00:00:00
Dan Loney: Well, has has been mentioned in
the past, almost every aspect of
00:00:03
business is being examined to
see what impact artificial
00:00:07
intelligence can have to improve
operations. Today, we take a look
00:00:12
at it from the customer service
perspective. Pleasure to be
00:00:15
joined by Christian Terwiesch,
who's a Professor of Operations,
00:00:18
Information, and Decisions here
at the Wharton School. Christian,
00:00:21
great to talk to you again. How are you, sir?
Christian Terwiesch: Hey, always glad
00:00:25
to be on your show. - Thank you,
sir. This is obviously an
00:00:28
interesting topic, especially
when you think about the fact
00:00:32
that customer service has been
one of these elements of
00:00:35
business that, in many cases,
has been ridiculed by the
00:00:39
public, but now we're talking
about it as a way that maybe
00:00:43
we can improve it. So when you
think about the connection
00:00:46
between AI and customer service,
what first comes to mind?
00:00:51
- I think the big challenge is
that, I mean, everybody on
00:00:55
the consumer base is familiar
with it, that we have many
00:00:58
disappointments in the customer
service experiences that we
00:01:01
have. Think about interacting
with your airline, your
00:01:04
insurance, your bank. Most of us
have horror stories of bad
00:01:09
customer support. And let's
acknowledge it, it's pretty damn
00:01:13
hard because it's a very
tedious, a very labor intense
00:01:17
piece of work that the service
provider has to provide on
00:01:22
demand, on the time, and
in synchronization with the
00:01:26
needs of the customer. So it's
nothing that you can do like on
00:01:29
an assembly line. So it's really
hard work. And I think many
00:01:33
companies have turned to AI and
have arguably somewhat naively
00:01:38
believed that this problem is
going to go away because of AI.
00:01:43
And I think we've seen some
things work, but many things
00:01:46
that do not.
00:01:48
- So for companies, is the
expectation then the customer
00:01:52
service, or is it the cost
savings they may have, or is it
00:01:56
a combination of both? - This goes
to the heart
00:01:59
of the problem, right? So you
think, in general, in strategy
00:02:02
and operations, we like to think
about the cost trade off, a cost
00:02:06
quality trade off, right? So in
one dimension, you can imagine
00:02:09
the efficiencies of productivity
as measured in dollars per
00:02:14
episode. And on the other
dimension, you think about some
00:02:17
version of customer delight,
willingness to pay, customer
00:02:21
utility. And there's a trade
off, right? I mean, I can give
00:02:24
you an amazing service if money
doesn't matter. I can give you a
00:02:28
glass of champagne, I can give
you a back massage, and I make
00:02:31
you the happiest customer, but
it's going to be bloody
00:02:33
expensive. Or I see like, hell,
you're going to wait an hour
00:02:38
long on the on the call center,
on the call line, I'm going to
00:02:42
give you a crappy service with
someone who barely speaks
00:02:45
English. So there's this trade
off. And the hope is that we can
00:02:49
push this trade off curve
outwards and give you better
00:02:53
service at lower cost. The
empirical pattern that I've seen
00:02:58
most companies do is when
they've shifted that frontier,
00:03:02
they hope the AI will give them
lower cost. So they're solving
00:03:05
for lower cost, as opposed to
better customer service.
00:03:09
- But don't companies still want
to have a human component
00:03:12
involved in customer service
somewhere along the line? They
00:03:15
just can't, you know, go 100 percent AI.
00:03:18
- Well, so that is a challenge,
and I think, for two reasons. So
00:03:22
one is, what do you want as a
company? There's a certain
00:03:25
argument that you say, like
there's a human touch. I mean,
00:03:28
again, think about maybe a real
estate agent. Maybe think about
00:03:33
a cardiologist. You kind of want
to talk to a human being just to
00:03:38
be safe, because the stakes are
really high. The second piece is,
00:03:43
independent of what do we want
as a company, what do we have to
00:03:47
do? And I think we're waking up
to the reality that ChatGPT
00:03:52
and other tools are really good
of getting us 80 percent of the way,
00:03:57
but not to 100 percent. And so most
people then kind of wave their
00:04:01
hands at some point and say,
like, "We're going to -- we need a
00:04:04
human in the loop." And that is
when it gets expensive, right?
00:04:08
That human in the loop is
costing a lot of money, and the
00:04:12
workflow is even more
complicated than before.
00:04:15
- But the expectation of the
consumer, in many cases these
00:04:18
days, and it's been talked about
a lot by retailers, is that they
00:04:22
need to make sure that the
experience for their consumer is
00:04:25
as good as possible, even if
they have the AI, so that that
00:04:29
customer will continue to come
back time and time again. They
00:04:32
don't want to lose them because
of a bad experience.
00:04:35
- Absolutely, right. I think the
customer expectations are high
00:04:39
enough, they're only getting
higher. And at the same time,
00:04:43
many customer support centers
are struggling with the high
00:04:47
cost and the high customer's
expectations. And again, because
00:04:51
you, in most settings, need some
form -- you need to keep some form
00:04:55
of a human in the loop,
we're not at zero marginal
00:04:59
cost. And so I see a lot of
companies struggling with, how
00:05:03
do we educate the customers that
they can do self service without
00:05:08
the human in the loop? That's
when we make the big money. But
00:05:12
again, that requires training or
educating our customer beyond
00:05:18
just kind of having them call up
ChatGPT.
00:05:21
- One of the things I saw written
about when you talk about the
00:05:25
connection between AI and
customer service is the
00:05:28
potential to improve the
language barrier for consumers,
00:05:31
with AI being able to speak to
consumers in the language of -- in
00:05:36
their own language, and not have
to have that misconnection
00:05:40
because somebody can't speak
Spanish or German or French or
00:05:43
whatever it
00:05:44
might be. - Well, you shouldn't
ask the person who speaks with a
00:05:47
German accent here, right?
I think there's language
00:05:51
barrier. And indeed, I think
that is a good point. I think
00:05:54
that the more important point I
want to double click on is
00:05:59
eloquency, the persuasiveness of
ChatGPT. We've seen in study
00:06:05
after study, it is better
arguing, it's better -- it's more
00:06:09
empathetic. It's more
lovely to talk with an AI person
00:06:13
than with a human being. And
that's scary, and tells us a lot
00:06:17
about human beings. But again,
it comes out in study after
00:06:21
study that AI is really sweet
and persuasive, and I think that
00:06:26
makes even more difference than
the language capability.
00:06:31
- So then I'll leave you on this.
What do you think, then, is the
00:06:33
longer term impact of
00:06:35
AI in customer service?
00:06:38
- I think for the near future, we're
going to have some form of a
00:06:40
hybrid workflow, where we have a
human who is augmented through
00:06:45
some ChatGPT type of
technology. It's a hybrid where,
00:06:51
again, I might be able to get a
little bit more productivity out
00:06:55
of the agent by helping the
agent through text prompts in
00:06:59
the screen and the
interview by automating some
00:07:04
follow up chat messages. But
hopefully, we can get to a point
00:07:10
where, in five or ten years, a lot
of the tasks that still require
00:07:15
a human in the room will be
fully automated. But I don't
00:07:18
think
00:07:19
we're there yet. - Christian,
great to talk to you, and thank
00:07:21
you again for your insight, sir.
- Always a pleasure. - Thank you.
00:07:24
Christian Terwiesch, Professor of Operations,
00:07:28
Information, and Decisions
00:07:29
here at the Wharton School.
