Duration 26:27
16+
Play
Video

Improving customer experience through journey...By Srinivas Chilukuri, Data Scientist, ZS Associates

Srinivas Chilukuri
Data Scientist at ZS Associates Inc
  • Video
  • Table of contents
  • Video
Video
Improving customer experience through journey...By Srinivas Chilukuri, Data Scientist, ZS Associates
Available
In cart
Free
Free
Free
Free
Free
Free
Add to favorites
33
I like 0
I dislike 0
Available
In cart
Free
Free
Free
Free
Free
Free
  • Description
  • Transcript
  • Discussion

About speaker

Srinivas Chilukuri
Data Scientist at ZS Associates Inc

Srinivas leads the AI Center of Excellence at ZS Associates. He focuses on helping clients through their AI journeys to realize business value by tailoring, designing and implementing AI solutions for them Srinivas’s core expertise areas include automated machine learning, natural language processing, customer omni-channel next best action orchestration, recommender systems and longitudinal patient predictive analytics

View the profile

About the talk

Proactively customer experience management is increasingly the norm across industries. The rich granular interaction or transaction level data collected for each customer enables companies to understand their customer journeys in depth. Modeling these customer journeys to automatically classify which customer is on a happy vs. unhappy experience path can enable companies to proactively address friction points and thereby improve the overall engagement and as a result loyalty. We show how some of the state of the art deep learning based approaches can applied for this modeling and demonstrate the business impact when implemented in real world settings across industries

Share

LED lights on Tropics 1 billion dollar annual revenue company officers harass Puerto Rican tropical and focus on natural language processing. And in addition to that. I will see you later looks on the frontier and emerging topics and if you're interested. Contact details. All right there. Show me back. Let's jump into the topic for today, which is around to the course of a customer journey and witches and the customer has the Delirious channels and various promotional offer

letter that they are bombarded with desus and you can see I am here if you take Telecom each of these three customers in the context of a planning middleby some contractions before international travel. Black and the elements of the atom customer call center complaints on someone to be able to maximize the customer experience and make the experience. I just said I didn't switch data captured for customers along the journey to the interaction which is good and is desirable for us to be able to better understand the journey, but

at the same time it posed challenges because there is so much information about in terms of the volume as the last station here. Each of these is an interaction. The customer has is that an individual complexity but each of these contractions that will be quick to stream data that we have electric track where we can see which pages to customers going to did they do it first. Is an all these are valuable information intent of the customer and we can make a call and so on

ivr level two. How much time will it take for each of those levels are in the end of the managed to talk to a agent in the transcript of the ages of the conversation can have lots of critical information back to Summit which information but at the same time this introduces complexities in terms of the competition cost that is necessary to property as well. sophistication that is necessary and if you look at the conventional approaches that are used to understand the customer Journey one approaches the quality

there be typically have to figure out what we are looking for a plant and ask those questions and even then there will be limitations like the number customer service contact information highly into credible and you can directly answer questions. Like why do you make such a decision? So that's where you come up with various exploration data mining. address of the complexity to some extent but this also doesn't capture be Central aspects and they contact you last week and the longitude

and emerging methods of each other more and more increasingly and if it's been to be Black Box models, so we do have some leads to I'm back to Iran flight limitations. So if you look at the framework of Johnny Marlon LeFevre and anything you want to play something then we are clear with what they want to collect. So it's a different outcome and education are in for my tactical execution sometimes real time and sometimes so I can only support customer. The other one is which

Mad they want to look at the macro picture and try to understand what are the different patterns and then if there are some new emerging pathogens from that then if they have to design a new kind of 153 like the new kind of customer segments that kind of attention attention around the key is under customers. Like how do I get the likelihood of acquiring if I had a customer? What is the likelihood that a customer can buy? I got a very important question. I

have these 50 promotional offers, which of them is the best one that discussed Millennials with all these are prediction based questions that I know the intervention and just trying to figure out who will respond best to attend trying to Target that are engineering manually or alternatively we could use deep learning approaches like using long short term memory. Extract to treat customer Journey as a text and apply language models like Transformers and we have applied that

is more of an understanding and object to this request. Nothing bored these cases. We can actually augment them with the family presentation learning techniques. Leica autoencoders are dino and more recently or something like a process where you feed the sequence of events. And then it'll actually see what is the intensity of contribution and based on that. You can actually identify broadly diesel vehicle impoundment tactical execution and then identify the patterns that are happening at the micro-level to inform strategy.

What we have done is like I said, maybe can have the truck running and on top of it can be both architecture that can support both collection Thunder and Buttons for each of the events. I'll show you how to look like a separation between them. That'll eventually predict. What is the next day and then do the first time that will inform the correct? This kind of a unified approach will actually interviews tremendous flexibility accident car flipped in customer experience here trying to predict who is likely to call the call center and

let's 3 a.m. To call if you're able to predict the customer call center agent in a timely manner. So we are. Call operation as well as the customer experience and how does it look like you can see that you can't dad will be sequence of these and those sequences of the fact that we have one customer has done mobile search then have to call them. They actually traveled after they have to call. This is an unhappy customer. The customer did searched and searched again and mobile and the international travel.

I just so how can we introduce once we built our model what we have seen is the weight can be implemented is as the customers interact. You can calculate the expected satisfaction which in this case can be measured as MPS. Okay, I'm thinking questions. But I imagine we'll take those questions at the end so high LTV customer in a certain age group and using iPhone so they're happy. Significant be happy so you can see it, but then I have two sequences happen.

Okay. Yeah, so the customer in this case has done some research and not able to find the information. You actually have to chat with the agent and use the happiness a little bit and then eventually have to call but then the call of the car dealers on 3rd in the car because of the very smooth experience so they came back to their original state and then the travel center and it was not resolved in one go there to call multiple times and ask Dad to call again and again

because we have the model the model can actually predict what is the likely satisfaction level of the customer and then we could have based on this model intervened. Right at the time because now we know that the customer is likely to call next so you have to intervene right away and the customer is on the web website and not getting the questions answered. So either you can from the chat in the window, or at least you can't avoid that cause call you can have actually a call center agent

with the necessary information about the travel agent, San Juan. So that's the predictions set of books at the same time. Like I said, the pleasant side effect of this is the information that we have in this process because as you do this we have to project the high dimensional data in this case will have lots of events like calling a general called to call about this. Is it a message in the app, or is it the blue it again will have some types within them based on the content so we can actually now see that all of these are very nicely clustered. So

all of these behaviors that captured as they have similar How the customer's preferences on. What is the sequence of choices of bee stings for a given customer and based on that. So they can better anticipate Water customer looking for? No, we have seen the telephone example. If you look at the Patients go to and this will involve going to this is the Snapchat for a single patient related interactions are various conditions for patient is diagnosed with and they could be

that as well. So much Dimension. So one of the object is to see which of these are really important for me so I can and if different patients that's falling in the following different Pathways possibly because their profile under the conditions and the managing doctors different Etc. So based on that we can actually formulate strategies on how to reach each of these patients separately how to get them on the right path that will maximize the best outcomes for the patient. So the first step is like it's a

process that figures. Which of these interactions Are actually having a different intensity for the events 11th of interest. In this case, then I'll come back hospitalization and then you can only text citation of each of the event has to happen on the 5 months and we'll see which events led to the temporal intensity spiking and only those events can be considered for the Illinois and then we can train Auto and colors on the remaining data and then we can actually get some clustering that I'll go to look something like this can be an intimidating check. But if you

see each of these columns here to talk to patients that they identified and each of these images here each line here is a patient. Write a column here is the cluster of patients and you can see visually that they're all following one kind of talking to the first one is they are stable and one that bad one is one type of medication the second one. Is there somewhat stable on brand two, but the third cluster seems to be switching quite a bit and there is no specific

brand for front and back and d e is stable on Grand to and if they belong Brian 3, but eventually switching some of them and then others and what are the different parts that are that they are taking and what can I do? Sophie said that switching to any brand regardless then my approach to them has to be different from if they're switching from my bank to someone else and then my put student-athletes. That's the idea that you even on the healthcare examples of how early can formula.

These are essentially applicable to any industry as long as the data around customer Journey along with the message that we have to make reservations and some of the impact of chocolate that shown that a sequence is emerging we can try to figure out what is going on exactly. What combinations of Dynamics are actually driving the output not just important in some industrious like healthcare and financial services. We're even if it is not sufficient. Man, and the predictions are not driven by any

information. They are actually we have to do some actual testing as with Lisa's to make sure there are no buyers on the frontier is so far all the methods of you have seen they're pretty good handling great accuracy and doing the prediction according to call the Dynamics of the different. So there is a lot of money so that can I infer the causal structure automatically based on the data and then eventually or I can be sure that I'm not getting any. Collisions are attributed to the outcome of interest and

if I do find it. So loser be Frontiers to Sammy. I did it which data which is good. But it also poses challenges that support a unified solution architecture that can handle both collection and Discovery and we have seen multiple actually have been instrumental in introducing. I think we have.

Cackle comments for the website

Buy this talk

Access to the talk “Improving customer experience through journey...By Srinivas Chilukuri, Data Scientist, ZS Associates”
Available
In cart
Free
Free
Free
Free
Free
Free

Ticket

Get access to all videos “Global Artificial Intelligence Virtual Conference”
Available
In cart
Free
Free
Free
Free
Free
Free
Ticket

Buy this video

Video

Access to the talk “Improving customer experience through journey...By Srinivas Chilukuri, Data Scientist, ZS Associates”
Available
In cart
Free
Free
Free
Free
Free
Free

Conference Cast

With ConferenceCast.tv, you get access to our library of the world's best conference talks.

Conference Cast
561 conferences
22100 speakers
8257 hours of content