From Salesforce Einstein to Conversational Mastery

0:00

(upbeat music)

0:02

- And welcome to our next session with my Duke.

0:09

My Duke from QIN.AI.

0:11

My Duke, how are you doing today?

0:13

- I'm doing great.

0:14

How are you doing?

0:15

I'm just looking forward to the holidays.

0:17

- Oh, I am too.

0:18

I am too.

0:20

It's cool just to do this during the,

0:21

during like right before everything's happening

0:23

after Black Friday.

0:25

But I will say I am excited to talk to you.

0:28

I think you had quite a journey.

0:30

I'm very curious as to ask you questions on that.

0:33

And then get into the details of your,

0:35

of your new company that you founded with your co-founder.

0:37

But first, I want to call out for the audience.

0:41

I know that you were part of Salesforce for,

0:44

I think it's over over four years

0:47

working on Salesforce Einstein.

0:48

Do I have that right?

0:49

- Absolutely right.

0:50

Yeah, almost four and a half, five years.

0:52

Yeah, even I forget how long it was.

0:54

- So can you, can you talk us through just that journey

0:58

and that experience for you?

1:00

And then obviously how that integrated

1:02

and led you to founding Q and that AI.

1:06

- Yeah, absolutely.

1:07

So I was on the product side on Salesforce Einstein.

1:10

So this is when the, by the way,

1:12

Einstein is like Salesforce's AI umbrella.

1:15

It's a branding and--

1:16

- Thank you for clarifying that.

1:17

- Yeah, under then there are many AI products.

1:19

So I was a part of the platform.

1:21

And so what it really means is, you know,

1:24

it Salesforce has all of these clouds like,

1:26

yeah, the service cloud,

1:27

they have sales cloud, marketing and commerce.

1:30

To me it feels like each of them are very big companies

1:32

within this huge giant company called Salesforce.

1:35

And each of them, and we're talking about

1:38

like five or six years back, right?

1:40

You know, and you know, they're, they were still AI then.

1:43

There was no large language models in general,

1:45

but AI was very prevalent.

1:47

And there was all of these opportunities

1:48

to build AI into enterprise applications.

1:51

And so, and if you think about many of the underlying

1:55

technologies that being built are kind of common across,

1:58

whether it's a sales use case or a marketing use case

2:00

or a service use case.

2:01

And the, and the theme of this platform was,

2:04

can you build those underlying building blocks

2:06

which will power sales, marketing and so on and so forth.

2:09

So that's the platform that I was working on.

2:12

What is most interesting and I know I talk about this a lot,

2:16

I almost felt like an investor because you kind of see

2:19

this wide variety of use cases across all of this different,

2:23

you know, verticals.

2:25

And the more important thing, you also see which of them

2:27

are experimental versus which one actually gets mass adoption.

2:31

So you kind of start sharpening that instinct.

2:33

Okay, these are just experimental.

2:35

It's a cool demo.

2:36

It's great.

2:37

It's fascinating.

2:38

But you know, there won't be a lot of enterprise traction

2:41

into it.

2:42

So that's, that's.

2:43

- And you're talking internally, like, well,

2:44

looking at the data and the clients that are utilizing the

2:48

- Exactly.

2:49

Because, you know, because we see that the sales was a lot

2:51

of customers and some of them are, you know,

2:54

would say not innovators, right?

2:56

And, you know, they adopt something when it's quite mature.

2:59

So you get to see which of those are mature,

3:01

which of them are making impact versus which ones are not.

3:04

And so that was kind of fascinating as a part of that

3:06

platform.

3:07

And we were mostly working on something called automated

3:10

machine learning.

3:11

So like, how do you automate the machine learning for,

3:14

you know, all of these different plows,

3:16

all of these different use cases?

3:17

For instance, in sales, leads scoring,

3:19

but you have all of these leads,

3:21

how do you know which leads are better

3:23

and are more likely to convert in an opportunity?

3:26

If you think about service, like,

3:28

there are common use cases, obviously bots is there,

3:30

but there are also use cases around

3:31

you have tickets and cases.

3:32

Like, can you classify those things based on priority?

3:35

Who should it be routed to, you know,

3:37

you know, in marketing, you can think about recommendations

3:40

related use cases.

3:42

And my second half, I kind of moved more

3:45

into conversational AI.

3:47

So I was heading our Einstein bots.

3:49

I was a part of the service cloud.

3:50

So I was working on conversational AI.

3:52

And also had influence on our agent assist

3:56

as well as some of the voice programs.

3:58

So that's kind of what my journey at Salesforce was.

4:02

Also got lucky enough to give one of the keynotes,

4:05

Einstein keynotes in Salesforce's

4:07

annual dreamforce conference.

4:09

Very different experience, a lot of stress,

4:12

as you can imagine, behind what goes on,

4:14

but it was really fascinating to be able to talk

4:16

about some of the things our team was building

4:18

to the community.

4:20

- I'm sure.

4:21

And I actually came across some of those

4:23

and I tell the audience, go ahead and search that as well too.

4:25

Really interesting to see.

4:27

So what was the, for you, what was the then moment

4:30

of like, okay, I wanna, I wanna, like, what did you see?

4:33

Maybe you didn't, but did you see something

4:35

at when you're at Salesforce, specifically at Einstein,

4:38

that led you to the creation of your own company?

4:41

- Yeah, absolutely.

4:42

I think there was, there was definitely seeds planted.

4:46

By the way, you know, as someone who at one time

4:49

wants to be an entrepreneur, you know,

4:51

you're always looking for things for a while, you know,

4:53

and just, and not always the timing is perfect.

4:58

But working on the conversational AI side,

5:01

and this was both pre-pandemic and during the pandemic

5:04

and post-pandemic.

5:05

So I saw that whole journey.

5:07

You know, conversation workflows was quite important.

5:12

Like what it means is like, whether virtual agents

5:15

are bots or even live chat with human agents

5:17

or even voice calls.

5:19

And more and more of, you know, the Salesforce's customers

5:23

were actually investing in conversational workflows.

5:27

Definitely a lot in bots, especially during the pandemic

5:29

because of the obvious reason.

5:31

And one of the things that I noticed is

5:33

there are a lot of tools out there

5:35

which helps you to build these workflows.

5:38

But when it comes to evaluating these workflows

5:42

or, you know, doing analytics on top of it,

5:44

and I'm not talking about high-level analytics,

5:46

really understanding deep what's happening

5:48

in the conversations where things are breaking,

5:50

where users are getting frustrated,

5:52

there were not a lot of tools there.

5:54

And the problem itself was quite difficult.

5:57

And so that was one of the key insights that I had.

6:01

Like, you know, the observability, the mining,

6:03

how do you evaluate was quite important.

6:05

I think in a more succinct way

6:07

or a more emotional way to put it is like all of the CX teams

6:10

they are building all of this workflows,

6:11

they're building, designing their bots

6:13

or they're, you know, designing their knowledge bases.

6:16

But many times they didn't have confidence

6:18

in what they're building.

6:19

So how do we help them to gain,

6:21

hey, what your build works?

6:23

It helps customers, they make them happy

6:26

to solve their problem.

6:27

So that may be in one way what we're trying to solve.

6:30

How do we make the CX teams feel more confident

6:33

in what they're building?

6:34

- Yeah.

6:35

And at least for myself and the audience,

6:38

can you give maybe an in-depth definition

6:42

of what you talk about when you say workflows?

6:46

Because you talked about the chat and I get that,

6:48

but is there an overarching understanding

6:50

when you're talking about workflows for a country?

6:52

- Yeah. So any of these like,

6:54

if you think about a chat bot, like let's say,

6:57

if you talk about a commerce scenario,

6:59

someone says that, hey, they want to return something.

7:01

Like what are the steps involved to actually do the return

7:05

and how is it shaped in your conversations?

7:09

Maybe they're asking for the email address,

7:11

maybe they're asking for the order number

7:13

and maybe they ask which country they're from

7:15

and where they're bought it from.

7:17

And based upon that, the steps to return might be different.

7:20

Some cases you might have to go into a store

7:23

and some cases they might be able to do a return online.

7:26

Similarly, there can be for various different use cases,

7:30

the workflows can be also on the live chat side.

7:32

Like when it goes to human agent, how are they solving it?

7:35

Let's say we take a use case of a financial services.

7:39

Maybe they have an additional charge on the credit card

7:41

that they don't recognize.

7:42

So how do the live human agent solve that problem?

7:46

Again, this whole thing is happening not through a UI,

7:48

it's happening back and forth

7:50

through a conversation or a phone call.

7:51

- And I want to just ask you one-off question on this

7:57

to get your opinion.

7:59

As you're talking about this,

8:00

how I try to identify this a little bit is on the phone.

8:05

And specifically, like we've had now for years and years

8:08

and years, IBR's Interactive Forest Response System.

8:12

- Yes.

8:13

- And there's workflows on that.

8:16

And I recall a while ago in my career building

8:20

those as well too.

8:22

Is this brand new territory or can you analyze,

8:27

were people analyzing IBRs and now there's just no way

8:30

to understand this or is this two completely different things

8:33

that we're talking about?

8:34

- It's related but there has been a shift.

8:37

Like, of course there are ways to analyze IBRs

8:40

but more and more conversations are going,

8:44

there's voice bots which are not IBR driven.

8:46

Also the actual human conversations like,

8:49

sure you go through the steps maybe

8:51

and then you're actually going to a human.

8:53

And how is that conversation planning out?

8:56

Are, how are the humans interacting to address their queries?

8:59

Are they interacting well?

9:01

Are they even understanding?

9:02

In some cases it might be that the industry is shifting

9:07

or the products are shifting

9:08

and the humans don't have the right knowledge

9:10

of their fingertips to be able to resolve it.

9:13

And how do you figure out which of those areas

9:15

where they're not able to resolve?

9:17

Also underlying there is so much of wealth of product knowledge

9:20

that you can leverage.

9:22

For instance, like maybe going back to the example of return,

9:25

like you understand people are coming to return stuff

9:28

but what every company wants to know why.

9:30

So what is the root cause underneath it?

9:32

It's like, and some of them are kind of surprising.

9:34

It's not always that they are coming for return.

9:37

You would think that, oh, sizing issue for e-commerce

9:40

or something like that.

9:40

But sometimes people are coming,

9:43

asking questions about return

9:44

because the return policy is not clear.

9:47

Or maybe the return slip was not included in it.

9:51

Or maybe they went to the store,

9:52

they didn't get the return.

9:53

So it's not about just returning the product

9:56

but also the process and where some of those bottlenecks are.

9:59

And you can imagine there are hundreds and thousands

10:01

of conversations happening across,

10:04

sometimes channels, different modalities, maybe voice, chat,

10:07

and bot.

10:08

How do you aggregate all of those and see

10:11

what are the top root causes around the returns

10:13

where my company can focus on to improve the CX?

10:16

And that is what I'm going to argue for.

10:18

- I think that's fascinating because I think right now

10:20

if you look at just the market,

10:22

building and creating, let's just use the example of bots,

10:27

seems to be table stakes at this point.

10:29

And it's really easy to do that now.

10:31

But the talk of what you're mentioning of iterating

10:36

and how to iterate isn't talked about much at all.

10:39

It's just about getting it up there,

10:40

but not like, all right, well, how do you improve on that?

10:43

So walk us through now, qin.ai.

10:48

So I know you gave kind of the premise

10:51

and the theory behind it.

10:52

Like, talk to us about the platform itself,

10:54

what have you built and what is it

10:57

that you sell to?

10:59

I know we talked about mid-market enterprise companies.

11:02

- Yeah, so we are building a co-pilot

11:04

for customer experience teams,

11:06

mainly focusing on unifying the support data

11:10

across modalities channels or even vendors.

11:12

Some of these companies are using different vendors

11:15

and bringing them all together

11:17

and providing, they have lots of questions

11:20

about that conversational data that's going on.

11:23

And we analyze and provide answer to those questions.

11:26

And on the main problems,

11:28

the three key problems that we're helping to solve

11:30

is help them scale their CX so that it does better.

11:35

And some of the problems that they have faced

11:37

is like multiple channels and conversations that's happening.

11:41

How do they connect them together?

11:43

A second is larger volume of conversations.

11:45

Like, just imagine even listening to one call recording

11:48

or reading through a chat.

11:49

It's very hard, takes time consuming.

11:52

It's also, I've done it is like, after you do a few,

11:55

you feel like, I don't want to do this job.

11:57

Like, who's going to go read through all of these things?

11:59

It's just, it's very, very hard.

12:02

And so when you have hundreds and thousands

12:05

and sometimes millions of those conversations,

12:07

how do you analyze that?

12:08

Most companies do a sample from time to time.

12:11

It's not very comprehensive.

12:13

And the more, and another thing that's coming up is,

12:15

how can you do that in real time?

12:16

For instance, like, we just talked about Black Friday.

12:19

And we have scenarios where certain coupons

12:24

or discounts were given out.

12:25

But when the customers were trying to add

12:28

that coupon code in the cart, it somehow didn't work.

12:31

And, you know, and so there were two,

12:34

this is a real example.

12:36

And there was a spike in customer support requests

12:40

around they were trying to get price adjustments

12:42

because the coupon code didn't work.

12:44

Or many of them are trying to cancel their order.

12:46

And you want to know this in real time

12:49

because every hour you wait, you're actually losing revenue.

12:53

It's not even customer support, you're losing revenue.

12:55

So the sooner you come to know that there is something broken,

12:57

the sooner you would be able to fix.

12:59

So it's more like real time trends and anomalies

13:01

is another thing that we are looking into

13:04

and offering to our customers.

13:06

- So then I assume that a lot of questions

13:10

that companies need to ask,

13:13

like are there, like are there tablets?

13:16

Are there a common question, right?

13:18

Like you get to a point of, all right,

13:20

this company, when you're looking at all the data,

13:23

do you need to ask this, this, and this?

13:24

Or is every company unique from your perspective?

13:28

And how do you help out that, right?

13:30

Because it's like now you got access all the data,

13:33

you're saying to me that now you can ask questions

13:35

about anything, about products, about discounts.

13:39

Like where do you start?

13:41

Is I guess my question to you?

13:44

- Yeah, I think it's a very nuanced question.

13:46

It's a great question.

13:48

In certain industries, there are overlaps

13:52

around the kinds of questions that are asked.

13:55

And in certain industries, there is less and less overlaps,

13:58

more snowflake in nature, because their business is very unique.

14:02

And so, but what is interesting is the underlying technology

14:08

to understand language and what users are asking.

14:11

So we don't have any notion of, our AI model doesn't know

14:15

that this is an e-commerce return use case

14:17

versus this is a financial services,

14:20

and a credit card chart that they wanna refund.

14:22

It just looks into those conversations

14:25

and unsupervised identifies those intense inquiries

14:28

that pops up.

14:29

And it's really a factor of the volume,

14:32

if more and more volume of conversation

14:35

are asking that specific query or intent would bubble up.

14:38

So that taxonomy is almost like iteratively is built

14:42

and doesn't need to come from the customer themselves.

14:46

Having said that if they already have a taxonomy built

14:49

and they can influence that.

14:50

So there's ways to see that model.

14:52

So that's why we are kind of excited

14:54

because the technology works.

14:55

So we have customers in e-commerce, in health and benefits

15:00

and travel and hospitality and financial services.

15:03

And we're not doing extra additional work.

15:06

It's the technology automatically figures out

15:08

and are able to answer the question.

15:10

- Yeah, it does.

15:13

I guess like just to go one step deeper into that

15:15

for the audience, is it that from your opinion

15:18

working with your clients so far is that they have been

15:22

trying to figure out these answers about,

15:26

what are the discounts being used

15:27

or how effective is our chatbot.

15:29

And now you have provided the technology

15:31

that for them to do that.

15:32

Or is the kind of technology there first

15:35

and you're trying to educate the clients of like,

15:37

okay, here are some questions now

15:40

that you can start asking your business.

15:43

- I love that question.

15:44

It's actually both.

15:44

So they wanted to know like,

15:48

so this is going back to Salesforce.

15:50

They implemented the chatbot.

15:51

They knew first few five use case,

15:53

almost every company knows what are the top five

15:55

or six use cases.

15:57

But they wanna know how it's doing.

15:58

Is it breaking apart?

16:00

Like even even those use cases,

16:02

they're deviations.

16:02

Like maybe the bot gave an answer,

16:04

but it didn't satisfy the customer.

16:07

And they wanna know what is the percentage of cases

16:09

where the customer was not satisfied.

16:11

And when, and generally what happens

16:13

is they get escalated to a human agent.

16:14

So how did the human agent satisfy it?

16:16

And can we learn from that diff?

16:18

Like what the bot said and what the agents said

16:20

and inform or improve the design of the bot.

16:22

They always wanted to know it.

16:24

I don't think the technology was there

16:26

to do that very well,

16:27

especially deep into the conversations.

16:30

So they did more high level.

16:33

Look at these are the areas where the answers

16:35

are not given, they know that,

16:36

but they don't know how to improve.

16:38

What's the next step?

16:39

What's the action to take?

16:41

The second part is because of a large language model

16:44

and the advancement in technology to understand

16:48

some of the use cases,

16:49

which was very hard and has to be done manually,

16:52

can now be automated.

16:54

That's so there's new opportunities there.

16:56

Like this root cause analysis that I talked about,

16:58

like, I mean, it's quite interesting

17:01

because sometimes the root cause

17:02

of why the customer reached out to,

17:04

you cannot analyze that by listening to the customer.

17:08

You have to listen to the solution

17:09

because the agent have more insights.

17:11

They're looking into their account,

17:12

they're finding some issues and said,

17:13

oh, this is the problem.

17:15

And then we see that problem is repeating.

17:17

And from the agent sites,

17:19

you have to also look at the agent site

17:21

and understand that and it can be anywhere in the conversation,

17:23

not in the first, not in the middle.

17:25

It can be multiple user utterances

17:27

to be able to understand the entire conversation

17:29

and surface that that's the technology

17:31

which didn't even exist before.

17:33

And related to that, you know,

17:35

another common thing is QA of agent interaction.

17:37

That's a very common stuff in the industry.

17:39

And almost every company has some kind of a rubric,

17:42

you know, for how they actually do that.

17:44

And it's very manual.

17:45

They have rubrics like if the agent showed empathy,

17:48

did they personalize the conversation,

17:51

did they provide next steps in the conversation,

17:53

did they read what the user said before

17:56

and then didn't force the user to restate that.

17:59

I mean, you just have to look into the conversation

18:01

and score in each of these factors.

18:03

And there was no AI to do that before, you know,

18:06

before like generative AI and large language models.

18:09

So some of those use cases

18:10

and now becoming more prevalent can be done more easily.

18:13

So I would say the answer is both things that was there.

18:16

They wanted to know it was not possible.

18:18

And there are some of these things,

18:19

newer areas now which can be done

18:21

which was not being able to be done before.

18:23

- So I want to go one level deeper

18:29

and get a specific from you, which is if let's say,

18:33

and we don't have to use an actual client of yours,

18:35

but like direct to consumer businesses,

18:38

after Black Friday, let's say they have a large collection

18:43

of data, phone, email, chat,

18:46

and they're using QN.AI, you know,

18:50

but what would be one or two examples of questions

18:53

that they should be asking right now

18:55

if they had all the access to the data and the access to the QN.

18:58

- Yeah, I think some of the questions

19:00

they should be asking is like,

19:01

I mean, if you're talking about the CX

19:03

and the customer support

19:04

and they have some kind of a bot in place,

19:07

how is the bot able to self-serve and what percentage

19:11

and the cases where they are not self-serve,

19:14

what are those topics, other newer topics

19:16

where it's not able to self-serve.

19:18

And when they're not doing it,

19:21

how are the human age and self-serving that

19:23

so that it can inform the design of the bot

19:26

or optimize the bot design.

19:27

So in the future when the next Black Friday comes

19:30

or throughout the year, it's able to self-serve more of those.

19:34

So that's kind of another thing.

19:37

The other thing which I talked about,

19:38

especially in Black Friday,

19:39

understanding this trending stuff and anomalies that's happening

19:43

and trying to being able to identify it sooner,

19:45

I think would be quite critical.

19:47

And that's what they're trying to do.

19:49

Like let's say I gave the example of the coupon code

19:51

which is very common during holidays

19:53

and it didn't work.

19:54

And as a result of which the customers were complaining

19:57

about it or canceling their order even worse

19:59

and they wanna know it sooner.

20:01

And that's the kind of question,

20:02

what are the fiction in their buying process

20:05

and can they know it immediately in real time

20:08

so they can address that?

20:09

- I love that and thanks for that.

20:10

I think that's really insightful.

20:12

What, I have to give you another broad question

20:16

but it's because we're not specific

20:18

but it's like in your opinion,

20:21

what should a company optimize for

20:24

in your example with self-serve, right?

20:26

Like it should a company be optimizing for like,

20:30

okay, yeah, if we weren't able to self-serve 40%

20:34

of those questions, how do we get that to 20%?

20:36

Is it okay to be at 50, 50% getting the agent involved?

20:40

I guess at the end of the day, like is it, what's the answer?

20:45

And I know it's not every company is the same here

20:48

but I always like to ask the question,

20:50

like what are we optimizing for?

20:53

And how would you respond to that?

20:54

- Yeah, I think, I think a good way to think about it.

20:58

Like we should put ourselves in the,

21:00

I mean, we're all consumers at the end of the day.

21:01

Like it's not that we don't want to interact with the bot

21:05

or we always love to interact with human

21:07

and it goes vice versa.

21:09

It's like we want to get, we have a problem,

21:12

customers are a problem and we want to get resolved.

21:14

If the bot is able to resolve it faster, quicker

21:18

without me waiting for 10 minutes on a call,

21:20

I would love that.

21:21

But there are some complex topics, right?

21:23

Which I know a virtual agent cannot solve it

21:26

or at least not today.

21:28

For that, don't make me go through that process.

21:31

I would like to talk to a human

21:32

who can actually solve some of those complex stuff.

21:35

So I don't think the goal is like,

21:37

I don't think it will ever be like,

21:39

you know, 100 person self service.

21:41

There are cases where the humans are involved.

21:44

So, but the easier ones and some of the more nuanced ones

21:48

can be solved by the bot

21:49

and the more complex and difficult ones

21:51

can go to the human.

21:52

But again, when it goes to the human,

21:54

do the humans have, that human agents have

21:56

the right tools of their finger step?

21:58

Like for example, one of the companies

21:59

that we are working with, they're in the benefits management,

22:02

during the open enrollment time,

22:04

they have like huge, you know,

22:06

guides around benefits and people asking about the deductibles

22:09

and all of those complex scenarios.

22:11

And it would be great for the human to, you know,

22:13

ask this question to that knowledge base

22:15

and get the answer quickly to be able to support.

22:17

So that's also part of the optimization

22:19

for the human agents.

22:21

One other thing I wanted to say,

22:22

which is I see there is, it's catching up more,

22:25

is this concept of an angel concept

22:29

of customer satisfaction,

22:30

but it's more like a real-term inferred

22:33

customer satisfaction because the problem

22:35

with the angel method is like a survey based,

22:39

it happens later on, it has low take rates, it's biased,

22:42

and there's so many other things.

22:44

But the most annoying part is like,

22:46

if you read the customer conversation,

22:47

you would have a very good guess how satisfied they were.

22:51

But yet we do this annoying thing of asking later

22:54

on like, how did it go?

22:55

So the idea is, can we just go to the conversation,

22:58

take the, see the steps that the bot or the human agent did,

23:01

and kind of the conversation they had to infer

23:04

what the customer satisfaction is,

23:05

which is a leading indicator.

23:07

And the reason I brought it up,

23:09

that has to be balanced with the self-serve.

23:11

Like there are ways you can self-serve like 100 person,

23:14

you can just tell the customer to go away,

23:17

but your CSEL would be really done.

23:18

I'm talking about an extreme case,

23:19

but you have to balance the self-serve

23:22

as well as the customer satisfaction that's happening.

23:25

Because if you drop the customer satisfaction,

23:27

I don't think that's the greatest.

23:28

So that's another very, very interesting thing.

23:31

And it's actually even more important

23:32

with large language models to build that card rails,

23:34

because the bot can go off the rails.

23:37

And you want a real-time interception

23:39

to understand the customer satisfaction

23:41

is actually declining.

23:43

And maybe it's a good time to stop the bot

23:45

and go to a human agent.

23:47

- Yeah, I love your discussion

23:50

about the real-time CSAS scores.

23:52

And where I'm drawing the line,

23:54

and maybe the audience, that is a B2B SAS,

23:57

you can understand as well too,

23:58

which is that if you think of a tool like GONG,

24:01

for instance, and GONG's now at a point of,

24:04

it can tell you a certain prospect or deal,

24:07

the likelihood of a closing based on all the data and inputs.

24:11

And it's really that same concept of like,

24:13

okay, in real time, this customer,

24:15

what are the likelihood that they're like

24:16

actually 100% happier, actually,

24:18

they're never by again in real time.

24:20

I think that's fascinating.

24:21

Maybe one other kind of area and topic I'd like to ask you

24:25

about is, again, this data collection,

24:29

because all of this is based on every single customer

24:33

touch point that you have,

24:34

whether that be on the phone,

24:35

whether that be through the chat,

24:37

you can just website visits,

24:39

whether they're clicking out

24:39

and really understanding everything there is

24:41

about the customer.

24:42

I guess from your point of view,

24:44

has there been a shift in that?

24:47

Like in the last few, maybe just year, I'd say,

24:49

or even with the founding of QN.ai,

24:52

where are companies thinking of data

24:55

differently now than before?

24:57

And how important is that now moving forward?

25:00

- I would think, I mean, I don't know if you would agree.

25:03

I mean, more and more companies and execs and decision makers

25:07

are realizing data is more and more important.

25:10

And being able to collect and store

25:12

and have a clean data is quite crucial.

25:15

Also, other things that we are also seeing is there are more,

25:21

depending upon the business,

25:22

because we have some clients in different

25:24

kind of verticals and domains,

25:26

the security and the privacy of the data

25:29

because there are consumer rights involved.

25:31

So there are things happening there

25:34

where you need to make sure that when you're analyzing,

25:37

the PII and PHI is de-identified there.

25:41

And one subtle difference,

25:43

I'm not talking about reduction,

25:44

reduction is removing it.

25:46

Like it's because that hinders understanding

25:48

the conversation is de-identified.

25:50

We don't need to know what the email is.

25:51

We just need to know there was an email there.

25:53

That's good enough.

25:54

And that part is quite critical,

25:57

I think, especially in some of the financial services

25:59

or healthcare domains,

26:01

because I don't know if you have ever come across a bot

26:04

where it says, don't share your credit card information.

26:06

You know, they just give this disclaimer

26:08

and just hope that no one shares it,

26:10

but someone might share it.

26:11

And so being able to protect that,

26:14

I think is quite important as well, I would say.

26:18

The other thing is being able to,

26:23

you know, this is also a very common problem.

26:25

I think there are CDPs built around this

26:27

when you have these conversations

26:28

across different channels, especially vendors.

26:30

Sometimes it's actually quite often

26:32

it's very hard to connect them together.

26:34

So I think that realization is becoming more and more important.

26:38

Like is there some way, there's some ID

26:40

which can connect them,

26:42

the more you can do that,

26:43

the more you'll be able to stitch them together

26:46

and be able to analyze.

26:47

So I think that's also quite important.

26:49

- Yeah, I would agree with that.

26:51

I guess the last question I have for you,

26:54

it's more about looking,

26:56

it's hard to say looking forward

26:57

because I feel like we have so much

27:00

in the present right now,

27:01

but maybe just looking into next year,

27:03

which is less than a month away.

27:06

You know, what would you be excited about

27:09

specifically at qn.ai

27:12

that you can share with us?

27:15

- Yeah, I think some of these things I talked about

27:18

is definitely not done.

27:19

We are constantly iterating on it.

27:22

This anomaly and trend piece I talked about,

27:24

it's a very complex problem.

27:27

It's almost like, you know, Google trends, right?

27:29

You go to Google trends, you search for a thing,

27:30

you see how it looks.

27:31

Imagine the same thing for your customer support data,

27:34

right there at your fingertips

27:35

and you can see everything.

27:37

So we have made some progress there,

27:39

but there's so much more to do there.

27:41

Similar thing when I was talking about this,

27:43

understanding and analyzing conversations,

27:45

I'll give another analogy, maps analogy, right?

27:48

So think about when many of these conversations

27:51

that design, you know, let's say the user wants

27:53

to get a solution to a problem.

27:55

So the analogy is they want to go to path A to B

27:58

and the world that is designed things,

28:00

the map looks like there's one straight path from A to B

28:03

and that's what the map is all about.

28:04

But maps are not like that.

28:06

You know, users are taking so many different paths

28:09

to get from point A to B

28:10

and they're a bottleneck or higher traffic

28:12

and some of those paths.

28:13

So that analogy to the world of conversation and CX,

28:17

what are those different paths taken,

28:18

where are those bottoms being able to surface

28:20

that visualization, I think also key

28:23

to be able to solve some of this problem.

28:26

The last thing I would say we are working on it

28:28

is kind of a natural language BI.

28:31

It's like a chat typically for all your customer support data.

28:35

So, but it's very different.

28:37

It's not like, you know, you put all of those conversation data,

28:39

you ask the question and you magically get the answer.

28:42

It doesn't happen.

28:44

You have to do some of those analysis before

28:47

and then you do, you know, natural language to SQL

28:50

and you start getting those response

28:51

but it becomes very powerful because you can just go into this

28:54

and say, tell me what are the top contact patterns

28:57

and you fix one, tell me what are the top root causes around it

29:00

and then you can say, okay,

29:01

how are the agents giving responses to that?

29:03

Do you see this workflow right there

29:06

without going through dashboards and dashboards

29:08

and creating custom and, you know,

29:10

so that natural language BI can be another very interesting thing

29:13

that we are working through.

29:15

It's not straightforward as it sounds or, you know,

29:18

oh, you just put chat jippity for enterprise and it works.

29:20

It doesn't work that way.

29:22

But that's another area that we are looking into.

29:24

Probably there's some opportunities in the future.

29:26

- Yeah, super interesting.

29:28

And I think for those listening that are in operations,

29:33

there's a lot to look forward to and a lot of,

29:36

I'll be honest, a lot of responsibility come in their way

29:39

from like analyzing data that they could never do that before.

29:42

So that's really interesting and I feel like you're providing

29:45

so much more, again, opportunity to shift and change

29:49

in real time to customer support organizations.

29:53

So that's awesome.

29:54

Bye, Yuka.

29:55

Bye, Yuka.

29:56

I'm sorry.

29:56

Again, thank you so much for your time.

29:59

I wish you the best in 2024,

30:02

beyond for Q and dot AI for those listening.

30:04

Please check out what they have to offer.

30:07

And that's it.

30:09

Again, appreciate your time.

30:11

- Thank you so much Brian and for the audience.

30:13

Like we have integration with customer.

30:15

It's on the customer website live.

30:16

So do check us out.

30:18

- And what else do you have integrations with?

30:21

I just wanna make sure we're showing on by everyone here too.

30:24

- So we integrate with a bunch of vendors from Salesforce,

30:28

to Genesis, to Amazon.

30:30

There's a whole list.

30:32

So because that's one of the key things.

30:34

Like we made the integration quite easy

30:37

and once you do that,

30:38

and many customers use different vendors

30:39

and we were kind of agnostic.

30:42

So yeah, so it's also on our website.

30:43

If you go to Q and dot AI,

30:44

you will see all of those integrations as well.

30:46

- Amazing.

30:47

Thank you for pointing that out.

30:49

And thanks all.

30:50

(upbeat music)

30:52

(upbeat music)

30:55

(upbeat music)

30:57

(upbeat music)

31:20

(upbeat music)

31:24

(upbeat music)

31:34

(upbeat music)

31:39

(upbeat music)

31:44

(upbeat music)

31:51

(upbeat music)

32:01

(upbeat music)

32:10

(upbeat music)

32:17

(upbeat music)

32:27

(upbeat music)

32:32

(upbeat music)

32:37

(upbeat music)

32:44

(upbeat music)

32:54

(upbeat music)

32:59

(upbeat music)

33:04

(upbeat music)

33:11

(upbeat music)

33:21

(upbeat music)

33:26

(upbeat music)

33:31

(upbeat music)

33:38

(upbeat music)

33:48

(upbeat music)

33:53

(upbeat music)

33:58

(upbeat music)

34:05

(upbeat music)

34:20

(upbeat music)

34:25

(upbeat music)

34:32

(upbeat music)

34:47

(upbeat music)

34:52

(upbeat music)

34:59

(upbeat music)

35:14

(upbeat music)

35:19

(upbeat music)

35:26

(upbeat music)

35:45

(upbeat music)

35:52

(upbeat music)

36:11

(upbeat music)

36:18

(upbeat music)

36:38

(upbeat music)

36:58

(upbeat music)

37:18

(upbeat music)

37:38

(upbeat music)

37:58

(upbeat music)

38:18

(upbeat music)

38:38

(upbeat music)

38:58

(upbeat music)

39:18

(upbeat music)

39:38

(upbeat music)

39:58

(upbeat music)

40:18

(upbeat music)

40:38

(upbeat music)

40:58

(upbeat music)

41:18

(upbeat music)

41:38

(upbeat music)

41:58

(upbeat music)

42:18

(upbeat music)

42:38

(upbeat music)

42:58

(upbeat music)

43:18

(upbeat music)

43:38

(upbeat music)

43:58

(upbeat music)

44:18

(upbeat music)

44:38

(upbeat music)

44:58

(upbeat music)

45:18

(upbeat music)