AWS re:Invent 2018: Customer-Centric Contact Centers w/ Amazon Connect & Machine Learning (FSV301)

good morning thank you very much for coming early morning for our session my name is Hannibal I'm a principal solution architect based out of New York my name is Ken I'm also a principal solution architect out of Charlotte North Carolina both animal and I are in our global financial services group and this is part of the financial services tract if you know anything about Amazon and odds are you do because you've taken the time and spent the money to come out here and spend the week with us you know we have these 14 leadership principles right and you might say well boy that's a lot of principles but we think to continue to innovate and meet the demands of our customers we have to be good at multiple things we don't get to choose whether we obsess over our customers or whether we deliver results for our shareholders we don't get to choose whether we think big and to be a lot of announcements this week the result of our fellow Amazonians thinking big dreaming big daring to color outside the lines but we also have to invent and simplify right we find our financial services customers deal with similar tensions especially when it comes to caring for customers so the point of our talk today is to talk about one of those main tensions the tensions on the one side of increasing regulatory pressure demonstrating to regulators around the world that your customers are well informed about the products that they are purchasing that they understand all the terms and conditions and then you can prove that you met that burden to regulators that when you're giving them advice that that advice is in their best interest and again that you can prove that and then the pressure seems to be increasing not decreasing the amount of detail on the other hand the customers themselves have increasing expectations right will it have an era where they expect self-service 24/7 365 in any channel from any device and that is is quite a challenge so assertion today is that you can do both of those things through the power of the NBS platform the automation it provides particularly machine learning services which we'll talk a lot about today and odds are you can do at a lower cost than you're doing today so we'll spend the rest of the time defending those assertions and hopefully giving you a vision through some demos that we've built on how you might do that so first I've asserted this increasing regulatory pressure so if you are a securities dealer in the US or the UK or the EU you're subject to explicit regulations to record those interactions again to be able to prove to regulators that you've you've given proper information you've given unbiased advice again here in the US if if you're a broker dealer you're subject to SEC 1784 and FINRA 3010 and others like it and you'll see they're in what I've highlighted that it's not only enough to be able to retain the records and easily access them as the SEC puts it but you have to have a regular process for reviewing compliance if in the EU if you're subject to the upcoming method to requirements the SMA says you know this this requirement extends to all sorts of Records and explicitly calls out to telecommunications and this is not just broker-dealers dodd-frank here in the u.s. CFTC and even some more basic corporate requirements like sarbanes and GL DEA require that records be preserved if that weren't hard enough heard now limits we've all just come through quite a bit of work to ensure gdpr compliance if we do business in the EU and one privacy expert and gdpr expert says that our standard disclosure you know when you're on a call that this call may be recorded that may be enough may not be enough for to count as consent from the customer and when the customer does indicate that I would like my records removed that extends to call recordings if you're in the payments card industry the the PCI DSS 3-2 which came out earlier this year went from saying hey you have to encrypt the private material to not store that sensitive information so it's even more complex but there are also the customers and data shows what common sense and experience have probably already talking if you work in the customer service business especially in financial services TTI global is a market research firm based in the UK and their most recent customer satisfaction benchmark report in the financial services industry showed that 25 percent of customers would switch banks just because they've been kept on hold right again that expectation is instantaneous so they've had some problem they haven't been on resolved through automated channels and they had to call 25% indicate they would switch just for being on hold and 32% if they get bounced around and have to answer the same questions you see there on the right Deloitte in look at what our customer is expecting overwhelmingly the top two things customers want when they have an interaction with you and the content center is that it be easy and that they get accurate and high-quality information but if you work in contact centres I'm not telling anything you don't know intuitively right well for years we've spent a lot of time and energy looking at things like first call resolution metrics - spending lots of money on agent training but I would submit we're not making as fast progress as we'd like we're not keeping up with customer expectations when someone takes the time to make a call they've they've had some problem they couldn't resolve in other ways and they're probably it's a more complex problem they're more frustrated and I would submit it we're not making the progress we would like for at least two reasons the first is that just the complexity of these systems you know the the state of the art for Quality Assurance again is the the djoko's this call may be recorded for Quality Assurance purposes so state of the art is humans review a random sample of calls so most calls don't get reviewed so insights in there about where the customer experience is breaking down just don't get I listen to and you know budgets are also shrinking even if you could find the talent in this labor market which would be a challenge you can't simply throw more humans to screen a higher percentage of the calls but the second is the system themselves you know these customer service sends customer service systems tend to be built out of separate solutions that are integrated with a lot of bespoke code right so you've got your AC D system your IVR system call recording analytics workforce management all kind of tied together so when you when you do find an insight through that random sampling and you want to deploy it that's an increasingly complex problem so our session today is that by using AWS and the power of the cloud and machine learning services you can sample all of the calls right and leave no insights on the table apply your human resources to go deep on the ones that are interesting and then very quickly because that's the thing that the cloud is pretty good at very quickly deploy those new features to address those gaps in customer service so let's take a look at a few of the services that can back up that assertion that you can build a new services and respond to customers quicker on AWS the first the most foundational is Amazon Lex so sometimes we refer to this in shorthand as the brains behind Alexa and if you've interacted with Alexa you know that it provides automated speech recognition right it knows what I said and they can figure out what I meant right Alexa's a fully managed service that you can use to get those ASR and NLP services it does have a full set of model building API so you can programmatically build them but we find most developers uses the rich integrated development environment that's in the console like many services the extensibility model in Lex is lambda it's not required but if you need to go deeper in that automated speech recognition and determining intent you can do that and then for fulfillment once you've determined what the customer wanted to do you have all the information you need to fulfill their intent you can use lambda optionally as the extensibility model to call your existing back-end systems Lex also has a really interesting deployment model you can have many versions of the same chat bot but iterate on them so not only can you have those aligned to different sdlc phases so dev and test and UAT production but you have multiple productions and then it has a notion of aliasing and so if you start to think what you can do with that all sorts of things that are it can be quite hard today like a bee or Bluegreen type testing paradigms hard in these these complex customer interaction systems we have today like integrating voice response become very straightforward let's take a look at one of those chat bots in this case we've used a SS mobile hub to build a simple iOS app imagine this is your iOS app and somewhere in it customers have an ability to say I would like to speak to an agent so that's what our users doing here today they're navigating through our app and in this case they're gonna interact via text they could have been actually speaking to it and it supports both so first thing the customer is asking is a balance so their intent is to check the balance and they've given me enough information right balance of what my savings account and so the chat bot can quickly return that information now the user is going to say okay now that I know how much money is there let's move some money around and you'll notice that this is true natural language processing the customer is not having to you know punch a button DTMF style on an IVR say push 3 if you want to transfer money they can just say what they want to do and interact so again their intent is to transfer money they've given me some data how much and from checking to savings I needed one more piece ask them when confirm that I've understood them correctly and then tell the user that we've moved the money and then the user again is going to go back to the check balance intent and you know just to be sure these chat bots are new they're gonna double check on us that now the balance is properly recorded so simple lex chat pot in that case it was exposed in an iOS application you can also expose in other places out of the box you'll see slack there on the Left Facebook Messenger on the right we also support Twilio you can embed this in your own HTML website as you see in in the middle if it supports if the platform supports cards so instead of asking the user having them type or say it you can display a card and have them just choose checking your savings from a list there you can also expose it and build a virtual concierge so this is Amazon Sumerian and this is Christine one of the built-in hosts in Sumerian and use the exact same chat bot if you I wanted a even more interactive more lifelike interaction you could do that hello there this is GFS chatbot how can we help you today you can ask for account balance fun transfer or request to speak to someone thank you the balance of your saving account is one thousand four hundred forty-four dollars okay so what's going on under the hood I've mentioned that Lex is doing at least two things right it's taking an audio file so so sound and then figure out okay what what words were in that sound right automated speech recognition but then I still have to determine what the user meant by that so that's the natural language processing engine so in in this flow I've determined the intent of the user is to transfer money oh and the user has also provided me some of the data I need to fulfill that intent the from account I then iterate through a state machine by prompting the user to fill up those other slots so again we have intents and slots they've already giving me the from account I ate through that and figure out where do you want to go how much and when once I figured that out I confirm that to the user thing I think you wanted to move 500 dollars from checking to savings today if yes I can fulfill that using lambda as the extensibility hook and calling the real system of record to do the debit and credit and we were interacting via voice so we need to take that text of the confirmation prompt and turn it into speech using Amazon Poly which we'll talk about in just a moment and then speak it back to the user so that's the flow under the hood what does the development experience look like I said earlier there was this rich IDE so first I built a create a chat bot chat bot can support a number of intense you know things that the customer wants to do or wants the chat about to do on their behalf and for each intent I have utterances that tell me you know give me a signal what what the user wants to do and you see they can be as simple as a one-word answer weave right or wrong trained users to just sort of bark one-word commands at our IVR systems but what we can also support Lanta neo natural sentence sao some user might say i want to put some money others might say i want to move some money and you give it a few utterances and then the pre-trained and continually retrained a natural language processing model behind Lex can then and for other things that the user might say that would declare that intent once I've determined and define the intent i define the slots and whether or not they're required you see in this transfer money we have those four slots we've been looking at but we could maybe make transfer date optional and if the user doesn't provide it we assume it's today and confirm that and go on we also have to then define those prompts that the chat bot says or text back to the user again so how do i how do i fill the slots and then how do we confirm the transaction and you see at the bottom of the fulfillment and so you don't have to use lambda it's it's a convenient extensibility hook it allows you to really get anywhere you might need to get and your existing systems are record but if you just want to use the lex api's and you call it'll hand you back a structured JSON document with the results of the conversation what was the intent what were the slots and then you can fill that any way you like by just parsing the JSON document okay next service to talk about unlike teenagers who you know when their parents call them on the phone you know with initiative oiss call and they push ignore and then text back what's up most humans you know if they're speaking they like to be spoken back to and that's where Polly comes in so you already saw a Christine using one of the Polly voices it's got dozens of voices and more than 20 languages and a pretty simple API right where you give it text and it gives you back an audio file it also supports markup so it's SS ml speech synthesis markup language that's how Christine's avatar or Sumerian Avatar was able to synchronize you know her mouth movements but also can give you hints on pronunciation and other things it's how you can have it pronounce things properly even if it's a foreign word like a French word but but she's speaking English that's how you can give it hints via markup so let's get a quick demo there in this case I'm using the COI and I'll call call synthesized speech was one of the poly api's I'm gonna pass it some disclosure text so imagine in this scenario you want to ensure that your context and Regents give the same proper disclosure again you want to be compliant you're telling them about terms and conditions where they're signing it for humans are pretty bad at repetitive tasks and so you might want to just pre record this and then have it played as part of your call flow every time you say I choose a voice ID I'm partial to Russell Russell which is a Australian male reminds me of my good friend Hannibal here tell it I want it in mp3 and it's gonna kick it back to me and then we'll have a listen to what that sounds like we must provide you with the following important rate fee and other cost information for the gfs platinum credit card for gfs platinum credit card accounts the annual percentage rate or APR for purchases ranges from thirteen point nine nine percent to twenty four point nine nine percent based on your credit worthiness and will vary with the market based on the prime rate there is no annual fee so just like you had all right so again throwing a lot of services at you but hopefully you're starting to get a sense of how you might put these things together in a pipeline and build a customer service application so the next is transcribe so it's function is if I have an audio file let's say it's the recording of an interaction between a customer service agent and your customer I want to turn that into text again pretty straightforward API until very recently this was an asynchronous API so I created a transcription job I passed in a document it could be up to two hours long so quite a long audio file and then I got notified that the job was done so was asynchronous just last week the transcribed team added support for real-time transcription so if you wanted to transcribe this very talk transcribed could take the audio stream and do it in real-time and it doesn't just give you back a blob of text and just just words it applies some intelligent formatting and punctuation and that's actually within a no surprise a structured JSON document that's how you can get other information like what's the confidence level that this word that what's my confidence level that I've transcribed this word correctly and then you can make some decisions on that or apply humans to improve the transcription where you had low confidence when did it occur so you have time stamping so you can synchronize a transcription to the video and thing you can also recognize multiple speakers we'll see that in the demo and very important so if you're trying to ensure compliance that your agents are saying the things they're meant to say and not saying the things they're not meant to say you want to separate out what the agents said and then apply some analytics on that and transcribe has support recognizing multiple speakers built-in you can also go all the way down to 8 kilohertz so think about a a mediocre cell phone quality call you get that a lot in contact centers transcribe can go down to that level and still have full fidelity on the transcription you also give transcribe hints again we work in the financial services industry there's lots of jargon product names other things you can give transcribe a custom vocabulary to hint given hints and make it improve its accuracy at transcription so let's have a quick look at how that might look like here I'm doing it in the console just for simplicity but new and as part of a pipeline you would normally do this with api's so I'm just going to create a new transcription job and then I mentioned the the previous API was all asynchronous so I first create a job just just a demo job for our session I'm gonna feed it actually a a call recording you'll you'll see two parties Hannibal our faithful customer in many of our demos and Emily our contact centre agent transcribe expects that to be an s3 bucket again no surprise if lots of ways lots of new ways already announced this week to get data into s3 it's in WAV and I want to give it a hint that hey this is the recording from a context Center so try to look for two speakers you would normally dump this out in your own bucket but for simplicity sake I'm just gonna use the default bucket so it'll turn up in the console so through the tried-and-true method from cooking shows my async API this took about five minutes in real time but I didn't want you to sit around while I hit refresh for five minutes so well it's done and then we'll go have a look at the results of the transcription so you'll see you'll see the text there in the console again with some punctuation and you can see this is an interaction between Hannibal and Emily but as I scroll down you'll see this is actually a structured JSON document I get other information that I can parse about this audio file I will see you know time stamping and other information and I want to flip over to the speaker identification sorry for the choppy video you'll see speaker zero is Emily our contact centre agent speaker one is Hannibal our customer and then I can apply analytics as we'll see in a moment when we get to comprehend on just the speaker zero channel I just want to make sure that my agent is compliant with various regulations okay again we're building up this pipeline we are going somewhere this is not just Andy Jessie style firehose of new services so I love Andy sorry it's comprehensive now I in my pipeline right I have an audio file I've transcribed it I have some data about it I'm able to separate out my contact center but now what was actually said how am I going to do some analytics and the assertion I made earlier was that sampling alone you know left a lot on the table how can we have the Machine listen to all of the calls and and pull out the ones where users are upset so comprehend will do sentiment analysis it'll pull out entities which think about them as proper names so product names or people or locations and things what are the key phrases again in this in this text which were the most important phrases so I can do some categorization comprehend it certainly I should have mentioned transcribe today supports 5 languages 3 flavors of English us British Australian a u.s. Spanish and Canadian fridge comprehend supports six languages and today and according to the fact more are coming so not an announcement I'm just repeating what's in fact let's take a look at how comprehend work so again that same transcription we've just transcribed that call I want to feed that speaker zero actually when speed feed Hannibal's text into and see you know was Hannibal happy or sad did we you know how did we do so again back to our speaker channel identification I'm gonna grab Hannibal's text and say hey was this customer pleased or not and I will feed it into transcribe and you'll get a sense of the data that it gives me back so again I just paste it in could it be could be an API call we're doing the console for simplicity sake you see Hannibal just confirming that he is his own authorized charge for five hundred dollars he wasn't gonna be charged for that so we put in we and we haven't you know a confidence score on what was actually those key phrases we've identified this as an English conversation with 99% confidence they actually have syntax so if you're into a really deep sentence diagramming style analysis of language you gonna pull out the verbs and do some analysis on that you know what what verbs correlate to the customers being upset what adjectives maybe is more fitting and you'll see that back to sentiment you know we have a sort 88% confidence that this is either neutral or positive this is probably not interaction I need to go deep on the customers seemed pretty pleased with it okay all right last one before I'll turn it over to Hannibal all my examples have been in English I'm not a polyglot so and we've mentioned that some of the services do support more languages but there's gonna be a lot of cases where you have data in one language you need to get it to another so Amazon Translate is a fully managed translation service that supports 21 languages today and sports 417 ordered pairs of from this language to that language and I'm sure in a room this size someone's really good at in fact or math and they're doing in their head they're good wait 21 languages ordered pairs that's 420 possible combinations and if you're really curious the three they're not supported today are going from Korean to Hebrew and going to or from simplified and traditional Chinese so there you go there's a trivia for the day and we're really proud of the work that we've done in translate and there's also this rich tradition at Amazon of humorous and informative customer reviews this is an actual review for an actual product on our our German site that's a real knife that you can buy and there was a customer left to review I mean German speakers could read it but I do not so you you fed it through an open-source translation engine as commonly available you'd get something like this again there's you can kind of figure out what's going on and you see that yeah he bought that my Forte's toothpick capabilities but it's not it's not great English you know you can see word order grammar capitalization other other errors if you feed the same same customer review through translate now you should get is still not perfect we still recommend that customers they don't have to attribute it to Amazon Translate but we do recommend you tell customers this was machine translated we're not there yet but you see this is this is much more interesting much more accurate language and you see that yes he's a quite happy customer with his fancy toothpick again I hope you've gotten a sense for you know that the these services that can give you insights and you can whatever language the data is in whether it comes in audio or text how you can derive insights and then use that to guide the humans you have to go deeper and find those gaps where the customer experience isn't where you want it to be and was gonna come up and talk about how you can then put that into action using Amazon Connect you know thanks again ok so before we go into what is Amazon Connect it's one of those platforms that a lot of the times it gets ignored when organizations are doing transformation and there are reasons behind that so I will actually go through why that happens but before we do that what is Amazon Connect it is a cloud-based context in a solution that can scale to meet any business needs it is based on the same technology that's powering amazon.com contact centers now Amazon has over 70,000 agents servicing their customers in multiple languages in over 30 countries when Amazon was looking for a platform to support their contact center they were looking for something that was going to be simple easy to integrate to transform also it had to be scalable and secure all the existing platforms that were available with very complex very difficult to make configuration changes too often require professionals to be making those changes now one of the other challenges that Amazon was facing was the scalability of these contact centers you had to provision the infrastructure you need to meet your maximum capacity now if you think about how Amazon operates days like prime Day and Cyber Monday just like yesterday we add or Amazon adds thousands and thousands of agents just to service those customers just for those two days now what happens if there were to pay for that well they had to pay for the maximum costed associated for those because the cost typically was based on number of agents or number of maximum capacity you will need so they were looking for something not only that they could scale it but also needed to be able to scale it down where they were not using it so what they did to provide that customer centric experience for their customers that they were willing to do they actually build their own platform and AWS has made that platform available for you guys as well it has skills based routing which makes you which allows you to service your customers more effectively it provides audio management on call recording it has real-time and historical metrics we can actually run analytics from your own contacts honor and all obviously it has high-quality voice capability because if your customers are called normally calling in they actually want something and you want to hear them and at the same time you want to hear them so these are the simple things that you will really need to build a contact center but I mentioned that transforming contacts and it is actually very very complex even the context and a normally starts with a IBR it gets really complex there's lots of other modules that get integrated and a lot of times these modules are our solutions are managed by different partners or solution providers and if you want to make a configuration change there's a lot of integration points here which if one of those breaks well what happens how do I actually make my configuration changes so that's what people are actually typically scared to make configuration changes because it costs them a lot of money to make the Changez so what is Kinect doing is replacing a lot of those modules so that way your integrations are a lot simpler easier to for you to make those changes and also you don't need to have those professionals to make those changes for you and at the same time it's open and allows you to integrated with other products we have lots of third-party products anti-fraud and so on that you can actually integrate it with very easily and also integrated with your own databases CRM and workforce management solutions what else is unique about Amazon Connect what it's 100% cloud-based which means you don't need to provision any infrastructure it's pay-as-you-go you know with no minimum commitment you pay for only number of minutes that the agents are servicing your customers it can scale up at the same time it can scale down and cost you nothing also the agents that are servicing those customers can log on over browser and from anywhere to access the our control panel to service the customers so there's no infrastructure needed a as well it has self-service configuration which means you can actually spin up a instance or make configuration changes to your contact center very easily you can actually spin up an instance of a context and a meeting minutes instead of days or months they used to take traditionally to build something so let's take a look at how you can actually provision an instance within few minutes so I'm gonna go to the AWS console and I'm gonna go to Amazon connect once I got it I'm gonna get started first thing I'm gonna give it an instance name for this purpose the demo you can have as many as you want you can also integrate it with your single sign-on I'm gonna skip all these and just go through and configure the configuration I'm gonna enable incoming calls and outgoing calls and the next I'm just gonna select where do I want my recording logs and matrix to be stored and when I select those I'll just create the instance once I click within few minutes it takes something about five minutes for the instance to start up and when you're in you're going to the console and you're ready to go now what we're gonna do is we're gonna go and claim a number I'm gonna pick us you can actually put your own number if you wanted to once you get that number you're pretty much ready to start receiving calls so when we actually create the number for you we're also linked it to a sample workflow so let's have a look at that what that floor looks like there are lots of them there we've selected a inbound flow and you can see different samples where you can actually choose from so let's take a look at what that sample flow looks like it's pretty simple it just answers the call and if you want you can actually put a person on the queue and the agent can pick it up so I've logged in as an agent now made myself available and I'm ready to take calls so that took me less than 10 minutes to actually spin up a brand new instance allows me to actually start taking calls and make configuration changes and build experience or one now whereas organizations are going through transformation financial services a lot of them are doing digital transformation they do a lot of changes on their online channels but contact centers are ignoring and not being touched but the customers are demanding that they want the similar experience as Tim mentioned they want they're getting used to that experience from a digital channels they don't want to be repeating themselves they don't want to be put on hold just to do a simple basic task so they want to be serviced as quickly as possible so the contact flow which is the engine of contra connect allows you to provide a dynamic personalised and natural experience to your customers so let's take a look at an example what happens when a customer who is frustrated that they call into a contact center and they want to be serviced as quickly as possible the first thing that happens as soon as the court comes in we're gonna tap into the data that we know about this customer hi Hannibal based on unusual activity we have locked your credit card as your call related to this if so please enter the one time code sent to your phone thank you your credit card is now unlocked and you should be able to proceed with the transaction great thank you so in this example we were able to be dynamic and answer the caller before I even they asked the question why because we actually tap into the data that we know about them the experience is personalized contact law allows you to adapt and provides a personalized experience even though we actually answer the order in an automated fashion it felt very natural because we were using Amazon Lex which is the same technology that's powering Alexa so in this example if frustrated customer that called in because their credit card was locked to do to a unusual activity they were able to get unlocked and move on without being put on hold without repeating themselves and invalidating so many different things and put potentially go to different agents to try to do that they were able to unlock themselves and move on and at the same time we were able to do this without actually engaging agent so we provided a positive experience to this customer even though they were frustrated when they call in into contacts or not now the platform is also very open which means you can actually build the experience that you want for your customers using the tools that you might already have now the center of all of that is the contact flow which is the processing engine behind connect now that integrates with our AWS lambda which is our event-driven surveillance platform which allows you to make calls to your own databases to your own customer databases to your own business intelligence so you can actually get information about the caller as soon as they call in so you get all the information you need before we start servicing the customer also the contact control panel which is what the agents are using to service these customers can be integrated into CRM it of your choice and one of those is Salesforce so let's have a look at what happens when we integrated the contact control panel which is simple as you go to the our Salesforce Store install it and your integrated into your Connect and when this agent is sitting into in Salesforce and they can actually receive that call and start servicing the customer without even leaving the screen so when the call comes in Salesforce right away pops up the contact why because we've connected into the CRM and we know who the caller is based on the caller ID so in this case the agent was able to service this customer without leaving the screen start servicing the customers right away and write the notes or whatever they need to do to a service that customer now from the back end of the system connect provides call recording which gets stored straight into your own s3 buckets which can be encrypted through kms which is a team management service which means that you don't need to have those expensive call recording solutions anymore the data that you need for compliance reasons to store them and archival you can easily do that and put them into your s3 or do a lifecycle management on them and move into Glacia and keep them for many ease as you need as well as that now you can actually tap into that data and do analytics and apply machine learning on it and get these information that you need from those call recordings a lot of the times as Ken mentioned we may be sample some of those calls well this allows you to to actually sample a lot of those calls not only just sample them you're actually process every single one of them and I'll show you that in a second so how do you do a call recording it's simple as when you created the instance you go in and enable call recording you choose which buckets which s3 bucket you want your call recordings to end up in and you also select how you want those to be encrypted you choose the KMS key that you want those encryption happens and as soon as the calls are done the recording are stored into your nursery bucket which are encrypted using TMS now contact center also provides you lots of metrics historically as well as a real-time so now you can actually forward all of those thread into your own data warehousing could be redshift running AWS or any other database you might have so that way now you can actually run your analytics generate your business reports as you do and which you typically was very difficult from a traditional contact centers because it was just hard to get metrics admin so now you can actually get those very easily and load them into your stream and into your data warehousing databases also because the platform is open allows you to load the agent data into your own workforce management tools so you can actually provide the operational efficiency for your contacts on up and lastly the way that integrates with AWS which is what makes the power of connects now all those services that can mention about we can easily integrate them we can actually build a true omni-channel experience for our customers so let's take a look at a demo that we built now we took lots of those services that can just mentioned and with integrator and into connect so let's take a listen and what happens when a caller calls into our contacts enough hi honey ball thank you for calling TFS please select from the following options press 1 to get your account balance trust you to make a transaction such as a transfer press 3 to report fraudulent activity or a loss card press 4 to speak to an agent thank you do you need to report a fraudulent transaction or a lost or stolen card yeah I lost my card with it your credit card or debit card we are canceling your credit card now and putting a hold on all transactions you will receive a new card in the mail within 5 business days were there any fraudulent transactions yes when did you lose your credit card last night I can see a transaction for $500 at Best Buy in Paramus New Jersey with this the fraudulent transaction yes thank you your credit card has been lost and we will reverse the $500 charge and the funds will be returned to your account within two to three business days press 1 to speak to an agent press 2 to go back to main menu please hold on we are transferring you to the next available agent thank you for calling your call is very important to us and will be answered in the order it was received so in this example this is what the call our experience look like so let's listen behind the scene when the agent is about to pick up that call that was sitting on that queue to service that customer so this is the contact control panel that we've integrated into our own version of CRM very basic version just for demo purposes just to show you how you can actually integrate these this could have been done in Salesforce as well very easily or any other CRM that you might have so let's take a listen on what happens when that agents about to answer that call as soon as the call comes in they know who the caller is and they get some additional information from the caller it is acute hi Hannibal this is Emily with GFS customer support I can see you just reported a lost credit card and we are refunding a fraudulent charge were there any other fraudulent activities that need to be reported I won't be tired for that $500 time that's correct we are reversing the charge and we'll be back in your account within a few business days and we have already ordered a replacement card for you okay thank you is there anything else I can help you with today all I need oh thank you thank you so as you can see in this example when the agents are picking up that call the strata-ray know who the caller is they can also see the context of the interaction the user was having with the chat bot because we can actually forward all of those back to this agent so that way they can start servicing them that are and not repeat the interaction that they had so that that way is personalized or so they're getting to them as quickly as possible so let's take a look at what we can do with back end of the system so when the core recordings are stored into your s3 what we're going to do is we're going to forward those audio core recordings from s3 into transcribe so we can get the transcription of the call once we have that we're going to store that into our s3 data like so that way we can use a light on as well then we're going to pass that text or the transcription into comprehend so we can start analyzing those calls we can actually get the sense of analysis we can pick up the keywords entities or any topics that we've created ourselves and we also push those up put back into the s3 against that caller and against that call ID so now we've actually put not only the actual audio call recording to s3 we've got the transcription and we've got the sentiment and the keywords that we picked up from that transcription so in this example we're actually going to stream it into a real-time dashboard so we can see as the calls are coming through what's the sentiment analysis on it looks like is a positive is a negative how many calls are we actually getting and this is you know you can go imagine having what we're all seeking create from the dashboards base you can see the number of speakers because that's what's coming from transcribe you can actually see the full transcription you can see what's the confident level and all that what are the key for key phrases what our customers actually calling us about now if you think about what can we do with this data well typically we actually couldn't process that much data before because all those audio recordings were stored in two tapes for archiving maybe some of them were you know we will listen to them but now we actually processing every single call recording as it comes through and we now we can generate and access that data to do interesting stuff below what we can train our agents how do they service the customers to maintain a positive experience for the customers also based on the keywords and the and the calls that we getting we can determine what are some of the top calls are about so that way we can figure out not only through our context and but also through other channels what do we need to do to enhance that experience for our customers so we can serve them a lot faster and quicker so now we actually got access to all this data figuring that what is the main calls are coming through and how we can maybe service them automatically so that way we don't have to put them on hold and and and you know give them a bad experience so we got access to all these data so one of the let's take a look at how Amazon Connect actually contextual designer works its drag-and-drop it's very easy in this example the first thing I'm gonna do I'm gonna turn on the recording for both agent and the customer then I'm gonna call some lambda function and I'm actually passing the caller ID into lambda so I can actually make calls back into my CRM database now I'm actually going to do some interesting stuff here am I going to call and to CRM and look for my a pilot flag based on that caller ID if I do have a pilot flag I will send them to a different flow if they don't belong to a pilot group I'm gonna send them to a existing flow you can have as many flows as you want it doesn't cost you anything you're only paying for number of minutes that the agents are servicing the customers then I'm actually going to prompt the user to enter some you know select the option and one of those is to interact with the chat bot to integrate it you select legs and you give it the name of the chat bot and the integration is done now the actual customer or the user when they're calling they're interacting with the chat bot now when they decide to speak to our agents we can put them on a queue by from what they've selected or what were what agents are available and again you can have multiple queues there and the way agents are sitting on those queues and they make themselves available they can actually pick up that call and start servicing the customer and the thing is he that the actual interaction that they had with the chat bot the context is being passed back in to the agent so they can actually see what the user was doing so it's very easy to make these configuration changes is drag-and-drop you can save these you can publish it you can roll it back you can export these and importer back into another instance if you want to so that way you can actually have full SDLC process devtest UAT production we can create multiple pilot groups multiple ad testings and so on it doesn't cost you anything because the only time you pain is for a number of minutes with the agents or servicing the customers again so I've talked about the piloting let's listen are at a at a call that comes through and the user doesn't belong to a poly group thank you for calling GFS press 1 if you have a technical issue press 2 to speak to a customer support so that caller didn't belong to Apollo Group I'm gonna go into my CRM update that flag and just called back the same number again hi honey ball thank you for calling GFS please select from the following options press 1 to get your account balance press Q to make a transaction such as a transfer press 3 to report fraudulent activity or a loss card press 4 to speak to an agent so it's very powerful to be able to do par learning to do a/b testing on your contact centers similar to what you do in your other channels now you can actually put people in different groups it were based on the caller ID based on percentage that you want and forward those calls and figure out what calls or what flow is making your customer have a positive experience based on that you can do a/b testing and then when you're happy you make that your main flow and so on so actually you can start do what you typically do with other channels and transform and quickly based on the feedback you get from your customers it's very powerful imagine doing that on your traditional contact centers you will need so much infrastructure to be able to do that with this it's very easy exported modified centered request to a different floor and you're done so one of the question is which can mention well how do I store some of these credit card details for example how do I do that to make sure that I'm still compliant well with the flow it's very easy what we're going to do in this case we're going to put the agent on hold so that way I can ask the user to input to input their credit card details so the agents actually on hold they're not list that they can't hear the conversation and I'm going to prompt the user to enter their credit card details the credit card details once they enter them we're actually going to encrypt those as soon as they enter it so that way the agents actually not hearing that and also restoring it in encrypted way and then once that's done we put the agent back on the call so they can continue service the customer so again this is a simple flow that shows you how you can actually capture secure information without agents listening and also maintaining your compliance so let's take a look at what we've done when we brought all those different AI machine learning services and AI services and we've integrated wit connects so what we did we took the core recordings from connect and we pass them into transcribe so we can get the transcription then we're going to pass that in to comprehend so we can get the sentiment analysis done on it and also store that against ice tray data like for additional analytics if you want to to provide that dynamic conversation we use the chat board using Amazon legs now legs can also be integrated we translate to support multi-language now to pro to those dynamic prong prompts that we were sending back to the couple users they are gone through Pali so we convert the text into a lifelike speech so that way they're still natural and we can actually select how you on the what voice and what language did we want to send those back so in this example that we've put together we were able to show you how you can bring different AI services integrator really easily we connect to provide that true customer centric experience for your customers so it's truly an omni-channel experience and you can do that very easily at the same time meeting your compliance and regulations and being able to transform your customers easily and make changes as you need to I can see ken is like dying to ask us some questions well yeah I hope you've caught some of the vision of what's possible but I didn't want to leave you with some questions upon there right we're builders that's where we're here at reinvent we are going to attack these challenges of compliance and customer expectations by automation by using services like machine learning but just some things to ponder if you put your compliance hat on you know what's your level of confidence that all your calls are being recorded and then you can retrieve them again and the sec 17a for requirement that they're easily accessible if you take nothing away from today I would say at least consider s3 as a place that's great for storing call recordings right the price performance is very attractive durability lots of ways to get data in and out but once it's there it's a data leak that you can run analytics on top of and then what about those calls or portions of calls which shouldn't be maintained again that portion of the call we had to take the payment card data or compliance with GDP are like how easy it is is it to find and not retain ones that shouldn't be retained and then let's be honest about what that's costing you right if this is traditional storage if there are many systems you know what is what is that cost of retaining and producing evidence that you're compliant with these regulations really costing you and compare that to what it might be possible on s3 but once we have it there you know think about what you can do so compare you know the vision we've cast today with the you have today like what percentage of your calls you actually reviewing for compliance and quality and customer satisfaction how much business insight are you deriving again by being able to take an audio recording and very simply transcribe it and then do some sentiment analysis and keyword analysis on it you know what insight might you be finding right how can you have your humans that are context and they're professionals who are used to reviewing calls for Quality Assurance purposes more focus their limited hours during the day on the outliers the customers who had really good experiences calling about a problem that usually leads to dissatisfaction or you know which agents are their top performers which ones need coaching right how can you apply those better by listening to all the calls or letting the Machine listen to all the calls and telling you where to focus your energy and you know how many of those insights are just being left on the table because that the call never gets listened to and then once you derive an insight which you have found something like oh I I found an opportunity for improving the customer experience of increasing my odds of retaining that customer how long does it take to put it into action yeah I think the the this space we have lots of best to be products right we have an ACD product an IVR product we have workforce management products and and the slide they Hannibal showed where Amazon Connect you know which we built because we surveyed the market and we didn't we didn't like the notion of stitching a lot of things things together we didn't like the lack of agility that would give us to once we find insights to pivot very quickly validate them through a be testing and then get the the best ones out to customers we didn't like that which is why we built connect so hopefully you found this interesting if you want to go deeper the contact center team maintains their own blog as does the machine learning team I would definitely encourage following both of those there are a few tutorials you can walk through on line one you know walking through building out a connect contact center another building a similar pipeline that we showed to transcribe blogs and then do insight analysis with comprehend while you're here when we're done Hannibal and I will be around for questions but we're gonna then walk down to Bellagio where we're gonna assist a couple of our other colleagues for a two-hour hands-on workshop where they will take call transcriptions they'll put them in a data Lake and they'll get some hands-on with Athena and other services to to query so how can I find those dissatisfied customers and and take corrective action and then their couple other sessions there are many more on connect and contact centers and machine learning and I've just called out a few so thanks for spending some time with us handle I will hang around as long as they will let us and please remember to fill out your surveys [Applause]

Channel: Amazon Web Services

Views: 1,514

Rating: 5 out of 5

Keywords: re:Invent 2018, Amazon, AWS re:Invent, Financial Services, FSV301, Amazon Comprehend, Amazon Connect, Amazon Lex, Amazon Transcribe

Id: erayXNsBT0Y

Length: 56min 13sec (3373 seconds)

Published: Wed Nov 28 2018