Customer Story – G2IT / US Navy

Video Transcript

So my name is Jeff block me, I, as Matt said, I work for a company called GE to it. We do a lot of work with the Navy higher education and stuff like that. One of our big Navy customers moved to kinetic about four years ago, they had some specific authentication and branding requirements that their current self service technology couldn’t support.

So the first service request that they were all that was rolled out for this program was the one you kind of see on the screen, it’s a two page service request, where the customer comes in the first time, they have to kind of fill out all the form with all the contact information.

Hit next. And then they get the second page where they can describe their issue. And then we give them some pulldown menus where they have to select the product system or service that they’re having trouble with. And they have to select the like, it’s kind of the product and operational categorizations, the subject IDs that they have, like what type of issue that they’re having. And what we saw was customers kind of get frustrated by the time they get to the second page, they’ve already filled out the contact information, they’re going in filling out, you know, description of their problem. And now they’re gonna sit here and search these menus to try and find their product.

And Navy 301 is the project we need to throw in supports, it’s kind of a catch all help desk for the Navy. So if you don’t know who to call in the Navy, you call maybe throw one. So there’s tons of products and services that they get called on. Because they’re being the catch all. So the customer having to search through those and try and find the product that they want.

It’s kind of cumbersome, and then from there to try and hit these three tiers of menus to try and like, describe the actual issue was kind of frustrating. So one of the things that are the analytics team within maybe throw one was doing, they started they were doing some work with predictive analytics.

So one of the things that they did with predictive analytics was they took a look at the old tickets, ran the problem descriptions through a predictive analytics routine, and and had an output what the problem categorization the product categorizations, and the operation categorization should have been.

They compare that to what the helpdesk people selected. And they determined that they could actually more accurately predict what those product operational categorization should be versus having the helpdesk agent select it. So we then said well is integratable as kinetic is, could we take that same routine and plug it in on the client side, so the customer comes in, fills out a shorter form, which is, you know, the contact information is pulled down, they fill out a subject a description, run it through predictive analytics, and then determine that on the back end, and then we feed that into the helpdesk.

So that’s kind of what what ended up happening. So you’ll see it on the poster board out there as well. But it’s a more abbreviated service request that kind of looks like an email. So you come in the first time you would register. In this case, I’m already registered.

So it showed my name there, I fill out the subject, I filled out the message body, I chose a priority for this, certify that it wasn’t classified and hit send. When I hit send it called a predictive analytics routine using Java. It’s a Java function and calls calls predictive analytics routine and then returns a bunch of data from that.

And there’s a bunch of hidden fields on the service requests that get populated by the predictive analytics routine. So you can see we’re in this one, I said, I can’t log into CDM DOA, which is a navy application.

So running it through predictive analytics, it came back and said, okay, the commodity name, which is the product is configuration data management, the subject IDs are security access, you know, for an information system. And on top of that, we actually extended it to then predict who that ticket should go to. So now when it gets to the helpdesk, they not only know the product, operational categorizations, but it also predicted who that ticket should get routed to saving them some time.

So and for email requests early, right now we’re piloting with a bunch of different actual products and services within the within the within the project. And on average, it was taken about 23 days to from the time the email came in to actually getting assigned to a person and running it through predictive analytics that’s been shortened down to about two days. So this might seem like a lot to most help desks. But being a catch all help desk for the Navy, you get calls about all kinds of different things. And sometimes it takes helpdesk people a while to research what actual product it should be and actually who it should go to to get support. So any questions on that?

On that one? Okay. So the other non traditional thing we’re kind of doing with kinetic is we implemented a customer agent chat capability. So our service desk with the Navy through one was using a third party chat, which they were kind In the toll that they needed to shut down, because we were using a third party tool hosted off site to for Navy communications.

So they asked us, we had already done a service item and kinetic that was a little bit of a collaboration, where folks could just kind of come in and discuss on an issue back and forth a little bit. Not as robust as some of the Kinetic tools. But so they asked us if we could actually do a real time chat application.

So two of us worked about four weeks, and got the entire customer chat console developed and the customer, the agent, chat console developed it in kinetic request. So the previous example was a, this is the customers experience, they can actually see on the right hand side there, the request is Help Desk Agent logs it.

And then this is the agent experience where you kind of see the queues on the left hand side, the chat interaction in the middle, and then the ticket information show up on the right. So I can actually log in and demo that if you’d like. Pardon was that. So here’s the the, I’ve gotten to the customer chat item.

So basically just get a description box, it already knows who I am based on my cat card, which is the reason I kind of had to use my PC. So I could get in authenticate using it. So I just put a description in here, what colors are in? I can type in rainbow. And pardon? Yeah, this is live.

So my coworker Rob is actually on the other end in the agent console. So he should see. He said he would be monitoring it. So he says, and there it is. So it says Robert has joined the session. And then he says Welcome to nab 301 How may I help you. So he should have my problem description, as I said, How many colors are in rainbow?

So we can, we can kind of chat back and forth here and interact on his side. I’ll show you the agent console in a second. But on his side, there’s some canned responses. So that you know Welcome to the bathroom was a canned response that he put pulled from a pulldown menu. And so it saves them a little bit of time.

There’s also on the agent side, there’s a whisper capability. So if a supervisor you know is monitoring the chat cues, and see someone interacting and wants to kind of express something you know, in the chat session that’s only seen by the agent, they can whisper into it, and the agent will see it but the customer won’t see it.

So Rob must be researching this there are seven colors in the rainbow. Okay. Kim, can we provide further assistance? No. And when I’m done, I can hit and chat and I can actually end the chat from my side. And it’ll actually close out the session. So on the agent side, this is the experience that was kind of being seen by Rob as soon as he came in.

So up here at the top, if you click on agents, you can see who’s logged in. There’s three other agents logged in at the current moment, I’m going to set myself available. And you’ll see me now I’m showing available.

Back on the customer side, I can actually pull up a new session you’ll see me coming in. So I’m just gonna take test here and usually within a few seconds you’ll see the chat come in. I’ll assign it to myself.

So I’ve grabbed it before Rob did. So now you can see it’s in my active it’s blinking down here. So I can click on it, you’ll see the information here. This is where I can do the the canned responses.

And also where I can whisper I can type something in there and whisper it so that only agents can see it. And it kind of shows up kind of shadowed I can reassign it to different people. So if I’m leaving if it’s the end of my shift I’m in the middle of a long chat, I can reassign it to another agent that’s available. Perhaps a reboot would help.

So Rob is watching so good stuff. I can also within the console, so I can scroll down and see any open tickets for this individual. So if they’ve already have a ticket open on this particular chat session, I can just select the ticket from the list at the bottom here. And then I can choose link here and it’ll actually link this chat session to that ticket.

Once the chat session is over, and I archive it, we pass it over to kinetic task, kinetic TAS builds up the chat log, and dumps it into the work history of the of the incident. So you actually see the whole whole ticket in history here.

So that one was linked up. If I were to create a new one, if I click on New, it launches a new service item, I fill out a few fields and hit submit and kind of task goes off and creates the incident and links it up automatically with Jad session. So so that’s kind of the the customer agent chat interface, Jeff, what what parts of this app are using kinetically backbone back things like so this is a this all of this is a kinetic request service item.

Then we use kinetic task on the incident creation, the customer like people record creation, and on the linking of requests. So when a chat session starts as well, you’re putting like data in a data store too. Yeah, so when you actually start the chat session, we’re actually using callback in kinetic requests to actually create the chat session log. So, so it’s more real time.

Some of the challenges we had doing a live chat on here is when we first created it, we had each one of these tables kind of refreshing. And and those intervals started kind of colliding with each other, and freezing up the client a little bit and stuff. So we actually only have like one JavaScript interval, we have one JavaScript function running on an interval. And within that, we define some cycles.

So they’re stored in a form as data, we can go in and tweak them and reconfigure them, you know, as we want to, but we go in and say, Hey, every one cycle, you know, refresh the chat log, every two cycles reached refresh the incoming queue. We can kind of tweak those based on how network performance is.

Anybody else have any questions?

Where you using BMC? Remedy? Yep.

Anyone else? All right. Brendan, have a question. Jumping back to your earlier slide about predictive predictive analytics.

I was actually just curious. You said that the number of days went down from was it 24 Almost to two? What about the accuracy of the predictability? 

How did that turn out? So if you notice on the right here, let me bring it up a little bit on the wrong slide. It’s okay. Yeah, here. So on the right hand side, you’ll see next to each data element that the predictive model returned, there’s a reliability.

So it was like a confidence level on how confident the predictive model was on that value. So we bring those values back, and then we can act on them in the kinetic service item. So if it comes back and says, Okay, we think the product is this with a 70% accuracy, we can tweak the service item to say, if it’s less than 80%, don’t take it. Don’t don’t fill up the commodity, which is the product? Or if it’s, you know, so we can kind of tweak those thresholds. But you know, as we want to in the service item. Yep. Exactly. 

So just wondering, like, how many records did it take in order to like, build the brain? Like, if you only have five requests, it’s pretty hard to get any smartphone app. Yeah. 10,000 1 million.

So I think we’re going back a couple years on request right now for the model, I think we’re going on like a rolling two years, in two years will be about 2 million requests. So but on the like, when the customer hits the send button, the and it runs through the model, it’s I mean, it’s a brief second before the model is completed. I mean, it runs really fast.

The, the Java libraries are all cached. So it runs super fast. When it comes to the predictive models, consistency is is important. So like, if you’re if your helpdesk tool is using templates and stuff, it’s a lot easier to predict what they should be because the data is consistent.

If you have a lot of erroneous data in there, it really meant mocks you know, messes with your the reliability of your model. So, when they when they first start training the model, you know, you have to go in and look at outliers, you know, that you want to eliminate, so they’ll tweak the model and eliminate some outliers. And though they’ll put in some error handling trying to say you know if these values are novel, this is how we want you to handle it.

So there’s some stuff you have to go through to actually train and develop your model and tweak it. But then once you get it in place, it works really well. And if anybody was interested in predictive analytics stuff I could hook you guys up with with our guy that does stuff I’m sure you’d be glad to share. Thank you.