Nick Mehta & Allison Pickens 23 min

AI Meets SaaS with Allison Pickens, Board Director, dbt Labs and Commvault


Explore AI investments and creating Human-First Products in this expert discussion covering how AI's role in customer success.



0:00

>> I love it.

0:02

>> We're going to go on mute and head out and then you guys can take it from

0:05

here whenever you're ready, Nick.

0:07

>> Great. Three, two, one, go.

0:11

Hi, everyone. I'm Nick Meda, CEO of GainSight,

0:14

and I welcome you to the next episode of AI Meet SaaS,

0:18

where we talk about the implications from generative AI and AI more broadly on

0:22

the world of B2B SaaS,

0:23

because we know it's affecting all of our personalized,

0:25

but we think it can transform business as well.

0:28

I think I'm more excited about this episode than anyone I've ever done,

0:31

although don't tell the other presenters.

0:33

I want to welcome Alison Pickens back to the GainSight stage.

0:38

Alison, how you doing?

0:39

>> I'm doing great. I felt the same way when I interviewed you on my sub-stack.

0:43

It was so much fun doing that episode.

0:45

Also, I think people could tell we knew each other really well and we also got

0:48

great feedback.

0:49

I'm excited to join yours now.

0:52

>> Love it. Alison helped create GainSight.

0:54

The early early days of GainSight came in, did so many different roles.

0:59

Eventually, ending up as our chief operating officer and actually being prior

1:04

to that chief customer officer,

1:05

so helped build everything that we have today,

1:07

and I'm immensely grateful to Alison forever,

1:09

and then has done amazing work since being a very active investor herself,

1:14

and lucky to be an investor in her fund,

1:17

as well as being on the board of DBT Labs and Conval.

1:21

She's done so much together.

1:22

We actually wrote a book together years and years ago.

1:25

Over many flights to India, if I don't not mistake it, that was great.

1:30

Today, though, we're going to talk about gender and AI more broadly.

1:33

You've also written a lot about this and spoken a lot about this.

1:36

I think there's a lot we can get on this.

1:38

Before we dive into that, though, we always have a nice breaker as you know

1:41

very well.

1:42

I'm going to go back to an oldie but goodie because I think I have a guess of

1:45

what you're going to say.

1:45

What's your go-to karaoke song?

1:48

Oh my gosh, I can't say I've done karaoke in a while being the mother to two

1:53

very young children.

1:55

My favorite all-time karaoke song is "Timber,"

1:58

which you might remember me performing at an early gig site event,

2:03

and whenever I think of karaoke, I think of that song.

2:05

Me too. I better think of you. I think of "Timber" too.

2:08

So "Pipple Timber."

2:09

You might be doing a little more like "Kids Bop Music Now" in the next few

2:13

years or frozen.

2:14

Maybe "Bumble Bee."

2:15

Yeah, exactly.

2:16

That's recurring for us.

2:18

You're a karaoke that now, if you want.

2:20

Okay, so we searched to AI.

2:22

The first one is a more general question about how you think about AI.

2:27

How do you use it in your own life personally, professionally?

2:31

I've even heard perhaps you use it on weekends sometimes.

2:34

The one I was using on a weekend was, again, being the mother to two young kids

2:39

My husband and I have not gotten out all that much lately, I would say,

2:43

although we're trying to do better at it.

2:44

We did spend a Saturday night, date night, at home,

2:48

playing around with the main journey earlier this year.

2:51

And it was so much fun poking around and being in the Discord forum

2:55

and seeing how other people use it.

2:58

And it gave me early exposure to what prompting means and why it matters,

3:04

how you phrase it.

3:05

I also have used "Pilot AI," which I thought was really interesting

3:10

as sort of a conversational partner.

3:12

And I experimented in a few ways with it.

3:15

One, let's look at probably a topical cover later.

3:18

Like, if you're a customer success manager,

3:20

you can use this AI as a conversational partner,

3:23

potentially to brainstorm about client situations.

3:27

Like, what to do?

3:28

You could give it the total scenario.

3:30

I'm trying to get a hold of the stakeholder.

3:33

They won't respond and hear all the things that are going on.

3:35

And so I poked around with it in that way.

3:38

And also started asking it for feedback on my sub stack.

3:41

And it actually had some interesting ideas,

3:44

which I initially rejected maybe out of hand because I'm like,

3:47

this isn't AI.

3:48

Like, what does it know about, you know, what I'm doing?

3:50

But actually, it's inspired lots of trains of thought later.

3:54

So I think that's been sort of a useful endeavor.

4:00

And then actually, the way that I use it,

4:02

probably on a more ongoing basis professionally,

4:05

is with my investment memos.

4:07

I actually have a contractor who used to be a gain site intern of mine,

4:11

by the way, who's phenomenal.

4:13

And I gave her the broad mission to go figure out

4:19

how to streamline the creation of investment memos

4:22

based on notes that I've taken about the company,

4:25

other observations, decks and things like that.

4:28

And figure out how to create investment memo very quickly.

4:31

And notably, I had heard do it because there has to be human in the loop.

4:37

It's not enough just to, like, for me to like,

4:39

be like chat GPT, like create something there.

4:41

There are certain like actions that still need to be done by human

4:45

and certain forms of like, you know, judgment calls and error checking

4:49

and things like that that she needs to do.

4:51

But that's been, you know, become pretty core to my investment memo creation

4:54

process.

4:55

That is so cool.

4:56

And actually, I read your investment memos are quite good.

4:59

So I didn't realize that chat GPT is working behind the scenes.

5:02

But I know your judgment goes to too.

5:03

Of course, yeah. And of course, like there's a layer of me on top of it.

5:06

Exactly.

5:07

But it's nice actually eliminating a lot of the busy work.

5:10

Well, it's a good, it's a good example to where you're like using

5:13

chat GPT and AI to augment kind of human work, right?

5:16

Versus totally replace it.

5:17

So that's going to be a prime theme that comes up.

5:20

So, okay, so let's talk about it because you're in,

5:22

you've done SAS for so long and you're also thinking a lot about AI.

5:26

So what do you think are some of the biggest opportunities right now

5:31

to really have AI change the world of SAS?

5:34

So, you know, people talk about there being different layers of the generative

5:38

AI stack,

5:38

not just the application layer, which is probably what people are most

5:42

interested in

5:43

when they're thinking about what the opportunities are.

5:45

But, you know, just to kind of go through the layers,

5:46

of course, we have the foundation models and they're ones, you know, produced

5:50

by OpenAI

5:50

and Thropic that are the best known.

5:52

But actually, a lot of people think now that there are going to be tens of

5:55

thousands of models,

5:56

many of them very specialized with a small number of parameters

5:59

that each one becomes very good at specific tasks.

6:02

And whenever you have a task that needs to get done,

6:05

it can be routed to the best specialist AI.

6:08

So, I think there's actually a ton of opportunity for folks to create,

6:12

you know, very smart individual models that are very good at a particular task

6:16

and can be easily observed and monitored because the number of parameters is

6:21

small.

6:22

Then you've got, you know, vector databases where, you know, companies like

6:26

Pinecone are taking off.

6:27

I'm not, I can't say I'm an expert on the space.

6:29

I don't know if there's like opportunity for more vector databases,

6:32

but, you know, Pinecone has emerged as a leader in this space,

6:36

helping you figure out how to, you know, search and store, you know,

6:39

cross, you know, vector data, which is really important for the creation of AI

6:43

models.

6:44

And then you've started up to just so the audience can track it.

6:46

Do you want to describe like how, if you, if you have anything more

6:49

than that, how a vector database fits with the foundational model, like how

6:52

that helps a use case?

6:54

Yeah, again, can't say I'm like an expert on the space.

6:57

But I think the general idea is that if you're a model,

7:00

you basically got to ingest, you know, thousands of different attributes about

7:05

what you're scanning.

7:05

Like if it's an image or something composable and a vector is essentially like

7:11

a series of numbers,

7:12

each of which might characterize like a particular attribute of that piece of

7:16

content that you are

7:17

analyzing. And so, and we do that actually. We use it today because we're

7:21

bringing in all the

7:22

note data from Gainsite, a lot of the timeline posts, like basically, notes CS

7:26

Ms are taking.

7:27

And we put them into a vector database and then use the LLM to kind of generate

7:31

basically like a

7:32

understanding like it's actually called cheat sheet, like a prep note for an

7:36

executive about

7:37

everything that's happened with this client, similar to your investment memo

7:40

idea, but actually for

7:41

like a customer prep meeting. And so that vector database was quite important

7:46

because there's just

7:46

so much data. We can't just pass it all into the LLM as a prompt. So yeah,

7:52

quite similar.

7:53

Oh, interesting. Okay, great. Then you've got the middleware layer where, you

7:57

know, you've tools

7:58

like laying chain and fix the AI, which make it easier to build applications

8:04

using LLM. And then

8:04

you've got the application layer, which again, I think is what most people

8:07

think about SAS.

8:07

They're like, this is where things get interesting. And they're, you know, I'm

8:11

seeing some notable areas

8:14

of opportunity. I think one is selling into automating services, which are

8:21

inherently, you know,

8:23

composed of a lot of human effort. So examples of this might be, you know,

8:27

company called EvenUp,

8:29

which is one of my portfolio companies, sells to personal injury attorney

8:33

attorneys and their

8:35

firms, right? And the idea here is that you want to automate the process for

8:40

ingesting data about

8:42

cases that are similar to the injury case that you're working on, ingest data

8:45

about, you know,

8:46

that particular cases, you know, the medical situation, the harm to the client

8:51

done, the fault, you know,

8:53

all that stuff. And then, you know, you can auto create a demand letter, which

8:58

sent to, you know,

8:59

the accused, I guess, saying like, this is the amount that will settle for. And

9:05

so it basically

9:07

automates like huge percentage of the work that apparel legal might otherwise

9:11

be doing painstakingly

9:12

and probably in a lot of cases, not enjoyably. And it actually gives up time

9:17

for the apparel

9:18

legal to go do more like client relationship work, which for customer success

9:21

people, of course,

9:22

is like probably more exciting, especially. So that's an example of like, you

9:26

know, automating

9:28

services. And then there are also companies that are trying to automate

9:31

services done by,

9:32

you know, boutique firms, like consulting firms, not necessarily boutique, but

9:36

you know,

9:37

the likes of Deloitte and others that are helping to configure software. You

9:41

know,

9:42

with for example, like Salesforce.com. So, you know, companies like Superframe

9:47

and SwanTide are

9:48

trying to, you know, go after that challenge. And then I think there's another

9:53

opportunity

9:54

application layer that has to do with unlocking new markets that haven't

9:58

previously bought software,

10:00

personal injury attorneys, I think probably a good example of that. But micro

10:04

SMBs are

10:04

probably the most notable example where these are, you know, small companies

10:08

that haven't even

10:09

bought, you know, gusto or other SMB oriented software that we're familiar with

10:13

. They don't

10:14

have a need for it, but they might have a need for generative AI, which is very

10:18

easy to do.

10:19

Super streamlined onboarding and can take a lot of costs out of their efforts.

10:26

For example,

10:26

if you're a hairdresser and you're spending all day, you know, cutting people's

10:30

hair,

10:30

you may not have the ability to answer the phone when someone calls to make a

10:34

hair appointment.

10:34

And people like actually often call to make their hair appointment instead of

10:37

like booking it online

10:38

in some way. So there are now, you know, a couple of companies I've seen that

10:41

are cropping up to

10:42

create AI voice assistance that allow a customer to interact with AI and like

10:48

easily book a meeting,

10:49

which is, you know, cost effective for the hairdresser and generates a big

10:53

revenue lift for them.

10:54

I love that so many good examples. So it sounds like there's certain types of

10:58

use cases you can

10:59

enable. Like the personal injury example, there's certain customer segments you

11:03

can target.

11:04

And I thought that middle one is pretty interesting for a lot of folks

11:07

listening,

11:07

because we're going through this too of making software easier to manage and

11:11

configure, right?

11:12

I think that that was a point around sales force and actually against that, we

11:15

're working a whole

11:15

bunch of stuff to let you configure gain site through prompts based creative

11:19

report through a

11:20

prompt set up a health score, things like that. So it could change the way of

11:23

how people use

11:24

software in general, where it's like, I don't know what you think, but like

11:28

reduce the friction

11:28

of using software more broadly. Totally. Yeah. And I think it's you. I mean,

11:33

the fact that

11:33

gainset is working on this brings up an interesting point about, you know,

11:36

should every company be

11:38

developing its own way of auto configuring their software? Or is there a more

11:42

horizontal play here?

11:43

And these early stage companies, I think, would like to think that there is a

11:46

horizontal play,

11:47

but you know, we'll see how how it pans out. Yeah, I'm sure both will happen to

11:50

some extent.

11:51

So that's great. And then if you've looked at, we talked to us a little bit,

11:55

but people really

11:56

innovating in AI, one thing I'd be really interested is the challenges they're

12:00

running into. So like,

12:01

what are the, like it could be like you said, a small company doesn't have the

12:05

data or doesn't

12:05

have the workflow. What are some of the challenges, both small companies, as

12:09

well as more mature

12:10

companies are dealing with in like actually doing this, like making it happen?

12:13

You know, I would say there are a couple challenges that I've seen. You know,

12:19

one is that I think

12:20

in enterprise companies have a lot of budget available to try out AI products.

12:26

If you're in a

12:27

boardroom of a big company, chances are AI is on the agenda every time it's

12:32

over. And you're seeing

12:33

a list of all the vendors that they're trying out. Notably, they are trying

12:38

things out. They're

12:39

playing playing around with things is like another question that I've heard.

12:43

And so, you know,

12:44

startup might sell into an enterprise, get a six figure contract, but then the

12:48

key is can they

12:48

actually deliver on a value proposition that's enduring? And at the end of that

12:53

one year contract,

12:54

you know, can they renew that contract? I think for a lot of vendors, there

12:58

will be a lot of churn,

13:00

you know, after a year because of this. And so it's, I think it's really

13:04

important that AI

13:06

startups invest in customer success. However, you define that like just making

13:10

your customer

13:11

successful. The founder has to dedicate a lot of attention. I've also seen

13:16

challenges from

13:17

startups that were built before the release of chat GPT. So, we've entered an

13:23

AI way that,

13:24

you know, envisioned a different kind of ecosystem that they're selling into

13:28

another,

13:28

trying to figure out how to adapt. And a lot of them, you know, started

13:31

incorporating LLMs into

13:33

their product, but in ways that felt kind of ancillary to the search for a

13:38

product market fit and

13:39

maybe a distraction, you know, they might be building a chat bot.

13:45

Maybe, maybe you could argue that that's just going to be table stakes for all

13:48

software applications

13:49

growing forward and you like have to have an AI chat bot. But you know, is that

13:53

really worth

13:54

your time when you have 12 to 18 months of runway and you're trying to deliver

13:58

on like a poor value

13:59

prop and show that you're an aspirin and not a vitamin, you know, I don't know.

14:03

So I do worry

14:05

about like some companies getting distracted. That's really super interesting.

14:08

One of the things

14:09

we wrestle with and tell me if you have seen this, probably my guess is it

14:12

happens at bigger teams,

14:14

is, is AI like a team inside engineering or is AI something everyone's working

14:20

on? And that's

14:21

something I think as you get into hundreds of engineers or whatever, it's like,

14:24

okay,

14:25

we create a separate team for AI or do you create like get everyone to work on

14:29

it? But most people

14:30

don't know that much about it. So that's something we're wrestling. I don't

14:32

know if you've seen that

14:33

at some of the companies are on the board of or things like that, but I'm

14:36

seeing them more and

14:37

more. Yeah, I can't say direct exposure to that. But it reminds me of a

14:39

conversation you and I have had

14:41

earlier about how in several different technology waves, there was initially

14:45

one department that's

14:47

possible for it, whether it was like, you know, your mobile strategy, or maybe

14:51

before that,

14:51

your internet strategy, more recently, your staff strategy. And then eventually

14:57

that theme

14:58

just becomes the whole company and everyone needs to be up to speed. So maybe,

15:03

maybe going back to

15:03

your question, it's just a matter of time before everyone becomes an AI

15:07

engineer. Yeah, it's

15:08

interesting because that's what when I we're spending a lot of time on it

15:11

because we have a whole

15:11

bunch of stuff coming out, but also the possibilities feel endless. Like, like,

15:15

I don't know if you

15:16

feel that when you start ideating on this stuff or talking to entrepreneurs, it

15:20

like even in our

15:21

business, it's like, Oh my gosh, every single thing can not 90% of things get a

15:25

lot better

15:26

with with the general AI. And yeah, just came off another customer event where

15:31

that we're brainstorming

15:32

about gender, to AI stuff to do around survey feedback and NPS and things like

15:36

that. So that's

15:38

really, really cool. So we talked about innovation. And you know, you get

15:41

pitched a lot of startups,

15:42

which is must be super fun in terms of just seeing the ideas out there. I can

15:47

imagine also,

15:48

there's an question of filtering. So how do you everyone's going to say they're

15:52

an AI company now,

15:53

I'm guessing like 95% or something. How do you figure out like the ones that

15:57

are serious about it,

15:58

that it's foundational to their company versus like a little bit of icing on

16:02

the cake? Like,

16:02

how do you filter? Yeah, you know, I am very suspicious, less of AI companies

16:11

that make it

16:11

sound like they're AI and not doing it and more companies that are trying to

16:14

build in a very

16:15

crowded space. Many really crowded spaces, like examples include note taking

16:23

applications,

16:24

like ingest, you know, transcript of a call and generate summary. That could be

16:29

just a feature

16:29

of gain site as I thought we're adding we're shipping it next month actually. I

16:33

think I remember

16:34

learning about that. So, you know, that's an example of a very crowded space.

16:38

You know, other

16:39

spaces like, you know, scan publicly available data sources about a company,

16:45

say from 10Ks or

16:47

posted job descriptions and surface talking points to sales reps, more, you

16:53

know, potentially

16:54

like customer success folks so that they can more easily sell or renew or

16:58

otherwise like

16:59

align with the customer. Again, like probably a feature of another company,

17:03

maybe gain sites

17:04

product. So, you know, and tons of startups are going after that problem. So, I

17:09

think that's

17:09

where I tend to be suspicious. I think, you know, what's particularly

17:14

interesting is founders that

17:16

have a lot of direct exposure to a very specific pain point. They have a lot of

17:22

domain knowledge,

17:23

great founder market fit. There's a lot of white space and, you know, they're

17:30

in like potential

17:30

to take out a ton of tons of cost. I love that. Yeah. I mean, your personal

17:34

injury example seems

17:35

like a really good one where that's so specialized that it's not just, oh, use

17:39

chat, you know,

17:39

chat GPT APIs and run some data through it. I'm sure there's a lot of workflow

17:43

and things as well.

17:44

Yeah. I think there's a there's just to give it a couple other examples. There

17:49

's a company called

17:49

Fulcrum that I invested a bit in recently. And, you know, that team has a lot

17:55

of experience

17:56

at consulting firms, you know, McKinsey and others helping to, you know, reduce

18:02

like basically ship

18:03

products that cut out work for back offices. And so now they're delivering a

18:08

product that's

18:09

essentially UI path, but in an AI context. So like AI for RPA, robotic

18:14

automation.

18:15

Yeah. Imagine, you know, dramatically taking out costs from BPOs or like shared

18:20

service teams

18:21

that are based in, you know, India or like other places that, you know, are

18:25

sort of known to be

18:26

like lower costs. So, you know, very interesting opportunity there. And then I

18:31

think also, you know,

18:32

kind of to my point earlier, I'm really interested in companies that are going

18:36

after enterprise

18:37

budgets because there's a lot of budget there. And the key question is like,

18:41

are the founders

18:42

enterprise ready? Do they have prior relationships, the companies? Do they

18:46

exhibit like

18:47

tremendous go to market instincts, which of course, a lot of technical founders

18:51

and AI, you know,

18:52

may not. But if you can have that combination of like great technical skills

18:56

and really strong

18:57

enterprise go to market instincts, you know, I think it can be unbeatable.

19:01

I love it. Well, let me close out on the human side. So, you know,

19:06

Allison actually helped create our purpose statement, which again,

19:08

insight, which is to be living proof, you can win in business while being human

19:12

first. And that's

19:13

been very durable since 2016 when we all came up with that in an offsite. And I

19:18

also remember

19:19

vividly you did this keynote at Pulse, which is our annual conference. And you

19:24

talked about like

19:25

human first products and like your experience of technology. And sometimes all

19:29

of us getting

19:29

overwhelmed with technology and having, I think I remember something about you

19:33

not using tech

19:34

right in the morning, right? And so that was, you know, five, six, seven years

19:39

ago, right,

19:39

quite a while ago. And you were thinking about back then, and now obviously,

19:43

generative AI is

19:43

right here. And we're thinking about the human aspect versus technology. And

19:48

you had this quote

19:49

recently, when you're chatting with generative AI, you're chatting with someone

19:52

who's supremely

19:53

creative and more attuned to divergent thinking. So what's your latest

19:57

reconciliation of human

19:59

first and like AI and everything we're talking about? It's a great question.

20:03

And I can't say that I

20:04

have like a new framework for describing how AI should be human first. If I did

20:10

, I would be out,

20:11

you know, promoting it guys. But a few things that I would call out, I think

20:16

one is, if you are

20:18

attuned to your market, chances are you are finding a way to make humans more

20:25

productive using AI.

20:27

So you have you almost by definition have a human first mindset. If you were

20:32

looking at people and

20:33

thinking, what needs can I solve for people? Because at the end of the day,

20:37

people who have

20:37

money to like spend on AI, if you are more messianic about AI, if you have sort

20:44

of a religious

20:45

attachment to it, which I do think some people do, right, I think you will end

20:51

up building things

20:52

that are not human first. And I think and would hope would not actually be

20:57

bought by people.

20:58

I'll give you an example actually of a story I heard recently that I found

21:02

troubling and hopefully,

21:03

you know, it will be defeating in some way in the end. Like I heard that,

21:07

again, not to be like too provocative here, but I heard that at an open AI, you

21:13

know,

21:14

meeting, there was sort of a discussion about how GPT was like officially autom

21:20

ating what

21:21

writers like professional writers were doing. And maybe it was a response to

21:24

the writer strike.

21:25

And there was like a celebration in the room about like, yay, like we've

21:33

automated the job of

21:34

writer. And there's an open AI engineer who I know quit in response to this

21:38

meeting.

21:39

Oh my gosh.

21:40

That's an example where I think like, I don't think we should be celebrating

21:45

automating

21:46

professions in itself. Like that's not an interesting celebration to me. It

21:51

actually

21:52

speaks to like a religious attachment to all of me. I think people need to do

21:57

work for money,

21:59

but also for purpose. Like I think we need stuff to do when we get up in the

22:03

morning. And if AI

22:04

can help us do that stuff or do better stuff or stuff that we want to do, like

22:11

that is success

22:12

for AI. And I think also the market will reward products like that.

22:17

I love that. That's such a good way to kind of close out, which is just

22:20

thinking about how

22:21

AI can help the human first, which obviously will help the business too. So let

22:25

me close

22:26

out with word association every time somebody has say three things. Tell me the

22:29

first thing that

22:29

comes to mind. So generative AI. Future. Future customer success.

22:39

We start the open over. Yeah. We're preparing answer to the

22:43

okay. We start the whole thing over. Yeah. So we're gonna do no worries. We're

22:47

gonna do word

22:48

association to close this out. So I'm gonna say three things. You just say the

22:50

first thing that

22:51

comes to mind first term or phrase. generative AI. Future. Customer success.

22:58

Happy. Happy. Gain site. Pulse. Pulse. There you go. Awesome. Well, we'll

23:06

hopefully see you in

23:07

another pulse sometime soon. Allison, thanks so much. This has been incredible

23:11

kind of

23:11

walk down memory lane, but also some great insights for the future. So really

23:14

appreciate it.

23:15

Thanks for having me Nick. Awesome. I was really

X

This is a test comment.

X

This is a longer test comment to see how this looks if the person decides to ramble a bit. So they're rambling and rambling and then they even lorem ipsum.


author