Amazon Connect Delivers Personalized Customer Experience for Your Contact Center

hello welcome to this session I appreciate this the last session of the day and we're all that stands between you and the networking drinks but well done for making it this far and we've got a really great session lined up to help you understand how Amazon Connect can deliver personalized customer experiences for your contact center so my name's Wes Neary I'm a Solutions Architect in the UK and Ireland public sector team and I'm going to give a quick introduction to Amazon Connect then I'll be handing over to my fellow sa guy who's going to show you a really cool live demo there's saying in tech about never doing a live demo but we're going to and then we'll hand over to Pete Clarke from CAFTA who's going to give you his story and their organization's story about their experiences with Amazon Connect so it's important and we always work backwards Amazon so it's always good to start with what are some of the common challenges with traditional contact centers and understand this will help you understand where Amazon Connect is different so first of all lots and lots of contact centers are really struggling to deal with rising levels of demand whether that's through physical location closures or just more and more people demographics changing more more people being comfortable dealing with that way than in-person transactions upcoming refreshes again hardware has a shelf life in traditional contact centers some of that hardware can be very expensive and we're getting lots of customers telling us they've got some fairly major and costly refreshes coming up unsuitable licensing models so contact by its very nature is a very piquant Rafi workload yet traditional contact centers licence it on either a per agent or a per line basis and what that means in reality is that a lot of the time you're wasting money on capacity that you don't need to deal with that people at work when you do need it traditional contact centers are quite often tight physical locations whether that's through physical telephony cabling which again is fine if you've got a stagnant workload but contact is in that way and so if you want to scale that you have to build bigger facilities lack of interfaces and integrations again traditional contact centers aren't the most open things in the world and it can make it really difficult fully integrate that with your on-premise systems and slow development cycles so again traditional contact centers generally on a six monthly a yearly release which means you've got to wait that time before you can deliver changes or value to your customer so at Amazon we strive to be Earth's most customer centric company and to do this we obviously support millions of customers in dozens of languages across 32 countries and at peak times we can have over 70,000 customer service agents supporting Amazon so this is quite a large scale and quite a large challenge and being honest we looked around and we couldn't really find a contact center solution that could meet this scale so in true Amazon fashion we built one so Amazon Connect is the contact center solution that was originally designed born to enable Amazon to deliver an exceptional level of customer service it includes all the standard functionality you would expect in a traditional contact center so skills-based routing allows you to maximize the efficiency of agents calls are recorded in high quality making easier to improve and monitor agent performance you can get real-time and historical analytics on the platform and you get high quality voice capability that comes as standard so all of that most traditional contact centers can do that so let's look at some of the differentiators for Amazon Connect so firstly setting up Amazon Connect is easy with a few clicks in a console your agents can begin taking calls it is possible to do it within about five minutes I won't say you'll have a fully functioning contact center in five minutes but you will have the ability to take phone call and pass it to an agent the other thing about Connect is it can be managed by business users your contact center supervisors you don't need people with highly technical highly specialized skills to manage it using natural contact flows you can drag and drop yet elements within the editor which allows you to really change as your customers change we're not talking months to change IVR flows anymore Amazon Connect provides open interfaces that makes it easy and straightforward to interface with your back-end systems third-party products and AWS services this allows you to provide a richer depth of capabilities and ultimately a better customer service and Amazon Connect can leverage the full ecosystem of AWS technology and consultancy partners to help you deliver this so will now just do a quick example of what one of these natural contact flows might look like hopefully the audio works hi Nikki wolf I see your flight was canceled today how can I help you can you please rebook me for the same flight tomorrow okay you are now booked for a 9:00 a.m. departure tomorrow out of San Francisco arriving in Seattle at 11:45 a.m. great thank you so what you saw there is a potentially normal customer interaction but actually there's no agent involved in that and a resolution has been found to the customer the less unpack how that actually worked and what actually happened there so we had Amazon connect there and we had a contact flow that was handling that communication we then used AWS lambda as a flexible to interface with our CRM that allowed us to use the calling line ID to identify the customer from there we were able to look into our flight booking database and understand why that customer may be calling have they got a flight that's canceled have they got an upcoming flight and we could also use the same flow to interface with our business intelligence tools so that's fine for an automated response but sometimes an automated response or your default patterns won't solve so what we look to do here with Connect is really provide a consistent user experience so our same customer nikki has run back and said well since I was inconvenienced can I get an upgrade for my flight the Lexx bot that's powering that chat engine hasn't been able to deal with that so it's passing it over to an agent but what's important is when we're doing that the agent is actually given the history and the information there this means they're not looking into another system to understand what's happened with that call it's a natural flow and a consistent customer experience so agent experience is also vital and often within contact centers the really lost thing that people often forget about it's just as important to give your agents a really good experience because this ultimately is what empowers them and enables them to deliver a good customer experience with Amazon Connect it's simple to integrate the contact control panel you see up on the right-hand side of the screen there into other applications whether that's your CRM or whether it's a web portal that agents use so as an example here on the left is a screen shot of the customer control panel using the Salesforce pre-canned CTI adapter this allows the control panel which is where the agents make and receive calls from to be embedded into a single window meaning they have single pane access to all the relevant data that they require for the call if your Salesforce user within your organization's already the adapters available it's and it's available for free and it's constantly being updated as new integrations and new features are released this goes down really well with agents the fact they can do everything they need to do in one window is a real game-changer so we talked about an open platform and easy integrations and there's a number of elements that do that but again AWS lambda are event-driven compute service is really the glue that allows you to stitch them together so for example we can take your call recordings are recorded into an s3 bucket this highly available highly durable storage means that actually you don't need to worry about scaling and replicating them call records lifecycle management allow you to easily manage them and delete them as per your regulatory requirements when they're no longer needed but also allows you to start using the breath of other AWS services to actually start analyzing that data the agent data itself can be available through the connect console but can also easily be passed into workforce management systems to allow you to optimize your workforce call metrics again are stored but again these can easily be passed into your data warehouse whether that's an AWS service or an on-premise existing data warehouse so just to sort of really talk about this and this is applicable to all AWS services when you consume them you're not just getting a service you're getting the power of the AWS ecosystem so at Amazon we have a saying that security is observe so you can use our Identity and Access Management Service to control access to the administration of the platform you can use the AWS directory service or any sam'l compatible IDP to authenticate users and agents to the platform as we've said you store call recordings in s3 if you have a long-term compliance requirement to keep them you can make use of cheaper storage through s3 Glacia you can use lambda step functions or API gateway to provide custom integrations with other systems you can make use of our database platforms to store that data as it comes out whether that's the agent contact data or call metric data you can use cloud watch and cloud trail to monitor the performance and provide audits of activity if you have a regulatory requirement that's how you can achieve it you can start to use some of them all our messaging services to allow you to interface with other applications if there isn't a native integration and because we're storing most of our call date you're in a in s3 we can then start to leverage some of our democratized AI and m/l services whether that's Lex poly transcribed comprehend and you can perform some real big data analytics once you collect that dataset using our analytic services so just to pick out on one of them we talked about improving contact centers with artificial intelligence when guy comes up on stage you'll see us using a call into connect will then be using Lex our natural language understanding service to build a chat bot to handle that call once we've recorded the call we'll pass it in to Amazon transcribe which will allow us to create a near real-time transcription of the text which would be pretty awesome I think a lot of customers we speak to already would be like if we can just save the agents typing there that's a huge saving in itself but we can take it further we can then you start to use services like an Amazon comprehend to do sentiment analysis on that transcript and then as guy will show you you can apply machine learning and analytics on top to really start maximizing the value of that data so it's no longer just data you're holding it's now useful and providing a business value to you so just quickly to wrap up for this section we talked about the entire e AWS ecosystem so there's an ever growing number of connect technology partners providing integrations into the platform you see there we talked about Salesforce but ver in a nice have integrations that are available and if you need some help with this there's again an Evan growing number of connect consulting partners that can provide the skills and resources you require to help you on your journey with their so actually pictures and demos paint a thousand words and no one really wants to listen to me talking anymore so I take this opportunity to welcome guy to the stage who's going to take you through a live demo of Kinect okay sound is good can you hear me perfect great thank you very much for being here my name is Gabe in the walk I'm a solution architect in AWS I work with customers from the media entertainment and telecom verticals so basically today what we're going to do we're going to do a live demo so we need to pray for the gods of demos that it will work and it will work so basically what we did we we had a thought about how we can actually integrate AI and machine learning services in order to expand the capabilities of the Amazon connect and to show you a little bit of the different integration capabilities and how we can actually enhance the experience of your customers and how we can actually enable your agents so what we actually did here the first way thing which I'm going to demonstrate is basically real-time speech analytics which automatically grades every call so you're gonna see a life course that we're going to do a I have here my colleague in the room is going to do a couple of calls and you're going to see the dashboard that we've built the UI we're going to see how the agent can actually see in real time there's an understanding of the sentiment analysis if either if it's a positive sentiment if it's a negative if it's a natural one and you can also see how we can actually accumulate that so the agent can understand in real time whether there is any risk during the call either it's a very negative code and it can maybe transition the call into a an expert q for example to look at that call so that what this is part of the things which we are going to cover today the second part basically is a instant indication as I mentioned before so we have the customer sentiment and we have the accumulation we're going to go through it during the demo the last part is basically we're going to leverage the customer data to personalize the customer experience so basically what we are going to do in the demo we are going to route the call so my colleague will make a phone call to the Amazon connect a call center and we are going to determine according to trend prediction so we have built in the background a model using Amazon sage maker to better understand if the customer that is calling is going to likely a whether to change it to turn or not and so we actually trained a model in the background we used around 7,000 different customers so we based it upon how many calls the customer has done how many minutes overnight does he have any voicemail plan international plan so we took all those parameters and we actually trained the model with sage maker and we expose an endpoint so what you can actually see is that we connect route the calls before we the agent even picks up the phone whether it to an expert queue if we think that you know the model predicts that the customer is likely to churn over a regular standard agent queue if looks like the customer is not going to churn so that's just an example of showing you how we can actually use machine learning in the background right so before I go to the solution architecture because it's better to show it after the demo let's move on to the demo part perfect so just a couple of sentence around that so this is basically the UI the dashboard of the agent this is what we've built you can see here on the left side we have the Amazon connect so that's like a standard panel once there will be a incoming call the agent can pick up the call you have also the path for the call transcript this is where you can see live transcription so this is basically using I will show two I will show it afterwards also in the architecture this is the live transcription of the customer calling in and then you can see the Kohen sentiment you can see here the god turning between red and green this is according to the sentiment analysis whether it's a negative sentiment where there is a positive of a natural one and on the right side you can see the coal attributes so we store the the data from the customer as I mentioned before either if it's voicemail plan international call how many times actually the customer has called previously to a call center so we can take this data and present it to the agent as well as we can present the churn decision to the agent so exactly immediately when the agent is actually picking up the call you can immediately get this data to his dashboard including whether this customer is about to churn or not so that's basically the UI so what we're going to do right now we're going to make a call so my colleague is going to call the call center so it's a similar thing the customer is calling the call center he has an issue a specific issue whether it's Wi-Fi or phone or anything like that and I'm gonna pick up the call as the agent and we're gonna just do a basic short quick discussion so I want you to see how it actually looks yeah so once the incoming call will reach to the Amazon connect you can see here actually the inbound call and I'm just going to pick up the call you can see it's connecting I'm just going to put myself on mute so I just muted myself and then actually we can actually start the discussion perfect good another problem with my bye-bye I can't connect any okay we can definitely try and help you with that okay which kind of device do you have can you please restart my router yes what I'm going to do right now I'm going to restart the router let me low and let me know if that helps thank you very much that is great yeah we started it perfect glad we could help you that was very helpful it's great thank you very much thank you very much have a nice day perfect right so that was actually just a short call for you to see how it's actually working so again we have the live transcription where you can see the call transcript that's actually using the Amazon transcribe a real time API and you can see the you can see the actual sentence and you can also see the idea if it's a what is the level of confidence either if it was positive so of course thank you very much this is great it's very high 99% confidence if it's positive and what we did on the right side you can actually see the accumulated sentiment right so we actually accumulate different sentiments and we aggregate that so then the agent can understand you know what is the overall sentiments during the discussion and then it can make decisions whether you know you move to the right side and then you can pass it to an expert or supervisor and so forth what is interesting that on the right side you can actually see the call attributes those are actually coming from the dynamodb so this is the data coming in for that specific customer so you can see different parameters like international plan or a voicemail plan but basically you can use whatever data your customer is providing so for example we have customers that we work with them on the quality of Wi-Fi in their in their house so you can have like Wi-Fi signaling quality of service or whatever data you would like to use and you can also see here the pout for the current prediction not likely so that's exactly what happened once the call went through it actually we had a lambda function which got the data from dynamodb and send it over to a sage makeer endpoint and this is actually the results which is not likely so that's the reason why the call actually been routed to a standard agent what we're going to do right now we are basically going to do a call and my colleague is going to call now to a from a different number so this number basically we have data around this number that is actually a it is likely to churn so we're going to see how it actually looks in the background the lambda function will inquire with the data coming from dynamodb the sage make your endpoint and we will see how it actually routes directly to an expert queue so here is my inbound call as an angel I'm just going to pick up the call expert here you can see the call is connecting I'm gonna put myself on mute and here is the call first of all you can see that on the right side if you look on the call attributes you can see churn likely this is coming from the same to make your endpoint own I cannot make any outbound calls okay let's see if we can help you with that you please check my account yes let me take a look in your account thank you very much perfect that is great I'm very happy for that yes I said there is an issue there and I would be happy to help you have a nice day perfect right so again you see you saw the real-time sentiment analysis and the gotcha turning with the accumulated score and again we have two queues one is the regular agent the other one is the expert and we just immediately routed before the call actually started to the expert queue because we suspected that this customer is actually going to churn it called according to the model we have in our sage maker amazon sage make your endpoint perfect so let's look a little bit on the architecture so on the left side you can see the customer actually doing the inbound call which goes all the way to the Amazon connect instance on AWS so the first part is actually the authentication part so the lambda function goes and authenticate the customer it goes into an Amazon DynamoDB verifies and authenticate the customer and get back the data right so we mentioned different data international plan voice volts meal plan how many calls nightly calls and so forth the data comes back into a lambda function and the lambda function actually sent this data over with the you know an API call to the churn prediction endpoint which we've built with Amazon sage maker an Amazon sage maker bring back bring back the result either if it's like likely to turn or not so when lambda a lambda function bring back the result to the Amazon Connect then you can decide if how you want to out the call to which queue and that's actually what we are doing period to the to the second the agent is actually picking up the phone in those different queues the next part is basically a the customer audio streaming which goes out from Amazon connect through a kinesics video stream so we actually stream the audio which is coming from the Amazon connect instance all the way to a lambda function again and then lambda function does two things the first part which you saw on the screen as well is the real-time API so we do the transcription the live transcription that you have saw if you've seen and the second part is basically the sentiment analysis you mizzen using Amazon comprehend right so both of them are being done via a lambda function and those are actually API services so we just trigger the API and we're getting the result and we visualize them and that what you can see here and we've actually built an agent web application so we have the iaws up saying so it's in real time updates the dashboard so when you have a real real time sentiment coming over and updates in actually updating the UI you have the Amazon DynamoDB lambda function we use kognito Amazon kognito to authenticate the agents were actually logging in into our a call center so though we are doing it with the kognito and in the end of the day you have the agent console and then you have the contact center agents so just in a nutshell about what we're seeing next and what we are seeing in the industry and we work with customers so then you can take it to for example predictive maintenance so you can have your using machine learning in the background to understand better about your customers issue maybe predict even issues that your customers have or will have and you can actively contact your customers and give them a call and tell them hey where you might have an issue next week maybe with your Wi-Fi or we suspect that you might have an issue upcoming issue with your cell phone so we might help you with that personalization right so we can route customers to VIP cues we can give them a personalized experience about their issues as we already know you know from the history different issues that they face so we can try and help them with that and of course we can also look on virtualize virtual assistants like Amazon like so essentially we can have rather than a real agent we can have a virtualized a virtual assistant and we can have the same integration with machine learning in ni AI in the background and this virtual assistant can actually leverage that integration of the ia IML stack right so thank you for that a I'm gonna pass on the mic to a Peter clock from a capital customer management is going to talk about their case study thank you hello everybody I'm very conscious on between you and the beer so I will not take too long so I'm Peter Clark I work for capita customer management where the element and part of captive runs the contact centers we run 20,000 agents across Europe India and South Africa we have about a hundred million conversations with our customers annually and we work across multiple sectors financial services utilities automotive and retail we have large clients o - probably one our biggest ones British Gas Volkswagen LC National Trust and and Deutsche Telekom so we have a diverse array of clients and we have multiple sectors across multiple multiple geographies so why so why did we look at connect so a current landscape as probably many of you in this in this room we have lots telephony we have where you client telephony in some cases and we have our own telephony we have a large and Avaya state multi tenant 'add on-premise Hardware as was already mentioned by wares and guy it's on a per seat licensing model there's been quite a lot of that front investment to get at this point and there's also continuous investment maintaining such like PCI and upgrades so that is driving high set up costs probably comparable to other businesses and other sectors but it is it is inherently an a high capex it's very voice centric of civ I got some elements in actual language you've got IVR but it is quite limited to the voice voice channel there are you can do home working but again it there's elements that's limited and there's a considerable amount of ongoing support that's not just there's ongoing support to contain the the platform and infrastructure but also all the upgrades are necessary to keep a platform like that on going so we started speaking to Amazon last year Kinect was actually released in in March 2017 but we started to look at it very seriously last year and we decided that the probably the the best thing to do was to pilot it was to look at a client relatively small client but look at a client that was self contained in one one office that we could look at migrating from our Avaya platform in interconnect so not dissimilar to what you've oh right not dissimilar to what you've already seen what we needed to do was we couldn't just move to connect we needed to put connect into our current landscape so that meant we had to integrate it into lots of existing systems so as it's already been discussed agents now connect via a soft phone on the control panel to contact control panel we do record all as already being discuss we record all our calls they go into s free and we have made starts using transcribe and and comprehend and as has already discussed transcribe will turn your voice to text and comprehendible then to sentiment analysis and reason for call I mean one observation I would make on that and it was brought up on another session is you do need to train transcribe and comprehend you need to give it the context of the calls it has a vocab ree but you need to extend that vocabulary to the nature of of the course we have a workforce management platform that aspect it's not the latest version so there is a quick start for aspect but we couldn't use that so what we decided to do was feed plan data from aspect into connect for a Kinesis stream into a dynamo DB we then compare that plan data with actual that comes out of the Kinect that's the Kinect race cream and the agents dream and we then demonstrate that to the team leaders through a real-time at Darren's screen so we've built our own real-time endurance screen using cloud front in AWS and that provides our team leaders and our managers real-time information about who is taking calls who should be in who isn't in who's on a break that shouldn't be on a break such like as already been discussed we've also plugged connecting to our business objects data warehouse we day to day we use the Kinect dashboard to give us the flexibility and the ability to move our agents around and move cool flows but we also push that data into business object so we can do monthly monthly reports we use Kinect 2 for authorization so the Kinect I am it enables us to define agents team leaders and suchlike but we actually get the users to log in via ad so they use a standard username and standard passwords to log in to ad and that then authorizes them to use the correct facilities one really good feature amongst many of the Kinect is the ability you've got cloud watch and cloud formation so cloud what we've been able to do is we've built a test environment and a production environment which is we didn't have one of them before we had production we didn't have tests so we now have a test environment and a production environment and we use cloud formation to move the tests to the production environment you can't move all the Kinect call flows yet but you can extract them and then move them into into production that's been very powerful to be because as we'll start to move into the AI area into the app space and into the self-service space having a test environment is going to be critical to being able to develop those new functionalities and we've also take payments so we have a pci gateway so we push calls to that pci gateway but what you can also do if you could take calls in the connect IVR so you can push a call into the IVR from the agent you can then take the credit card number the expiry date the CSV into the IVR that gets encrypted and your name can pass that to a payment gateway so you can do it in the connect API you can do it in the connect IVR or you can do it in an external provider what we're then doing so that's that what that's done is build us a contact center solution from so on the right the capital contact solution this is now a repeatable architecture but we are now live with we have run in an account taking calls on it and we can now move that repeatable infrastructure to multiple clients and we can move that because it's all in the cloud all via software as a service we can now move that across from multiple accounts what we're now looking at is the more interesting part so this is what we're working with our client we're now looking at the more interesting aspects of integrating into our clients CRM into their business objects into their business warehouse and also to drive self-service so this is where connect really starts to deliver the power and and the flexibility and using lambda and other capabilities we are now you drive in interactive Ivy ours and we can now drive self-service so this is really where connect starts delivering its power and this is really where you've got the access to the wider ecosystem of AWS connect at the end of the day it's great it allows you to replace a legacy telephony platform with another cloud-based telephony platform but it's the connectivity to the whole of the AWS ecosystem that gives you the power so what what benefits have have we have we delivered my screens got enough unfortunately but there we go so we're delivering this into the kingfishers group for our trade UK account we did as it said first we did a short pilot so we got the pilot up and running in six weeks now it could have been quicker at six weeks was more down to operational timescales and our ability to move rather than than connect we then spent twenty weeks building the architecture I've just gone through so that that did take a bit of time to do it but we've now got a repeatable an architecture what we are observing if you if you compare like for like we are expected we are observing about a 30% reduction in in OPEX now that is because connects is per minute per call that will obviously be dependent on how many calls and how many minutes you use but we are we are observing a 30 percent improvement now picking up on a point that's already been made it's quite key for our agents feel better for use and this not just our customers and we are getting a better performance from our agents and and user experience and what we are also being able to do is customize reporting so in connect in the dashboard you can customize reporting for what you want you can have standard reports you can customize reports you can report on things like tentacle answers in 20 seconds 60 seconds and so forth and that is all through a browser browser interface right so what what are we doing looking looking forward and I think actually the demo that we've just had really really highlights that where we are now looking we've now got a repeatable architecture that we can deploy into our into our business and so we are now looking at moving this across other clients we have another implementation within our government services business that will go live in May and also our local government services our business are also using the Lexx capability to drive self-service on front of our Avaya so where we can't move to connect per se we're still using Lex and the self-service capabilities that delivers in front of our vial at forum we are also looking at some of more the aspects that we've just seen around route real-time voice analytics I think being able to help the agent through a call with next best action would be really really powerful and also about personalizing the experience for the for the customer so we're very much looking at voice bots are we working on some voice bots so this isn't just IVR as an ID and V this is true self service capability within a voice bot that enables the customers to self-serve but importantly if they can't self-serve or need to talk to an advisor we will then hand that customer over to advisor but the advisor will have the whole context of the conversation that you've had today it's absolutely key that if we're going to hand a customer over to advisor that's being in a voice bot that they understand the context of that conversation and that's where we can use comprehend and transcribe to make sure that we bring out the important elements of what's being discussed we're also working with a partner mission labs with their smart agent tool to look at chat BOTS so not only can we then manage voice within the connect with connect platform we can then start having chat bots on mobile devices Facebook Messenger and the website and that can then be automated through self-service or pushed to an advisor as required so what what's happened is by spending a bit of effort and a bit of work a few months worth of effort we've now got a repeatable platform on and which we can grow on and hopefully deliver a lot more personalization for customers and a better experience so I think that's it thank you thanks Peter thanks that's a really really good story and to all of you I just want to say thank you I hope you had an awesome day please do complete the session survey it may not be live for a few minutes till after we've finished but please do complete that the reason is that's what allows us to see what's worked what hasn't what's valuable what's not iterate and make this better for the future but thank you again [Applause]

Channel: Amazon Web Services

Views: 2,523

Keywords: AWS, Amazon Web Services, Cloud, cloud computing, AWS Cloud, AWS Summit

Id: SwbGkd53HN4

Length: 44min 19sec (2659 seconds)

Published: Mon Jun 03 2019