2020 Cisco Forums Profile

It’s nice to be recognized

I got a nice surprise in my inbox today. An email from Cisco letting me know that I was the first ever winner of the English Community Developer of the Month. Per Cisco the Community Spotlight Awards:

… recognizes members whose significant contributions designate leadership and commitment to their peers within their respective communities, including the Cisco Learning Network (CLN) and Cisco Community. Spotlight Awards Program is designed to recognize and thank individuals who help make our communities the premier online destination for Cisco enthusiasts.

I get a cool badge to show off too.

2020 Cisco Forums Profile

2020 Cisco Forums Profile

You can find current and past winners here or try to spot me in the picture below.

Current Spotlight Winners

Current Spotlight Winners

Looking back through my blog posts in 2008 I talked about getting my first star due to my contributions in the Cisco NetPro forums and how happy I was about it. In that blog I have a picture of my profile showing a total of 103 posts made with 8 questions resolved. That number, 12 years later, has ballooned to 3030 posts and 208 solutions.

I encourage anyone starting out or a seasoned veteran to contribute in the various Cisco Communities it’s a great way to network with your Cisco peers and try to tackle some very interesting technical problems while you procrastinate from your not as interesting technical problems.

~david

Another round of simple things you can do to create a better call center.

Back in September 2019 I talked about some minor and inexpensive things you can do to improve your customer service. This topic comes up often as many customers want to make incremental improvements without breaking the bank. The focus on this follow-up post is to try and provide another round of simple things which will yield improvements. Use these tips and the ones in my previous post before making any huge investments in your customer service strategy.

Have consistency across all your inbound numbers. This one is specifically important for healthcare. If I call your pulmonology department or I call my PCP, it’s ideal to have the same menu structure and same get out mechanisms. Trying to remember what options work for my pediatrician and for my neurologist creates unnecessary friction which really shouldn’t be there. If you absolutely have to have different flows, use this opportunity to compare and contrast which flows behave better and use data to use the best flow in as many departments as possible.

Have your agents live where the information is found. There’s nothing worse than hearing agents banging away a novel on their keyboard when they are talking to you. Surely I’m not asking a question which they have never heard before and surely they don’t have to type these many words for every call, right? CTI connectors for CRMs/ERPs are getting cheaper and cheaper and there are plenty of tools available which allow keyboard shortcuts and templating. If your agents are repeatedly typing out the same phrases this is an easy win for automation and get immediate returns.

Agent training and retraining. The best run call centers have a lot of communication between agents, supervisors, management. There is constant reminders about the work they do, why they do it, and how to do it better. Training and refreshers happen constantly and they expand beyond what to say to the customer, but also how to better navigate tools, how to deal with tough calls, and how to improve their writing. All of these things create a better experience for everyone around.

Collect some information. Every call center dreams of 100% deflection. Bots, virtual agents, etc., all with a single purpose to prevent the caller to talk to a human and have a computer answer their question. However, not all call centers even have any type of self service, but even if you don’t you should still have your customers provide you with some piece of information. It can be something simple like their zip code or what state they are calling from or more complex like their customer or account number. Either way, training your callers to have some information to give you does a few things: paves the way for adoption of self service to be easier, makes you seem like you’re more advanced than you really are, and give some extra data which you can later use for analysis. What data you ask for is certainly depended on the call center, but in my opinion asking anything from the customer is better than nothing.

This is getting longer than I expected, I’ll work on releasing part 3 of this at a later point.

Be well,

~david

Bringing Amazon Lex into your Amazon Connect flows

In this blog we’ll continue our discussion around Amazon Lex. Talk about a few things to keep in mind when integrating your Amazon Lex bot with your Amazon Connect flow. In my particular use case I wanted to use Amazon Lex to look at my Gmail calendar and book a meeting if I’m available. If you want to skip to the very end you can see the end result via video. You’ll see one video of the voice interaction and one of the Facebook Messenger interaction.

First, you might want to reference my previous post around Lex validation. Now let’s talk about our use case:

  • Lex easily allows you to build a bot which understand both voice and text, so our bot needs to handle calls into our call center as well as Facebook Messenger interactions.
  • Bot needs to to ask a few question in order to find out what time the user would like to meet.
  • Bot should only schedule calls between Monday-Friday and 10 AM – 4 PM Easter Time
  • Bot (using Lambda) should schedule a meeting and if slot already taken then suggest an alternate time to meet.

Second, let’s take a quick look at the Lex screen. The bot I created is very simple and it follows closely the Flowers example provided by Amazon. These are the slots I’m requiring my bot to confirm.

image

I used two different Lambda functions. One for validation and one for fulfillment. While most examples seem to focus on using the same function for both, for me it was easier to have different code bases for each with the added benefit of keeping the code manageable. As it is both validation and fulfillment both came in at around 250 lines of code, but fulfillment had around 9 megabytes of dependencies.

image

Finally, here are sample utterances I used for the main intent.

image

What this gets us is the following. The first video is the voice interaction. I went about it the long way to show some of the validation rules being set by the bot, such as no weekend meetings and no meetings too early in the day. At the end of the video you see I refresh the Gmail calendar to show the new appointment has been saved.

In the second video I go through the same Lex bot using Facebook Messenger and then show the calendar to prove that the appointment was saved.

Ultimately, Amazon makes it extremely easy to create a mutli channel bot, however the integration to back end systems is the tricky part. This bot needs a lot of tuning to make it more natural, but for just a few hours of work there’s very little out there that can get your call center to have some bot integration for self service.

~david

Creating a Lambda Function to Validate Lex Input

In this blog we’re going to step a bit away from Amazon Connect and focus on building a conversational interface using Amazon Lex. As you can probably guess down the line, this interface/bot is going to be connected to Amazon Connect for even more contact center goodness. Here we’re going to focus on creating a Lambda function strictly for validation, not for fulfillment.

First, let’s talk about what I’m building. I’m building a bot which can schedule a time to have a call with me. You tell your intention to the bot “schedule a meeting/call” and the bot will then ask you a few questions using directed language to figure out when you want to meet. Once Lex has all the information it needs it goes out to my calendar to figure out if I’m free or busy. Second, the validation code I have is mainly based on one of Amazon’s great blueprint for ordering flowers. I recommend you start with that before trying to write your own from scratch. Finally, read through the code and pay close attention to the comments marked in bold as these were the biggest gotchas as I went through.

A couple of things to keep in mind when building a conversation interface with Amazon Lex and you’re using validation.

– Have a clear scope of the conversation. I’m not a VUI designer by any means, but if you’re planning on going with an open-ended prompt “How may I help you?” you will be working on this for a long time. Instead try to focus on the smallest possible outcome. Ultimately, it is my opinion that no IVR is really NLU and they are all just directed speech apps with a lot more money sunk into them so they can be called NLU IVRs.

– If you’re going to use input validation, every user input will be ran through Lambda. This means that you must account for people saying random things which aren’t related to what your bot does and these random things will be processed through the validation function and might generate errors. Thus, you need to ignore this input and direct the customer to answer your question, so you can move on.

– Separating validation from fulfillment makes the most sense. Other than making your code easier to read and manage, you’re also able to separate responsibilities and permissions between your two Lambda functions.

– Play around with the examples Amazon provides. They are a great tool to get started and give you a ton of building blocks you can use in your own bot.

Here’s the validation code as well as some notes, hopefully this helps someone else along the way.

'use strict';

// --------------- Helpers to build responses which match the structure of the necessary dialog actions -----------------------

//elicitSlot is in charge of building the request back to Lex and tell Lex what slot needs to be re-filled.
function elicitSlot(sessionAttributes, intentName, slots, slotToElicit, message) {
return {
sessionAttributes,
dialogAction: {
type: 'ElicitSlot',
intentName,
slots,
slotToElicit,
message,
},
};
}

function close(sessionAttributes, fulfillmentState, message) {
return {
sessionAttributes,
dialogAction: {
type: 'Close',
fulfillmentState,
message,
},
};
}

function delegate(sessionAttributes, slots) {
return {
sessionAttributes,
dialogAction: {
type: 'Delegate',
slots,
},
};
}

function confirm(sessionAttributes, intentName, slots){
return{
sessionAttributes,
dialogAction:{
type: 'ConfirmIntent',
intentName,
slots,
message: {
contentType: 'PlainText',
content: 'We are set, do you want to schedule this meeting?'
}
},
};
}
// ---------------- Helper Functions --------------------------------------------------

function isDateWeekday(date) {
const myDate = parseLocalDate(date);
if (myDate.getDay() == 0 || myDate.getDay() == 6) {
console.log("Date is a weekend.");
return false;
} else {
console.log("Date is a weekday.");
return true;
}
}

function parseLocalDate(date) {
/**
* Construct a date object in the local timezone by parsing the input date string, assuming a YYYY-MM-DD format.
* Note that the Date(dateString) constructor is explicitly avoided as it may implicitly assume a UTC timezone.
*/

const dateComponents = date.split(/\-/);
return new Date(dateComponents[0], dateComponents[1] - 1, dateComponents[2]);
}

function isValidDate(date) {
try {
return !(isNaN(parseLocalDate(date).getTime()));
} catch (err) {
return false;
}
}
function buildValidationResult(isValid, violatedSlot, messageContent) {
if (messageContent == null) {
return {
isValid,
violatedSlot,
};
}

return {
isValid,
violatedSlot,
message: { contentType: 'PlainText', content: messageContent },
};
}

function validateMeeting(meetingDate, meetingTime, meetingLength) {

if (meetingDate) {
if (!isValidDate(meetingDate)) {
return buildValidationResult(false, 'MeetingDate', 'That date did not make sense. What date would you like to meet?');
}

if (parseLocalDate(meetingDate) < new Date()) {
return buildValidationResult(false, 'MeetingDate', 'You can only schedule meetings starting the next business day. What day would you like to meet?');
}

if (!isDateWeekday(meetingDate)) {
return buildValidationResult(false, 'MeetingDate', 'You can only schedule meetings during the normal weekday. What day would you like to meet?');
}

if (meetingTime) {
if (meetingTime.length !== 5) {
// Not a valid time; use a prompt defined on the build-time model.
return buildValidationResult(false, 'MeetingTime', null);
}
const hour = parseInt(meetingTime.substring(0, 2), 10);
const minute = parseInt(meetingTime.substring(3), 10);
if (isNaN(hour) || isNaN(minute)) {
//Not a valid time; use a prompt defined on the build-time model.
return buildValidationResult(false, 'MeetingTime', null);
}
if (hour < 10 || hour > 16) {
//Outside of business hours

return buildValidationResult(false, 'MeetingTime', 'Meetings can only be scheduled between 10 AM and 4 PM. Can you specify a time during this range?');

}
}

if(!meetingLength){
return buildValidationResult(false, 'MeetingLength', 'Will this be a short or long meeting?');
}
}
return buildValidationResult(true, null, null);
}

//--------------- Functions that control the bot's behavior -----------------------
function orderFlowers(intentRequest, callback) {
const source = intentRequest.invocationSource;
//get appointment slots
const meetingDate = intentRequest.currentIntent.slots.MeetingDate;
const meetingTime = intentRequest.currentIntent.slots.MeetingTime;
const meetingLength = intentRequest.currentIntent.slots.MeetingLength;

//For fullfilment source will NOT be DialogCodeHook
if (source === 'DialogCodeHook') {
//Perform basic validation on the supplied input slots. Use the elicitSlot dialog action to re-prompt for the first violation detected.
const slots = intentRequest.currentIntent.slots;
const validationResult = validateMeeting(meetingDate, meetingTime, meetingLength);

if (!validationResult.isValid) {
slots[`${validationResult.violatedSlot}`] = null;
callback(elicitSlot(intentRequest.sessionAttributes, intentRequest.currentIntent.name, slots, validationResult.violatedSlot, validationResult.message));
return;
}

//Pass the price of the flowers back through session attributes to be used in various prompts defined on the bot model.
const outputSessionAttributes = intentRequest.sessionAttributes || {};
callback(delegate(outputSessionAttributes, intentRequest.currentIntent.slots));
return;
}
}

// --------------- Intents -----------------------
function dispatch(intentRequest, callback) {
const intentName = intentRequest.currentIntent.name;

//Dispatch to your skill's intent handlers
if (intentName === 'MakeAppointment') {
return orderFlowers(intentRequest, callback);
}
throw new Error(`Intent with name ${intentName} not supported`);
}

//--------------- Main handler -----------------------
//Route the incoming request based on intent.
//The JSON body of the request is provided in the event slot.
//Execution starts here and moves up based on function exports.handler => dispatch =>orderFlowers=>validateMeeting=>buildValidationResult is the most typical path a request will take.

exports.handler = (event, context, callback) => {
try {
dispatch(event, (response) => callback(null, response));
} catch (err) {
callback(err);
}
};