CX Weekly Live: Autonomous Agents And Automation in CX
Michael Fauscette discusses autonomous agents and automation in customer experience.
All right, I think that's it.
Michael, hello.
Thanks for coming.
Thanks.
Thanks for inviting.
Yeah.
Let me-- let's see here.
I always have two screens going on.
I think we're good here.
So first of all, for those that are listening here,
and we'll kind of join the stream,
I just want to thank you, Michael, for joining us.
We just were introduced a day ago.
So first and foremost, I'm going to let
us do my goals, reduce your style, because obviously, I would not
do any justice there.
So let's start there.
Sure.
I've been in tech for a long time.
I'm an X-Nable Officer, and I got involved in software.
Then that's what they do with liberal art
to people when you go in the Navy.
They turn you into engineers.
And so then when I got out, I started
working for software companies.
I've worked for seven vendors.
I work for companies like PeopleSoft in the mid '90s,
Autodesk, I've been involved in a bunch of startups.
And I was eventually moved to an analyst role at IDC.
I ran the Enterprise Apps Group there for 10 years,
went to G2, which a lot of people have heard of probably now.
And I was chief research officer there for five years.
And I'm still an advisor to them.
But I started my own analyst firm three years ago.
It's called Arianne Research.
And we tend to focus a lot on AI and CX,
but I've been around the space for a long time.
So I still do some work on the ARP and some other things.
And I'm a chairman of an IoT company
and on a board of a sales tech company as well.
So I kind of dabble across a bunch of different things.
What I'm actually excited about, Michael too,
is that I've had--
and I know you've had these type of conversations longer
than I had about AI and customer service.
But I love branching out and just getting another person's
insights that I think I got a higher level too.
Because my insights were more of the tech company,
like working at customer, hearing from prospects and customers.
But only those prospects and customers
that are within our industry.
But to get your insight there as well too,
from someone that's outside of it.
And when you brought up autonomous agents,
and especially automation, obviously,
that was music to my ears.
So I'm happy to get right into it, which is like--
and I asked you this before-- but you're
just helping us be on the same page of when
you mentioned autonomous agents.
What do you mean by that?
Because it can have several different definitions
across different industries.
Yeah, the simple way that I like to look at it
is any AI-enabled system that can operate with independently
or with minimal human intervention
to make decisions, take actions in real-time data.
That is an autonomous agent.
And there are obviously a lot of different use cases
around that.
And we're, of course, with anything in AI,
I should do the preamble of.
I may say something today, and it changes tomorrow.
So that's--
Or the next couple of.
It is evolving rapidly, hang on.
But no, the AI piece of it is obviously real important.
And then when we moved to generative AI two years ago,
as a topic and as a tool, really, to use,
then I've seen that evolve as well into this automation space,
especially around autonomous agents.
And it really is moving from doing things that are generative
around generative AI, like making content to doing things
actually taking action for you, which is an evolution, for sure.
So from your opinion and standpoint,
when it comes to customer service and autonomous agents,
where are we at today?
Are we basically at the--
the Professor, Iran Chapter 1?
Are we in the middle of the book?
What are we at from your opinion?
The part-- the hard part about this
is that I don't know how long the book is.
So I'll say that we're a few chapters in,
and we are making really rapid progress.
So to make that simpler, I guess, to say that for the first,
what, 18 months of generative AI, there
was very little of it taking an action independently.
There was a lot more of it generating content
and doing some content dynamically and personalization,
all the sort of things in marketing.
But now we've gotten to the point where you could take
an agent, a chatbot, is a simple way
to think of it in customer service, right?
You could take that chatbot and you can now train it
with some contextual data.
So you're using retrieval augmented generation.
So in other words, a vector database that
stores your information so it doesn't go into the large language
model.
And you can take that, and I had the context around it.
And then on the other end, you enable it now
to go review documentation.
You train it to own all of your information
and then when it interacts with a customer,
it can interact with the same level of knowledge
or maybe even more than a human agent would.
Then as it evolves more, and we're just on the edge of this,
but as it evolves more than that agent could also take action.
So I'll give you a simple example.
Let's say you order something from SACS,
and it comes in, and it's the wrong size.
So you jump online and you're interacting with their chatbot.
And they're using an autonomous agent,
but it doesn't have full capability yet.
So what does it do?
Well, it interacts with me.
It knows who I am.
It looks up my order.
It sees that, in fact, I ordered something that was monogrammed.
But what happened was the monogrammed thing was incorrect.
So normally that would be a, we can't take anything
monogrammed back.
And an old chatbot would have just gone, sorry,
can't help you.
In the new world, that one goes, ah, okay, well,
we did this incorrectly.
So I've, it thinks past that and it goes, okay,
here's how you return it.
Talks you through the process, points you to the portal.
That's stage now.
Stage next week or next month or soon is that it then says,
okay, I can help you with that return.
And it processes all the return actions for you.
So that's, you know, we're right at the edge of that.
And then the next step will be moving into things
that can solve problems for you.
And this could even, you know, think about it
from a customer service, think about it internally
from an IT service perspective.
You could even have it take, you know, take actions like,
I'm having problems with this on my laptop and the agent
could go out and repair something, fix something,
change them, you know, so it'll be close to
and soon doing those types of actions
in a customer service environment.
- Yeah.
And I want to get your feedback on something
that I've heard a lot about, which is the educational stance
on this, right?
So in your example, that you just used,
this is again, brand new technology,
we're solving newer use cases.
So it's the education of the tech company
because this is still brand new technology for us
to output myself and the technology.
So we have to, I'd start educating ourselves right now.
We have to start learning about the products ourselves.
And then obviously the consumers have to learn
how to engage with this.
And then you have the brands themselves learning
how to implement this sort of technology.
Because like my question is like,
how big of a hurdle is that?
And could that also hold up how fast we actually execute
on all of this?
- Well, interestingly, I mean, there's always some level
of education there, but the way that a lot of these products
are being put together, they're being built on top
of some type of a platform that lets you
with load code or no code, generally no code,
be able to go in and configure the thing.
'Cause you're not really, you're not training the language model.
'Cause you presumably have an integration
to some language model.
Maybe you had your own, maybe you're using chat GBT
or, you know, entropics, cloud or something.
So you have a language model and then you have a system
that uses this retrieval logmented generation.
I told you about that is a way to get all your data
in there safely, contextually.
So you provide context.
'Cause if you know anything about generative AI,
one of the most important things in any prompt
is giving it the right context.
Because if you don't and the more context you can give it,
if you don't do that, then you see the old hallucination problem.
But the way things are being built today
is then with those no code tools on top,
I would take that agent and train it,
but it's not like training a language model
that takes a lot of effort, a lot of heavy lifting,
a lot of tech, all those things.
It's simply letting that data and letting that agent
interact with that data and be able then to work
from that data, not just the language model,
but the language model is important,
but also that contextual data from your company.
And so then it learns all of that.
The same, I mean, honestly, same thing,
pretty much you'd be training a human agent on,
except that it obviously can, has more data capacity
than a human would.
So it's gonna retrieve that much more easily.
- So on that note, Michael,
what would you tell, maybe you've had these conversations,
like what would you tell leadership
in terms of implementing this?
Is this a whole 'nother department in a company?
Is this just changing people's roles from like,
okay, you used to do this,
now you have to be a workflow AI editor
or you have to be understanding the feedback?
Like, what, you know, I would you tell a leadership team
to make sure that they're ready for this sort of implementation?
- That's a big question, and I guess, you know,
we can stick to talking about autonomous agents,
but if I was actually given a leader advice,
I would say the first thing you need to do is step back
and build out a complete strategy
because AI is a lot more than just this one place
we're talking about, but let's say for the sake of discussion
that they've done that, they have an AI strategy,
it aligns with their business strategy,
it aligns with their IT strategy,
and then they would look for use cases,
and that's, frankly, I've spent a lot of time over the last year
so digging into different use cases of generative AI
and traditionally, I like to separate them that way,
just so you can kind of keep it straight
'cause traditionally, I, machine learning, algorithms,
deep learning, all these things that we've had for a while,
those are important, they're still important in what we're doing,
you still need them even in some of the applications
of generative AI underneath it
because the generative AI ends up being the front of it.
But anyway, I would tell them that if they're,
that use case makes sense that they need,
you know, customer service automation,
then that's when they would evaluate the tools that are available,
and with the idea that it's gonna evolve very quickly,
so pick a partner that you believe will grow that capability
with you because there's still a lot of things that you,
you need that agent to learn and do.
I mean, a simple example is empathy,
and we've talked about empathy a lot,
that's a very important thing in customer service.
As a human, I want you to, you know, in certain cases,
I want you to feel empathy for my problem,
AI doesn't have empathy 'cause it's not human,
but we are working towards a way that they call it synthetic empathy.
I kind of hate that, but it is a thing.
That, you know, that's the language model
and the agent be able to interact in a way that is,
is reasonably sensitive to the language
and the way it's using things and the way it phrases,
things the way it responds to things, that sort of thing.
So, you know, those are problems that'll be solved,
but they're in process, I guess, right now.
- That's still interesting 'cause again,
we're talking about AI, right?
It's artificial intelligence,
so artificial empathy,
and then trying to like disguise that as real empathy.
That's so interesting there.
- Yeah, now I will say one thing that I would always tell you
is that I believe personally,
you should always disclose that a person's talking
to a chatbot or a person's talking to a human, don't.
And we're getting to the point now
where you could actually fool them.
Like the new chatbot.
- No, you can, can, yeah.
Intelligent bots anyway.
So that's the one thing I'm big on from an ethics perspective.
I feel like you need to be transparent about that.
- I would agree too, and I will say, like obviously,
I don't wanna call out anything, anybody.
But there are, I mean, there are services, right?
That are like, hey, we're trying to replicate
the human experience, we're not gonna say anything.
And what I've seen in like successful companies,
I mean, if you look at actually customers homepage right now,
we say AI bot, like we call it an AI bot,
so they know.
And I do think that is,
I think that's gonna be more of the norm
because as we start more innovation
and more companies start utilizing
these AI agents.
- People are just gonna be used to like,
okay, I know I'm gonna be interacting
with some sort of AI upfront.
And I do think that's gonna be the default setting
sooner or later.
- Yeah, well, and I did a survey last year,
consumer communication survey,
and we asked some questions about interacting with companies
and what preferences, consumer preferences were.
And if you ask somebody straight out,
if they wanna talk to a chatbot, they'll usually say no.
I mean, the overwhelming majority of people would go,
like, and I don't really wanna talk to a chatbot.
Just let me talk to a human.
But if you couch it differently and you say,
how many times have you been helped by a chatbot
when you called in for service
and it was a good experience, you'll find a lot of people
have positive experiences,
particularly now with newer chatbots.
And then if you go further and you go,
are you okay with talking with the chatbot first?
And if the problems result fine,
if it's not, that chatbot hands over to a human agent
and passes all the context,
so it's none of that, you know, forgetting who you are
and you have to go back through your story again.
It's all this contextual, it's passed on.
They're very open to that idea.
And I think that's the experience that we're gonna see
for the foreseeable future.
Automation's important and automation is growing
and being able to take actions using those autonomous agents
is important, but at the same time,
there's still a place for that human agent.
And frankly, it even lets you
from a customer service perspective,
start to think about your customer service approach differently.
If you're in a business where you can upsell,
acrosssell or interact with your customer,
you free up those human agents, teach them to be able
to help you generate revenue or improve
the customer service satisfaction levels,
those sorts of things.
So simple things get handled by the AI, by the agent.
And more complex things go further up the chain
to a human if need be.
And then as humans can also be encended
to do other things versus the old idea
of I'm gonna try to deflect.
'Cause deflections not customer service,
I realize why you want to do that,
but is it a really a good experience
if you just deflect as much as possible?
What you really wanna do is create the best experience
at the best price.
And automation can help you do a lot of that
and leave your human agents to do
that higher level, more complex things
that the AI at least right now couldn't really do.
- I love that, the best experience at the best price
and then companies that will understand,
like, okay, am I gonna send this to AI or a human,
which I mean, humans would be more expensive,
but who knows in a little bit what that could change?
But you also bring up something that I think
as a marketer myself,
like there's this preconceived idea of a chatbot.
Like I think, like you said, you ask people
what do they think of chatbots?
And I do think about the last five,
you know, I don't know, 10 years of chatbot experience
and probably hasn't been really nice.
And so going back to autonomous agents,
like if brands, I should say,
of technology companies start advertising this
as another chatbot or another version of a chatbot,
I do think they might run into problems like that.
I'm like, no, I'm not looking for a chatbot,
which is why we talk about autonomous agents
and all of a sudden it's like, oh, hold on, what's that?
And then obviously you have to make the connections,
like well, it's kind of like a chatbot,
this is, but this is totally different.
And I can see that a lot of companies running that
sometimes.
- Yeah, and the experience changes dramatically.
I mean, none of us liked the old IVR,
interactive voice systems because they were logic-free-based.
And the same thing happened when we rolled out chatbots,
they were logic-free-based.
So if it's in their realm of logic-tree, you're fine.
But if it goes outside of that,
then the chatbot simply can't do anything
because they don't have the capability
to generate new content or new bot, new interaction.
A generative AI-based autonomous agent
has the capability to synthesize all of that data
that it has and take other action outside of what a logic-tree
would be because it learns as it interacts.
So that's a very, very different experience
than trying to go through a logic-tree.
Well, one thing that I've never mentioned this out loud,
but I think it's fascinating is there's
a element of who's in charge of the conversation.
If you're talking to a chatbot or even an IVR,
you know that the IVR is in charge of it
because it's already predetermined
and it's up to you of which way you're going to go around.
But if you're dealing with an autonomous agent,
you're still technically in charge because you can actually
get, these are the questions that you want solved.
And you're going to have follow-up questions.
The AI agent is really, they have instructions,
but still there's this preconceived
like the visitor that actual customer is really in charge
of how they're going to get their answer, right?
Is that something you thought of?
Shift the interaction back, right?
It shifts the interaction back to the customer
so that they're in control of this.
And if they want, you know, in the old days,
you're yelling at your phone, "Give me a human, please."
It's actually probably not please.
And now, you know, they have that interaction
that can be more in control and control
when they want to go to a human if they're not getting
what they're trying to solve.
But the nice thing about these agents is they, you know,
they learn they evolve over time.
They can be trained to, in these situations,
this is when you would hand it to an agent
and here's how you would do that.
And that's important.
The other thing that's interesting now,
because we've, again, we've evolved quite a bit,
that agent can also, in real time,
analyze the sentiment in the conversation.
So it knows if you're getting angry
and it can then go, "Why are you getting angry?
Am I doing something incorrect?"
Maybe I didn't give you the right information.
Maybe I need to pass you to a human now
because you're getting upset.
Like it can know that you're happy, you're sad,
you're mad, you know, all those things.
And that, and it can take that into account
as far as the interactions concerned, which is pretty cool.
- Yeah, just one other, one other kind of footnote
on this is that, maybe for like a learning
for other companies here, like on the customer side,
we've put in AI bots and we're doing a little bit
of like customer service on our side
and trying to eat our own dog food.
And what I've seen is that when you,
when people do want to talk to a human,
it's so important to then bring that human on the call
like as fast as possible.
And I don't think we're ready for that yet.
It's like, hey, they're asking for a human now.
Be ready to join that conversation
because what has been set up beforehand
and historically it's like, hey, give us your email.
We're gonna follow up with you at a later time.
And now this gives that ability where that's now,
it's not a great experience.
Like you were answering questions so fast,
getting answer so fast and all of a sudden
you got a day for a response from an actual human.
- Yeah, that's a terrible experience.
Yeah, that's a terrible experience.
I mean, if you're trying to solve a problem,
if I'm your customer, I'm trying to solve a problem,
that means I'm trying to solve a problem now.
Like if I took the time to get online
and do a chat with you or whatever,
however you wanna interact,
maybe it's a voice call or whatever.
If I'm doing that, I'm looking for an answer now.
I'm not looking for an answer tomorrow.
And so yeah, these systems become real time
and that's important.
And then when you think of automation,
I like to break it down this way
because I think this is,
and this goes across all sorts of different areas
in the business, what I call decision intelligence,
for example, so ways to automate certain routine decisions.
There's really three ways you can approach this.
One is fully automated.
And there are a lot of things now,
and there'll be more and more things that agents can do
that can be fully automated
because you trust them, it's simple, it's phenomenal.
But there are other things that aren't ready
for full automation,
and there are two ways you could deal with them.
One is you could build a system
that puts human oversight in the loop.
And so there's human oversight to the decision
and this human could step in at any point
if it's not going the direction that makes sense.
The third way is human in the loop,
in other words, the human is making that decision all the time.
So you're not giving the agent the capability
to make that final decision.
And I'll give you the perfect example of that one.
That's your doctor.
So I love that my doctor could have access to a chatbot
that could pull all the medical data in the world together
to tell him what I might have,
but I do not want that chatbot telling him what I have.
I don't want the diagnosis to come from the chatbot.
I want the diagnosis to come from the doctor
because that's what we have doctors for.
So that's very much human in the loop.
And then human oversight would be,
like for example, a lot of companies have automated approval
for credit.
So you put in some little bit of data,
you're still security number, whatever.
And in the AI, then crunches in the background,
and goes, are you okay?
If you meet all the criteria, it makes the decision,
you got your credit, you're good.
If you don't meet all the criteria,
that's when the human can step in and look at it
and go, yep, this is a rejection or no, this is not,
this is a special case, we're going to approve this.
So those three ways to think of automation,
I think are important because they're three different things.
And those will change over time
because the AI will get better and you'll trust it more.
But in the meantime, having those three scenarios
and thinking of it that way, I think is important
when you're thinking about how to deploy it.
And I'm going to add on to that point, Michael,
just because I've also, just in the last month or so,
I've also talked about human in the loop.
Only because I haven't seen it as much
from tech companies really talking about it.
It's either your first option of all automated
or maybe that handoff, like that to a human.
But again, I keep on pushing on that human in the loop
because I'll give you an easier example outside of the doctor
of what's happening in e-commerce, right?
Like if there was a travel website and the AI agent
is able to then change reservation dates
for certain people and change schedules,
don't you at least want a human to kind of like,
at least confirm that?
Like, hey, yeah, I'm good with you confirming that date.
And so what I've seen of some examples is like,
okay, AI bot, AI bot works on it.
And then just throws the question over to a human being,
like, hey, I just want to get an approval process.
Like, you're good with me changing the dates.
And that's another example of human in the loop.
And it's like, it's a good step because I do think,
oh, there's a lot of people out there that's like,
oh, my gosh, I don't want the AI agent to handle everything.
And even like critical decisions,
even maybe they can make it, and maybe it's the right decision.
And like, don't you just want some overlooking
the approval process.
And so I couldn't agree with you more on that.
And that's where, that's why I like this idea
breaking it out from human in the loop, human oversight.
Because human oversight is really what you're describing there.
Right, I want the human to look at it.
And if it's in the parameters, I'm not going to do anything.
Sure.
And if it's not, I'm going to take action.
And then having humans in the loop,
they always make this decision.
So decision intelligence is a systematic way
to approach decision making in your business.
Automation underneath that with some things
that fall in these other categories.
And so what you see is people start with the simplest
decisions and automate those.
Because there's no risk.
But then over time, as you trust the system more,
you can move that automation line up
and start to include other things
that you originally may not have thought would go.
But once you realize that this is safe,
this risk is very minimal or whatever.
And over time, the agents get more capable.
And so more things can be automated.
And it's not-- we have a lot of fear around future work
kind of issues, right?
But the truth is-- and it's not totally misplaced.
But the truth is, I'm not suggesting that your AI is going
to come in and make all the decisions in your business.
What I'm suggesting is you're going to free up the humans
to handle the more complex things.
In the service example we were talking about, right?
Now, your agents could help you increase revenue
or they could have a metric around NPS.
So rather than trying to have call to flex and call volume,
first call resolution, those kinds of things that we've seen,
you can shift that in the metric shift.
And then your customer satisfaction will go up
because you're providing a much more bespoke service
when it's needed.
But it's not always needed.
But when it is, you could do it.
Yeah.
I'm going to shift gears just a little bit here before we sign off.
But one question I have you Michael, too, is as you've
talked to a lot of different people in the industry,
I know you just talked to Salesforce recently, too.
Do you think that the companies that are embracing--
let's just talk about autonomous agents?
Because that's the subject.
Do you think that enterprises right now
are moving quickly?
Do you think the small businesses right now
are moving quickly used to that?
Do you think everyone's on the same timeline right now?
Because usually, right?
You're not going to put both of those degree
that you have large companies.
And they're just slow to move.
They're slow to innovate.
You have security issues and so forth.
Are we seeing that?
Are you seeing that with people you talk to
with autonomous agents?
Yeah.
So I will take the--
there's sort of two approaches to this.
One is, I just saw survey data this week actually.
And I don't remember the exact number,
but it was really high that small businesses
are adopting AI at a very, very fast pace.
Because it is such--
it's easy to pitch the productivity gains.
There are other reasons that you should look at it.
But that's obvious to a small business.
Interprises, they do-- they definitely move slower
in the evaluation, in the rollouts, all those things.
But we're seeing some really big brands
that have gone down this road because the technology
is evolving very quickly.
Now, last year, I would have said
that a lot of those companies are doing prototypes
and test and trying out these different ideas and piloting.
But now, that's not really true.
They're just putting it through their normal process.
So they have a tool.
They have a platform they want to go down.
In this path, they've gentied.
In the five, these use cases, they'll build it.
Maybe it's no code.
Maybe it's low code.
Maybe it's pro code.
It just depends on what you're trying to do.
And once it passes that QA, then it goes into production.
And you mentioned Salesforce.
So I'll say at DreamFourse, there were several very large brands
that told stories about what they were already doing.
The SACS example I gave was one of those.
In fact, that they are already using.
They've already deployed an autonomous agent
in their customer service.
And they're already using them.
And the next step is to increase the level of automation
and they're moving down that road.
And there were a bunch of others.
So I think you're starting to see real projects
go into production because a lot of those companies
started evaluating this a year and a half ago.
Ish.
Yeah.
And also for some context, too.
I usually come from the small business side.
And as I've started seeing this, I got super surprised
in last month because it was a combination of DreamFourse
only talking about agent force.
I mean, I think the quote from Benny off
was something like that.
He spends 100% of his time thinking
about autonomous agents, which for Benny off to say that it's like,
OK, this is a big deal for mid-site, like for all
types of companies.
HubSpot came out with, I think it was Blaze, right,
for a race, something like that.
But they're autonomous agents.
And obviously, they're more now in the red market
as opposed to smaller companies.
You have service now that is now
mentioning autonomous agents for all their enterprise
companies.
So it's super interesting because you have the startups.
You have the VCs putting a lot of money into the small companies
here.
And then you have all these large enterprises now
to really come out of autonomous agents.
So it's coming from all sides.
So that's why I asked that question.
Yeah.
And I mean, I think that it is gain attraction.
When you first started down this path with generative AI,
a lot of it was focused on productivity.
And maybe even to its own detriment,
because a lot of feedback I've seen since then
has been that we got this generative AI tool.
And if I'm a worker, they gave it to me.
They didn't teach me how to use it.
And they raised my quota or whatever the measurement is.
They quadrupled it because it's
supposed to make me that much more productive.
And I don't know how.
That was kind of the old problem.
And that started to mitigate a lot.
Now I think it's people starting to realize that there--
you have to understand the use case.
You have to make sure the tool is the correct one for the use case.
And the if your approach is the correct one for now.
And you may evolve that over time.
But it's been intentional about what you're doing.
And I think that we're getting out of just kind
of raw experimentation into very specific use case
examples that can be done.
And we're starting to see real metrics
that come back with success.
Yeah.
Well, I think that's a really good place to end it, Michael.
For everyone that's listening here, there's no script.
I think it's just awesome to have a conversation.
We could probably keep seeing you discover
station for days.
And then you have AI avatars taking on.
But this is the real thing.
So again, I appreciate all the time, Michael.
Yeah, enjoy it.
There's definitely a fun subject these days.
Yeah.
And it will continue to do so.
I don't-- like you said, I think we're--
I would agree we're probably at chapter 2.
Don't know when the end of this book is.
It could be 20 different chapters.
But let's see where it goes from here.
So thanks all.
Thanks, Michael.
Sure.
[BLANK_AUDIO]