, and so on. But this is an easy way to assess your projects and prioritize what you should do first, second, third, and so on.


So let's take a look at ITSM. ITSM is similar to customer service in many respects, but the problem that today's organizations have is more complex IT environments, right? With alert noise and cloud-native complexities, hybrid environments, data volume management, and so on. So the traditional ways of doing ITSM can be less effective. Enter process mining to diagnose your problems, and then advanced solutions like ServiceNow Assist to help understand and automate the solution to your problems.


You would, and we've got the same methodology all but here for ITSM. So for ITSM, you have again the left-hand side, you have the general model of the process, the happy path, or at least the most common general path that ITSM takes. And then you have the most common, you know, you're probably your business objectives: operational efficiency, service quality and reliability, risk management, customer satisfaction, and their relevant key metrics in an ITSM setting. And then your improvement opportunities, some of which are familiar friends that we just talked about, like AI-powered self-service and predictive analytics, but others are different, like automated ticket routing and even down at the bottom, dynamic translation.


But you see if you go to the next slide that here again you have them prioritized by impact versus effort, where AI-powered self-service and predictive analytics are typically the quick wins, and dynamic translation can be useful. It might be number one in particular in certain contexts, but on the other hand, it often doesn't dramatically affect the core ITSM objectives and is often difficult because for multiple languages, dialects, though that's becoming easier because of gen capabilities and translation.


Pavlik: I hate to interrupt you, but I want you to add to this. When I was looking at this and thinking through it, I forgot to mention in the effort part in the last two months specifically, but it has been coming up quite regularly is around AI governance and risk. And more and more AI governance committees want to start looking at where these LLMs are really pointing at, specifically, you know, for the chatbots and whether or not there's any bias and what the risks are. Segmenting that data into exactly what the problem and they're going to try to solve for is very critical at this stage in generative AI adoption. I don't know, Richard, if anything to add to that, I just wanted to add that. I know I forgot to mention that earlier on my slides.


Medina: The trend today is anything you do with data or documents or processes or AI should be governed. And, you know, having a unified view of how you're addressing it and then treating each one differently as it is different. So data and document IG, information governance, is the most mature. Process governance is becoming more mature, and AI obviously is the youngster, but it is very important that you have it.


The three key problems with AI, is as Marty indicated, is bias number one, number two transparency, what the heck is it doing? I don't understand, and I certainly can't explain what just happened. And then third is security and privacy.


If you are using large language models, and you're doing, folks are throwing stuff out into the Internet and bringing it back, that's a problem for PII and other stuff getting out there. But to have a rational, structured, rigorous approach to AI governance is essential moving forward.


Tucker: So as a wrap-up, you know what Marty and Richard walk through that there's lots of opportunities within ServiceNow. Part of it is to decide and triage what is most important, the resources you need, identify the metrics, what does the data set look like, and how do you tune and iterate because it's not a one-and-done fix, it's a multi-stage investment.

We use process mining to help organizations analyze your firm's readiness to adopt generative AI. We've gone through various examples today of what that looks like.


For every organization, it's going to be a little bit different because of your maturity and AI and the capabilities that you've grown out, but we're looking forward to the opportunity to help guide you on that journey as you go along and look at your process and automate what you're trying to do.


What does an engagement look like? We're going to review the organizational capabilities, we're going to identify five to ten use cases within the organization where candidates for generative AI capabilities in the next six months.


So looking at each of these processes examples like Marty and Richard laid out, each of those is going to identify a multitude of different AI use cases that could be important in various parts of your organization. We'll then help you prioritize and maximize the opportunity for you to grow. Typically, that engagement takes about eight weeks.


We are not trying to move in and do process mining forever. Process mining is just the beginning of your journey. The automation and the true use cases that you end up building in your ServiceNow instance will be what the real work is once you get done.


We have additional free resources available. And we look forward to helping you out!


  • Process mining tip of the week

  • Process mining training videos


Doculabs Insights Newsletter -- Read and subscribe here.

  • Bi-monthly updates on trends in Process Intelligence


Service Inquiries

42 views0 comments

Comentarios


bottom of page