Navigating the Future of Customer Experience

0:00

(upbeat music)

0:02

- Welcome everybody, we're excited to get going here.

0:07

We're gonna be talking about AI and CX

0:10

and to do that we brought on a little partner of ours,

0:13

company called For Thought.

0:15

We've got the CEO and co-founder, Dion Nichols.

0:18

Dion, thanks so much for joining and how the heck are you?

0:21

- I'm doing well, Gabe.

0:22

Thanks for having me, really excited to be here.

0:25

And yeah, how are you today?

0:26

- Yeah, yeah, well man, it's been hard to get you locked down.

0:29

You're a busy man, but we're excited to kind of get down

0:32

and into the weeds on AI and CX.

0:34

But before we do that, cool stuff going on.

0:37

Tell us a little bit about your background

0:38

and what you guys are doing over there at For Thought.

0:41

- Absolutely, so as mentioned,

0:42

I'm the CEO and co-founder of For Thought.

0:45

We are the generative AI suite for customer support.

0:50

I like to say we're one of the OGs in the space,

0:52

so to speak.

0:53

We launched in 2018 and since then we've grown

0:56

to working with high growth companies,

0:58

everyone from Asana to Marriott.

1:01

And what we do is we plug into their existing

1:05

customer service workflows and then fine-tune

1:07

and train models to discover insights,

1:10

solve problems and assist agents using generative AI.

1:14

- Love it, love it.

1:15

Well, that's why we're here today

1:16

to talk a little bit more about that.

1:18

So you obviously, I assume have a little bit of an opinion

1:21

on AI and CX.

1:22

Maybe let's start big picture.

1:24

Lot of stuff obviously coming down the pipeline

1:26

just in the last little bit.

1:28

Challenges, good, bad.

1:30

What's kind of just your take on this concept of AI and CX?

1:34

Where are we, where are we going?

1:36

- Absolutely, so I mean, I think it starts with

1:38

just the awareness and the noise, right?

1:40

Like five years ago when we launched,

1:42

nobody was talking about AI

1:44

and now suddenly everything and everyone is all about AI.

1:48

And that's a good and a bad thing for vendors,

1:51

but ultimately at the end of the day,

1:53

four customers, which is most important,

1:56

there are, I would say two or three key challenges

1:59

that they're trying to solve using AI, right?

2:02

And the first of that is agent empowerment.

2:04

How do you leverage AI?

2:05

How do you leverage a GPT, something like that,

2:08

to turn your ramping agents into experienced super agents?

2:13

The second, I would say is the age old problem

2:17

of solving customer problems, ticket deflection,

2:20

whatever you wanna call it,

2:21

but ultimately how do you put AI in a customer facing way

2:25

so that your end users can get their problems answered

2:29

efficiently and quickly?

2:30

And then the last thing I'd probably throw out there

2:32

is insights.

2:33

I think AI, LLMs, data, with all of that in the hopper now,

2:38

we're actually able to make better decisions

2:41

as support teams, as customer service teams.

2:43

And I think AI is actually gonna ultimately enable that.

2:46

- And what's your take on this idea that agents

2:51

will be dead, you say, the AI will take over

2:53

and I'm sure you get that one all the time,

2:55

but where do you see that balance?

2:57

And then I wanna dive into a couple of those areas

2:59

you just talked about.

3:01

- Yeah, so no, I think it's a really interesting question

3:06

when people are like, hey, is AI gonna take over everything?

3:10

Or is the job of the agent dead, so to speak?

3:14

You know, I think it's a little bit sensationalistic,

3:19

but here's what I think.

3:20

Not only will AI radically transform customer support,

3:25

it will ultimately drive much better value for customers,

3:29

drive down costs, things like that.

3:31

But I think one thing that we forget is that

3:34

AI will also kind of create new opportunities, right?

3:37

It'll actually open up new markets.

3:39

Like for example, 100 years ago,

3:42

the concept of a software engineer didn't even exist, right?

3:47

And now it's one of the most in-demand jobs

3:50

in the tech industry or in technology.

3:53

And so I think the same way we're gonna see

3:56

prompt engineers, we're actually gonna see people like agents

3:59

who are experts at their field, experts at answering questions,

4:04

be the ones training the models,

4:06

be the ones creating the content,

4:07

be the ones actually building AI,

4:10

because in so many ways, AI is now democratizing

4:13

the concept of engineering, right?

4:15

And so I don't think the agent job will go away,

4:19

but I do think it'll transform and actually be upleveled

4:21

in so many ways by AI.

4:23

- Yeah, I love it.

4:24

Okay, well, let's dive a little bit further

4:26

into a couple of these concepts.

4:27

I love insights, agent, customers.

4:29

So a lot of people are always wondering about, you know,

4:33

self-service.

4:34

It seems like it's something that customers want to do more.

4:37

Is that something you guys are seeing in your line of work?

4:40

And if so, how is that starting to play out?

4:42

Like do all customers,

4:44

they don't want to deal with agents?

4:45

There's this balance of complexity.

4:47

How do you see that line being drawn?

4:49

- The way I see it is that the ultimate role

4:54

of customer service or customer support

4:56

is problem solving for your customers.

4:58

And that's like the end all be all goal.

5:01

Your customer has a problem whether in many ways

5:04

it's often with your own product or it's onboarding,

5:06

it's whatever, or heck, it can be about anything else.

5:09

When you think about it more broadly,

5:10

our goal in the customer support world

5:12

is to figure out who our customer is,

5:14

figure out their problems and help solve them.

5:16

And so the reason I start with that frame of reference

5:19

is because when you think about it that way,

5:21

whether you're using AI, whether you're using human agents,

5:23

whatever you want to think about, that is the end goal.

5:27

And so as a customer, you just want your problem solved.

5:30

Like nobody actually has beat with AI, right?

5:34

Like if an AI could solve your problem in three minutes

5:38

versus a human solving your problem in 10 minutes,

5:40

you would love that, right?

5:42

And if it was correct, it was empathetic and so on.

5:44

The problem though is that most chatbots

5:48

in the past decade have sucked.

5:51

They haven't been able to solve the problem, right?

5:54

And so people don't have issues with chatbots

5:57

or AI or anything like that.

5:58

They have issues when their problem is not being solved

6:00

and they're being tossed around, right?

6:01

And so everything comes back to the customer.

6:05

And so yes, they do want self-serve.

6:06

They do want to be able to solve their problems.

6:08

But look, if you have an AI that's not empathetic,

6:11

that isn't able to solve the problem,

6:12

then it's actually really important

6:14

to create that handoff path to say,

6:16

hey, look, I'm not solving your problem.

6:19

Let's get you to somebody who can

6:20

and usually that's the human agent, right?

6:22

And then that symbiosis is what enables

6:25

a beautiful customer experience.

6:26

If you can have an AI that can solve those tier one problems,

6:31

but also not get people in these doom loops,

6:33

trust me, we've seen it,

6:34

but make it very, very easy to hand off to a human agent

6:38

when you need that level of judgment,

6:40

when you need that level of care.

6:41

And I think ultimately,

6:43

our end users and customers would value that.

6:46

- Where do you see us going from here

6:48

when it comes to kind of the customer side of the house?

6:51

You know, right now to your point,

6:52

a lot of tier one issues are being resolved.

6:55

Any thoughts on kind of where it goes

6:59

in the next 12, 24 months?

7:00

- Absolutely.

7:03

I've kind of an unhinged thought on this.

7:07

The, like, I actually think we're gonna be treating AI

7:12

almost like we treat agents,

7:14

just like a different variant of agent, right?

7:17

Like we're gonna have AI agent.

7:19

We launched a technology,

7:21

we invented a technology called auto flows

7:23

in the last couple of months,

7:24

where you can basically give an agent or an AI a prompt,

7:28

a policy, the same way you'd give, you know,

7:30

GPT a prompt and it responds.

7:32

What if you could give an AI customer support agent a policy?

7:35

Here's how we handle refunds here.

7:37

Here's how we do password resets here.

7:39

The same way you'd give that, you know,

7:40

playbook to one of your ramping agents.

7:42

And then imagine giving it access,

7:44

not only just to the policies, but to all your systems,

7:47

your customer with a K, your transaction system,

7:51

whatever your database is, your snowflakes, your Shopify's.

7:55

And then letting the AI figure out how to do the rest.

7:59

Okay, I need to ask about this order ID.

8:02

I need to ask about it,

8:03

and then I'm gonna go hit the backend system

8:07

and so on and then pull out the order ID,

8:10

interpret the results and then follow up, right?

8:12

Like that is the future of self-service.

8:14

That is the future of customer facing AI

8:18

is what we call these AI agents.

8:20

And in many ways, and this is like the craziest part

8:22

is I actually think that's like baby AGI,

8:25

you know what I mean?

8:25

Artificial general intelligence in a sense.

8:27

Like we're getting to that point in the technology

8:30

where these things can become so good

8:32

and actually so powerful for end customers.

8:35

Anyway, I think that's gonna be the future

8:36

in the next couple of months.

8:37

- I love that, I love that one.

8:38

That's a great, I think next step that's probably not.

8:41

So aggressive is eliminating the rep,

8:43

but kind of taking that next level in probably complexity.

8:47

Let's flip to the agent for just a minute.

8:49

Obviously we talked a little bit about agents

8:50

being eliminated and stuff like that.

8:52

They're probably, that's a little bit abrupt.

8:55

How are you seeing AI start to play out to support agents?

9:00

Like where is it working?

9:01

Where is it not working?

9:02

- Oh man, and this is where the variety of use cases

9:05

is actually so fun, so powerful.

9:07

So the first is like everyone is using GPT

9:12

or support GPT or whatever,

9:14

right now to make their writing better,

9:17

to take in more context.

9:18

So auto completion, being able to draft responses,

9:22

especially in cases for example,

9:24

where let's say you're operating in an English language

9:27

business, but English might not be your first language,

9:29

like you might be operating out of a call center

9:31

in another country, this actually can raise the bar for that.

9:34

So right, being able to assist agents on that side.

9:37

Second, bringing up resources.

9:39

So there's this, like I would say the bulk of generative AI

9:44

use cases in production today are RAGs,

9:46

so retrieval augmented generations.

9:48

And that is just like a fancy way of saying,

9:51

take all the data, pull it in, and retrieve something

9:55

and then use that to generate an answer using an LLM.

9:59

And so a lot of companies are just doing that on public's

10:02

knowledge bases and stuff like that.

10:03

So that's the bare bones.

10:05

But when you get something really powerful

10:08

that's plugged into your data,

10:09

plugged into your conversation history in customer,

10:13

plugged into your knowledge bases internally,

10:14

your policies, imagine all of that power

10:17

at an agent's fingertips.

10:18

So when something, you know, a new question comes up,

10:20

it's able to pull up, hey, here are the past issues

10:22

from this customer, past issues from other customers

10:26

that are similar, here's your playbook on this,

10:29

and be able to pull all of that

10:30

and give these agents superpowers.

10:32

So I think, you know, answer writing,

10:35

retrieval augmented generation.

10:37

And then lastly, I would even say Q&A.

10:39

This is, I don't know, more unhinged ideas.

10:42

But like, we've been doing Q&A, or sorry, QA,

10:47

quality assurance for agents manually.

10:49

Like most companies do that, right?

10:51

You're literally walking through every single ticket

10:55

and trying to be like, was this helpful?

10:56

And you're trying to give your agent a scorecard

10:58

and they're like, yeah, please give me the feedback.

10:59

I don't know.

11:00

But imagine now, if you, as administrators

11:03

or as leaders, customer support leaders,

11:05

have an AI tool that can then sip through

11:07

all of that conversation data and say, hey, you know,

11:11

Gabe is very, very helpful, extremely empathetic.

11:16

And, you know, Dion is very good at solving the problem

11:20

but is maybe not as empathetic.

11:21

And you can kind of learn that and pull out those insights

11:24

and then use that to coach the agents back, right?

11:26

And so I think all of these use cases

11:28

are gonna be super powerful from answer generation,

11:31

retrieval, augmented generation, and search,

11:34

as well as QA.

11:35

And I could go on, but they're like, oh, that's not true.

11:38

- That's interesting, you know, the QA one.

11:40

Yeah, I don't, what a manual process that's been,

11:43

that's one I hadn't really thought too much about,

11:45

but that seems like an obvious use case

11:48

to see if you can get QA in there

11:50

and really start helping people kind of see that

11:53

a little bit different.

11:54

I like that one.

11:55

Okay, great.

11:56

So as we look to wrap here a little bit,

11:58

you know, you're obviously in this space,

12:01

so many things going on.

12:02

You've got a lot of people tuning in

12:04

that are just at the beginning and they're like,

12:06

"Ah, I don't know."

12:08

Yeah.

12:09

"What advice would you give to people

12:11

who are starting this journey,

12:13

what they're thinking about really wanting to get,

12:16

dip their toes in AI warnings,

12:18

best places to start, things to think about.

12:21

What advice would you give them?"

12:22

- Yeah.

12:24

So the first advice I would give is,

12:27

one, just start dabbling, start prototyping.

12:30

I think that's super critical, right?

12:33

Like one of the things that I think,

12:34

opening I did, that they don't get enough credit for,

12:37

is that they made the APIs dead simple to use.

12:41

Like in five lines of code,

12:43

you can get up an interesting little mini chatbot, right?

12:47

And so I think like figuring out how to use it,

12:50

start getting your engineering teams using, you know,

12:52

GPT or anthropic or whatever the models are

12:56

of your choice is the first step.

12:58

The second step I would say is,

13:00

you gotta really tease out the difference

13:03

between mission critical products and operational products.

13:07

And what I mean by that is,

13:08

let's say, you know, you are building a,

13:11

let's use like, I don't know,

13:13

a new as an example,

13:14

like a weight loss journey app or something like that.

13:18

And so for them, what's product, mission critical,

13:22

is the weight loss journey, right?

13:24

And so building GPT products may be around

13:26

suggesting meal plans or suggesting routines

13:30

and things like that can actually be really, really critical

13:33

and become a defensible moat for their own business.

13:37

For other things though,

13:39

what I would argue is there's also operational use cases

13:42

for generative AI.

13:43

For example, customer service, customer support.

13:45

For those, I would actually highly recommend

13:48

leveraging a best in class vendor,

13:50

like customer, like boardbot,

13:51

because what you're gonna find is,

13:53

there's actually so much hidden complexity in that use case.

13:56

And vendors like us are actually benefiting

14:00

and learning from all of the different businesses

14:02

that we work with, right?

14:04

And so I would actually argue,

14:06

if you have resources dedicated to generative AI,

14:10

start by figuring out which product critical use cases

14:13

you can build and which operational critical use cases

14:17

you can buy, 'cause you don't wanna build out

14:19

your own retrieval augmented generation engine

14:21

for all of your support tickets.

14:23

It's like, yeah.

14:24

And so those are the things I would consider.

14:26

So one, start dabbling to tease out the product

14:29

critical use cases and operational critical use cases.

14:32

And then I would say divert, build resources to the product

14:37

and then figure out what is your best in class partner

14:40

on a lot of the operational use cases.

14:41

- I love that.

14:43

Do you feel like there's anything people should be careful about?

14:45

You know, sometimes you hear about this idea

14:48

that you're gonna maybe get with false information

14:50

or any warnings that you'd put on this

14:53

as people go down this path

14:55

or kind of maybe gotchas they should be expecting.

14:59

- Yes.

14:59

So the first thing I would say is security and safety.

15:04

Like that comes up all the time, right?

15:07

And so you're gonna wanna go with a vendor

15:08

or again, if you're building internally

15:10

for your product critical paths,

15:12

you're gonna wanna make sure, you know,

15:14

you've got GDPR compliance, you've got SOC2 compliance

15:17

and understanding how you send data to vendors

15:21

and how that gets trained, how that gets used,

15:24

all of that's really important.

15:25

So for example, us at Forthought,

15:27

one of the things we did from day one was we invested

15:29

in technology to automatically redact PII, right?

15:32

And so, you know, so basically anything that gets sent

15:35

to our servers, you can automatically redact them

15:38

and then like with 99.99% accuracy,

15:42

you know that nothing getting shipped

15:44

to any of these large language models

15:47

contains PII because the bots don't need it.

15:49

You don't need to know that my name is on

15:51

in order to answer the question.

15:53

You can replace that with a token that says,

15:54

my name is, you know, name_one, and as long as it's,

15:58

you know, homomorphic, so to speak,

15:59

and you can actually perform the mapping or whatever,

16:02

then the algorithms will perform almost identically

16:06

on those, right?

16:07

So being able to invest in those security measures,

16:11

open AI and others have like zero data retention powers

16:16

that you can sign up for, so they won't retain your data

16:17

or train on it.

16:18

So there's like a lot of these considerations.

16:20

It can be done, obviously, again,

16:22

if you're building information critical stuff,

16:23

you've got to do all of this yourself.

16:25

If you're leveraging a best-in-class vendor,

16:27

you should make sure and ask your vendor,

16:29

are they doing all this stuff?

16:31

- Very, very helpful, 'cause I think that's very timely.

16:35

Okay, as we wrap then, if someone wants to know

16:38

a little bit more about Forthought or dive a little bit

16:42

deeper with you guys, what's the best way to do that?

16:44

- Yeah, easiest way is to connect on our website,

16:47

so www.forthought.ai, request a demo of the product.

16:52

It's a lot of fun to use, and I mentioned autoflows

16:55

and a few other things there.

16:57

If you wanna get in touch with me directly,

16:59

I'm on all the social channels, so find me on LinkedIn,

17:02

I'm Dion Nicholas, find me on Twitter or X,

17:06

I guess they call it now @dogidon, D-O-J-I-D-E-O-N,

17:10

and just mention customer, and I'd always be excited

17:14

to jam.

17:15

- Awesome, awesome, well, I appreciate it.

17:16

Great session, very timely, I think,

17:19

as we dive into this topic of the AICX,

17:21

so Dion again, thanks so much for your time,

17:23

and for everybody else, enjoy your afternoon.

17:27

- Thanks, Gabe.

17:27

(upbeat music)

17:30

(upbeat music)

17:33

(upbeat music)

17:35

(upbeat music)

17:58

(upbeat music)

18:02

(upbeat music)

18:12

(upbeat music)

18:17

(upbeat music)

18:25

(upbeat music)

18:35

(upbeat music)

18:40

(upbeat music)

18:47

(upbeat music)

18:52

(upbeat music)

19:02

(upbeat music)

19:07

(upbeat music)

19:12

(upbeat music)

19:19

(upbeat music)

19:29

(upbeat music)

19:34

(upbeat music)

19:39

(upbeat music)

19:46

(upbeat music)

19:56

(upbeat music)

20:01

(upbeat music)

20:06

(upbeat music)

20:13

(upbeat music)

20:23

(upbeat music)

20:28

(upbeat music)

20:33

(upbeat music)

20:40

(upbeat music)

20:55

(upbeat music)

21:00

(upbeat music)

21:07

(upbeat music)

21:22

(upbeat music)

21:27

(upbeat music)

21:34

(upbeat music)

21:53

(upbeat music)

22:00

(upbeat music)

22:19

(upbeat music)

22:26

(upbeat music)

22:46

(upbeat music)

23:06

(upbeat music)

23:26

(upbeat music)

23:46

(upbeat music)

24:06

(upbeat music)

24:26

(upbeat music)

24:46

(upbeat music)

25:06

(upbeat music)

25:26

(upbeat music)

25:46

(upbeat music)

26:06

(upbeat music)

26:26

(upbeat music)

26:46

(upbeat music)

27:06

(upbeat music)

27:26

(upbeat music)

27:46

(upbeat music)

28:06

(upbeat music)

28:26

(upbeat music)

28:46

(upbeat music)

29:06

(upbeat music)

29:26

(upbeat music)

29:46

(upbeat music)

30:06

(upbeat music)

30:26

(upbeat music)

30:46

(upbeat music)

31:06

(upbeat music)

31:26

(upbeat music)

31:46

(upbeat music)

32:06

(upbeat music)

32:26

(upbeat music)

32:46

(upbeat music)

33:06

(upbeat music)

33:26

(upbeat music)

33:46

(upbeat music)

34:06

(upbeat music)

34:26

(upbeat music)

34:46

(upbeat music)

35:06

(upbeat music)

35:26

(upbeat music)

35:46

(upbeat music)

36:06

(upbeat music)

36:26

(upbeat music)

36:46

(upbeat music)

37:06

(upbeat music)

37:26

(upbeat music)

37:46

(upbeat music)

38:06

(upbeat music)

38:26

(upbeat music)

38:46

(upbeat music)

39:06

(upbeat music)

39:26

(upbeat music)

39:46

(upbeat music)

40:06

(upbeat music)

40:26

(upbeat music)

40:46

(upbeat music)

41:06

(upbeat music)

41:26

(upbeat music)

41:46

(upbeat music)

42:06

(upbeat music)

42:26

(clanging)