0:03 in the last two videos we looked at
0:06 writing and reading applications in this
0:08 video we'll look at chatting
0:10 applications in addition to the general
0:13 purpose chat boot like chat bot and ban
0:15 chat many companies are looking at
0:18 whether they can build specialized chat
0:20 applications if you're involved in a
0:22 company where you have many people
0:24 interacting with customers or having
0:26 certain types of conversations of
0:28 similar nature this may be a case where
0:30 you can consider what or Not A
0:32 specialized chat bot can help with those
0:35 types of conversations let's take a look
0:38 earlier we already saw the example of a
0:39 customer service chatbot that might be
0:42 able to take orders for cheeseburger
0:45 another example of a specialized chatbot
0:47 would be one that specializes in helping
0:50 you to plan trips so how can vacation in
0:53 Paris inexpensively and a bot could be
0:55 built to have specialized knowledge
0:59 about travel and today there are
1:02 companies exploring a wide range of
1:05 advice Bots for example can a bot give
1:08 you career coaching advice or give
1:10 advice on cooking a meal so a large
1:14 variety of specialized Bots that are
1:15 really good at answering questions about
1:17 one thing are being developed by
1:19 different companies today some of these
1:21 BS are capable of just having a
1:24 conversation and giving advice some of
1:27 these BS can also interface with the
1:30 rest of a company Software System and
1:32 take actions such as to put in an order
1:35 for a cheeseburger to be delivered
1:38 another example of a bot that might bu
1:39 take action would be a customer service
1:42 chatbot where it turns out that many it
1:45 departments get tons of password reset
1:47 requests and if a bot can take care of
1:49 that then it may take some of the
1:52 workload off your it department and a
1:54 bot like this that needs to be send a
1:56 text message to verify identity and
1:58 actually hope reset a password this is a
2:01 bot that would need to be empowered to
2:03 actually take action in the world such
2:05 as to get a text message to be sent to
2:08 someone next week we'll discuss more how
2:10 chat bars like these are built that
2:12 don't just generate text but can
2:14 actually take action because of the
2:16 number of customer service organizations
2:18 exploring the use of chat bot I want to
2:20 share with you a range of the spectrum
2:22 of common design points being used by
2:25 different businesses and for this slide
2:27 I want to focus on text based chat
2:30 rather than Voice or phone based chat
2:33 so at one end of the spectrum would be a
2:35 customer service center with only humans
2:37 so you would have Human Service agents
2:39 typing back and forth messages like
2:41 welcome to P burgers and let me play the
2:44 order for you at the opposite end of the
2:48 spectrum would be chat Bots only where
2:49 you just have software responding
2:53 directly to customers but between these
2:55 two ends of the spectrum of humans
2:58 typing at the keyboard or chat BS only
3:01 there are a couple common design points
3:02 one common design points would be to
3:06 have Bots support humans in which a bot
3:08 will generate a suggested message for
3:10 human but the Human Service agent will
3:13 read the message and either approve it
3:15 if it looks good or have a chance to
3:16 edit the message before it is actually
3:20 sent back to the customer this type of
3:22 design is often also called human in the
3:24 loop because as's a human that's looped
3:26 in and is part of the process before the
3:29 message actually gets sent back to your
3:31 customer and this is a way to mitigate
3:33 the risk of the chat bot maybe saying
3:34 the wrong thing because a human can
3:36 check over it before it's actually sent
3:39 back to your customer in the next video
3:41 when we talk about what lm's can and
3:43 cannot do we'll go over some of the
3:45 mistakes that LM can sometimes make and
3:48 so this design helps protect against
3:50 those mistakes of LMS a little bit
3:52 further on the automation Spectrum would
3:55 be if you have a bot triage messages for
3:57 humans so maybe the bot answer the easy
4:00 messages but escalate to a human for the
4:01 things that isn't quite ready to handle
4:03 yet sometime back I actually L A team
4:05 that build a bot that would
4:08 automatically detect if the customer was
4:10 asking for refund requests it turns out
4:13 that was about 10% of our total chat
4:15 call volume and by just detecting that
4:17 and automatically giving the customer
4:20 instructions this routed 10% or so of
4:23 the traffic away from the human agents
4:26 and so to save the agents a lot of time
4:28 and let the humans focus on servicing
4:30 the harder requests but this type of
4:34 triaging is another common design to
4:37 help your Human Service agents save time
4:39 and have to focus only on the harder
4:40 cases that they're more uniquely
4:44 qualified to handle in many customer
4:47 service centers a single human may be
4:49 simultaneously having chat conversations
4:51 with four or eight or in some extreme
4:53 cases maybe even 16 customers at the
4:56 same time and with B supporting the
4:59 humans it becomes easier for a human to
5:01 manage a a larger number of parallel
5:04 conversations given that Bots sometimes
5:05 say the wrong thing I want to share with
5:08 you what building and deploying a bot
5:10 often feels like in companies that want
5:14 to do this in a safe way often companies
5:16 will start with an internal facing chat
5:18 bot so many times I would build a chat
5:22 bot but let only my own team use it um
5:24 to say answer the questions about travel
5:26 or whatever the bot is supposed to do
5:29 and assuming your internal team will be
5:31 more sympathetic and more understanding
5:32 of mistakes and be more forgiving if the
5:34 bot says something wrong that one time
5:37 this gives you some time to assess the
5:39 behavior of the bot and also avoid
5:41 public mistakes that could be
5:44 embarrassing for the company after this
5:46 looks safe enough a common Next Step
5:49 would be to deploy with human in the
5:51 loop to let a human check over many of
5:53 the messages if feasible before it
5:55 actually goes out to the customer and
5:57 after doing this for a while if it looks
5:59 like the bot's messages are are
6:03 generally safe to send to customers then
6:05 you might allow the bot to communicate
6:07 directly with customers of course the
6:10 details of every business defers and for
6:12 some applications it may not be
6:15 practical to have humans check over
6:17 every message because of the Shar volume
6:19 of traffic but depending on the risk of
6:22 the bot saying the wrong thing as well
6:24 as the volume of traffic and thus
6:26 whether or not human in the loop is
6:28 feasible these are some of the design
6:30 patterns I've seen company used to try
6:33 to deploy Bots safely to summarize we've
6:36 seen how LMS can be used for writing
6:38 reading and chatting these three
6:40 categories are not meant to be an
6:42 exhaustive list of what LS can do but
6:44 there just a few broad categories of
6:47 what you might really use them for and
6:50 LS can do a lot but they can't do
6:51 everything in the next video let's take
6:55 a look at what LS can and cannot do and
6:57 better understand the limitations let's
