Artwork for podcast RevOps FM
Is AI a Game-Changer for Data Analytics? - Grant Grigorian
Episode 503rd October 2024 • RevOps FM • Justin Norris
00:00:00 01:07:06

Share Episode

Shownotes

This week we chat with Grant Grigorian, a long-time friend of mine and founder of Mogi.ai, to explore whether AI is actually a game-changer for data analytics.

We examine what’s broken in the current landscape of data analytics, the complex world of marketing attribution, and the “last mile of analytics” problem (that is., how do you get business users to actually read and use the analyses you produce).

Then we dive very deep into practical applications of AI for data analysis, looking at concrete examples to highlight the strengths and weaknesses of large language models (LLMs) and contrasting them with traditional machine learning approaches.

In closing, Grant reflects on the future of data careers in the age of AI.

Thanks to Our Sponsor

Many thanks to the sponsor of this episode - Knak.

If you don't know them (you should), Knak is an amazing email and landing page builder that integrates directly with your marketing automation platform.

You set the brand guidelines and then give your users a building experience that’s slick, modern and beautiful. When they’re done, everything goes to your MAP at the push of a button.

What's more, it supports global teams, approval workflows, and it’s got your integrations. Click the link below to get a special offer just for my listeners.

Try Knak

About Today's Guest

A startup guy who loves B2B marketing technology, analytics, and AI, Grant Grigorian is a serial entrepreneur. Coming from an operations background, Grant founded Path to Scale, a marketing analytics company, that sold to Engagio (later acquired by DemandBase). Today he's the CEO and co-founder of Mogi, an app that simplifies marketing data analysis by automatically delivering insights and recommendations to your team.

https://www.linkedin.com/in/grantgrigorian/

Key Topics

  • [00:00] - Introduction
  • [01:06] - What’s broken in data analytics
  • [08:45] - Using data to justify marketing’s existence
  • [18:15] - Analyzing marketing tactics
  • [24:39] - The “last mile of analytics” problem
  • [28:56] - Strengths and weaknesses of LLMs to extract insights from data
  • [45:49] - Difference between LLMs and machine learning
  • [50:37] - The importance of context for AI
  • [54:39] - Impact of AI on data careers
  • [1:00:07] - Motivations as a repeat founder

Thanks to Our Sponsor

Marketers: it's planning season. Time to live in your office with the world's biggest spreadsheet and hope that finance doesn't slash your budget.

Planning is tough, but at least you can put your best foot forward with a structured framework that covers all the bases. Our friends at Uptempo have just launched the Blueprint for Marketing Planning.

It's a totally free and comprehensive guide for enterprise marketing teams. You'll get a seven-step planning process that integrates top-down strategy with bottom-up execution and gives you a crystal clear picture of how you'll hit your number.

Get your free copy now:

Uptempo Planning Guide

Resource Links

Learn More

Visit the RevOps FM Substack for our weekly newsletter:

Newsletter

Transcripts

Speaker:

Welcome to RevOps FM, everyone.

Speaker:

Today, I'm joined by a longtime friend and collaborator.

Speaker:

He is a multi time founder, a marketing analytics expert, and an all around

Speaker:

innovative guy, Grant Gregorian.

Speaker:

Grant and I first connected through the Marketo community,

Speaker:

more years ago than I probably care to acknowledge at this point.

Speaker:

And then went on to found The company path to scale, which made a marketing

Speaker:

analytics app and was later acquired by Engagio later acquired by demand base.

Speaker:

And now Grant is back innovating in the analytics space.

Speaker:

He just can't keep himself away.

Speaker:

Uh, and as the founder of an app called Moji, which automates

Speaker:

marketing data analysis, using AI, turning data into action faster.

Speaker:

So we're going to learn a bit about what it does and all things AI today.

Speaker:

Grant, it was always a blast to hang out with you.

Speaker:

Thank you for coming on the

Speaker:

Thanks for having me.

Speaker:

Grant, maybe we just start, like you have been in this data space, I guess,

Speaker:

for, I don't know, 10, 15 years.

Speaker:

Like you keep coming back to this problem and I'm just curious, what

Speaker:

is it for you that, draws you here?

Speaker:

Like what's broken that you feel just needs fixing?

Speaker:

Well, you're right.

Speaker:

I do keep coming back to the same problem.

Speaker:

I would say more than a decade ago, I was a, data analyst and I found

Speaker:

myself at a tech company and what really caught my curiosity was that

Speaker:

the CMO was very analytical person and he wanted a analytical Model almost

Speaker:

of the way that they spent money.

Speaker:

It was a VC funded tech company.

Speaker:

And so he gave me this giant Excel spreadsheet and he said, Grant, your job

Speaker:

is to validate all of these assumptions that are, you know, it was like the

Speaker:

classic, we'll spend money here to get this many leads and then MQLs and

Speaker:

then did it like that whole thing.

Speaker:

And I had never seen that before.

Speaker:

And I, and then prior to that, my background is in, you know, more like

Speaker:

sciences, engineering, math world.

Speaker:

I didn't know how people.

Speaker:

Money.

Speaker:

I didn't think you could like, just quantify it like that.

Speaker:

And I was like, wait, wait a minute.

Speaker:

You mean we're just going to like engineer our way into, uh, sales.

Speaker:

And he's like, yeah, we're gonna, this is how many meetings and then

Speaker:

handshakes and did it and closed one.

Speaker:

and so he worked backwards from how much money do you want to make?

Speaker:

And I was like, wow, this is incredible.

Speaker:

This is like a way to engineer success.

Speaker:

And you're saying it's my responsibility to ensure that

Speaker:

this is all going to work out.

Speaker:

And I found it to be basically an impossible task.

Speaker:

So he gave me a task that he knew was going to be very hard for me to

Speaker:

actually achieve, given the reporting capabilities that we had at the time.

Speaker:

And, we had Marketo, early customer of Marketo.

Speaker:

We had Salesforce, all these things.

Speaker:

And I just fell in love.

Speaker:

I was like, wow, this is so surprisingly complicated.

Speaker:

It was a B2B company, large, deal size, many, stakeholders.

Speaker:

And it took a long time to get the deal.

Speaker:

And so all of those ingredients, and there are many different campaigns happening

Speaker:

at the same time, many different people involved, salespeople, SDRs, marketing

Speaker:

people, multi channel, go to market.

Speaker:

It was impossible.

Speaker:

I was like, wow, this is so interesting.

Speaker:

So complicated.

Speaker:

want to make this work.

Speaker:

And I've kind of been trying to get that CMO's approval since then.

Speaker:

what made it impossible at that time?

Speaker:

Was it just the technology that you had available to you or something else?

Speaker:

all of those things are still difficult today.

Speaker:

It's a classic attribution ish problem of how do you, capture all of those

Speaker:

interactions and then how do you decide what mattered and then how do you do

Speaker:

more of what matters and less of what doesn't, let's be less wasteful, let's

Speaker:

enable, and all of the same things that we have today applied then.

Speaker:

I just knew a lot less about everything.

Speaker:

So what ended up happening was at the time that company

Speaker:

didn't have a Salesforce admin.

Speaker:

And so I was like, well, where's all this data?

Speaker:

It's in Salesforce.

Speaker:

So, okay.

Speaker:

Who's in charge of it?

Speaker:

No one.

Speaker:

I guess you are now.

Speaker:

And by de facto, I did become a Salesforce admin there because I was like, okay,

Speaker:

well, it's standardized opportunity.

Speaker:

Let's figure out all the leads and how they're organized.

Speaker:

The lead sources, that whole thing campaigns.

Speaker:

And then he's like, Oh, Salesforce is fed by this other system.

Speaker:

Marketo who's in charge of that?

Speaker:

No, I don't know.

Speaker:

One you are.

Speaker:

and now doing Marketo and now I'm a business operations manager.

Speaker:

Now we're doing.

Speaker:

You worldwide business operation.

Speaker:

Now we have channel and their partners in Europe, and we've got direct

Speaker:

sales go to market in United States.

Speaker:

and so now the spreadsheet is getting more complicated.

Speaker:

My dashboards are getting more complicated.

Speaker:

roughly the same thing happens today in companies.

Speaker:

now, when I consult and when I go and I.

Speaker:

Open up Salesforce or open up.

Speaker:

It's either it's some vendor that has a flavor of that.

Speaker:

We've got all the touch points.

Speaker:

We, and then we're doing some sort of analysis to see which things mattered,

Speaker:

but there are so many obstacles to it.

Speaker:

There's like data capture, there's standard process, the way to

Speaker:

interpretation of the final results and what does it mean and what should we do

Speaker:

about it, all of those things are hard.

Speaker:

They involve alignment.

Speaker:

They involve good.

Speaker:

Business practices, they involve lots and lots of communication, and lots and

Speaker:

lots of non math things also, and that's what makes it really, really difficult.

Speaker:

My observation is that the state of marketing analytics has not

Speaker:

progressed, very much at all.

Speaker:

I think in the last 10 years, I think you've alluded to this too.

Speaker:

I think if anything, in some ways it's, it's regressed or at least

Speaker:

it's coming to a point of crisis.

Speaker:

In that there's very little consensus.

Speaker:

Uh, everyone thinks that every other method is wrong.

Speaker:

A lot of methods, like maybe certain types of attribution that historically

Speaker:

have been popular and now like under heavy attack all the time, it

Speaker:

really just feels like as a, as a discipline, we don't know what we're

Speaker:

doing, which is kind of embarrassing.

Speaker:

you comment on the situation at all and maybe where you see it going?

Speaker:

I have such mixed feelings about this, so I agree with you.

Speaker:

And in many ways, it's a good thing.

Speaker:

I don't know if we should be allowed to capture every interaction in a digital

Speaker:

and non digital form and put it on a spreadsheet for marketing purposes.

Speaker:

there will be by design and by legislation the way buyers want to buy from vendors.

Speaker:

They don't want all of the interactions to be visible to vendors.

Speaker:

We don't want to be spammed.

Speaker:

We want to take our own approach to how we're, we're going to buy

Speaker:

and have some level of control.

Speaker:

And when we're already engaged with the sales process, and I think there

Speaker:

is a kind of a natural tension, both for privacy purposes, for the way

Speaker:

buyers want to buy that's making it.

Speaker:

Basically an impossible equation to later solve.

Speaker:

So I agree that it will never be technically perfectly solved.

Speaker:

I do think that it's still valuable in a sense that it's directionally.

Speaker:

the best we could do with the data that we have.

Speaker:

And so we do want to make use of it, but we can't, say that it's a

Speaker:

hundred percent accurate, nor can we just completely throw it away

Speaker:

and say, well, if it's not a hundred percent accurate, why does it matter?

Speaker:

Well, nothing is a hundred percent accurate, And so I think there are

Speaker:

some nuggets of interesting insight that exists in that data set.

Speaker:

It's just not the end all that it's sometimes sold to be either both by

Speaker:

vendors and as a solution to some internal problem that sometimes get pushed.

Speaker:

You know, Oh, if only we had this dashboard, then we would know

Speaker:

we're just, well, you won't, it's not a magic bullet for sure.

Speaker:

the challenge I've run into a few times is the marketing leader, the

Speaker:

CEO, whoever wanting that dashboard that gives the magic formula.

Speaker:

Like, I think there's this myth that it's like, Oh, if we could just see

Speaker:

like two webinars, plus three eBooks, plus cold call equals an opportunity.

Speaker:

And then we'll just keep doing that.

Speaker:

And of course that I think we both agree is kind of a fictitious proposal, but.

Speaker:

when you come in and you're, and you're navigating these expectations.

Speaker:

What do you actually propose then?

Speaker:

It's like, here's what's useful.

Speaker:

what does good look like in your eyes?

Speaker:

it really depends on what the stakeholders want to do with it.

Speaker:

So there's so many different motivations for companies to

Speaker:

try to make that dashboard.

Speaker:

Sometimes it is an existential crisis for marketing.

Speaker:

Unfortunately, there's so many marketing teams that are under siege.

Speaker:

For funding or for a threat to be laid off or somehow replaced,

Speaker:

or are everybody questioning all of the investments all the time?

Speaker:

when the marketing to CEO relationship is broken, they need a dashboard that says

Speaker:

we did a thing that was valuable enough that wouldn't have happened without us.

Speaker:

So Please give us another day.

Speaker:

that's a big one.

Speaker:

and so for those folks, it's not about gathering all of the touch points.

Speaker:

It's about, can we show that we did something good?

Speaker:

Like, we've been working here for like a year.

Speaker:

What did we do?

Speaker:

Did we contribute in any way?

Speaker:

Can you please show some results, in a way that are compelling in the

Speaker:

court of that executive boardroom?

Speaker:

Okay.

Speaker:

All right.

Speaker:

Well, let's get to work on that, low bar.

Speaker:

And then other people are more sophisticated.

Speaker:

They want to know when, where do we spend the money in a more

Speaker:

incremental way, and they need proof.

Speaker:

They need some sort of a thing that's more than just their opinion.

Speaker:

But I find very often, they already know what the best campaigns are.

Speaker:

They already know all of the things.

Speaker:

They just want this additional validation, whether it's coming from a consultant

Speaker:

Or from this dashboard that allows them more political power in an organization

Speaker:

to then say, let's do this because see it's over here and it is just one of

Speaker:

many ingredients that they bring to the table to argue to spend more money.

Speaker:

So, then I, what I usually do is I literally ask what's the story I

Speaker:

literally want them to tell me what is it that you're trying to say?

Speaker:

If you're going to, you're going to go to this important meeting and you need

Speaker:

this dashboard to help you, say what?

Speaker:

What is your hypothesis about what the numbers will say?

Speaker:

Oh, I need to say that these deals are, you know, the deals

Speaker:

that have marketing touches and lots of them end up being bigger.

Speaker:

And so that's if we want bigger deals, you need more marketing.

Speaker:

That's what I want to argue that that we by having marketing teams, you're

Speaker:

going up market and, to the enterprise.

Speaker:

And so you need us to be able to do that.

Speaker:

Can you show that?

Speaker:

So then I want to know, what is it that we're trying to prove?

Speaker:

What is the hypothesis?

Speaker:

And then backwards engineer well, this dashboard won't work for that anymore.

Speaker:

This was something else.

Speaker:

Dashboard.

Speaker:

You want this whole other data set entirely that shows some

Speaker:

sort of a correlation with.

Speaker:

Number of touches and deal size, and you want to show something positive.

Speaker:

so that's how I approach it.

Speaker:

I don't think there is a singular answer and it very much is driven by the

Speaker:

internal business dynamics of the company.

Speaker:

So maybe let's keep drilling on that example.

Speaker:

Cause it's, it's an interesting one that I think we can all relate to.

Speaker:

the company wants to tell the story that more marketing touches

Speaker:

equals a bigger deal size.

Speaker:

what's your next step?

Speaker:

Like, are you then starting to think about data structure?

Speaker:

How do you work backwards from

Speaker:

Yeah.

Speaker:

I start, I immediately think, Oh gosh, how do we.

Speaker:

How do I reinforce that?

Speaker:

Can we go and interrogate the data and tell that story?

Speaker:

So then I need two things.

Speaker:

I need marketing interactions, so I can literally count them.

Speaker:

And they need to be classified as marketing or non marketing.

Speaker:

So I have a classification problem.

Speaker:

I have a capture problem.

Speaker:

are they all going to be there when I go to them?

Speaker:

I probably need to control for other interactions too, so that we

Speaker:

can say it wasn't just marketing.

Speaker:

You see, there's All these other teams that were involved.

Speaker:

So all of a sudden the scope is okay.

Speaker:

I need all the interactions for all everybody for all the deals.

Speaker:

And then I need, to connect them with deals.

Speaker:

So these interactions can't just float in the air all by themselves.

Speaker:

They need to be within some company.

Speaker:

So some important person.

Speaker:

Okay.

Speaker:

So which person?

Speaker:

So now I have like.

Speaker:

Which interactions are in and which interactions are out.

Speaker:

I get to throw away some interactions.

Speaker:

Maybe they're not really interactions.

Speaker:

Maybe we just, we sent them an email, but they never clicked on it.

Speaker:

Okay.

Speaker:

That's not an interaction.

Speaker:

So then we have to define what is an interaction?

Speaker:

We get that going.

Speaker:

Now we have another stream of work where, okay, but how are they

Speaker:

connected to what was purchased later?

Speaker:

Okay.

Speaker:

So they work at the same company.

Speaker:

Did they ever talk to a salesperson?

Speaker:

Okay.

Speaker:

So now we have another stream of work that's all about connecting

Speaker:

these interactions with your deals.

Speaker:

if you've got that going, then we can just count them, count them over what period

Speaker:

of time, count them forever that we've ever had interactions across everyone

Speaker:

at the company or across each person.

Speaker:

Okay.

Speaker:

How many people are there involved?

Speaker:

Okay, so now we have another multiplier, which is number of people

Speaker:

that we talked to in an organization.

Speaker:

And then how many interactions each of that person had.

Speaker:

So do you see what I'm saying?

Speaker:

So now I've got all these spreadsheets of people of interactions and of deals.

Speaker:

I've got some sort of a connection between them.

Speaker:

And I basically want a table that has some sort of a bucket that says

Speaker:

this is the deal size, from a hundred thousand to 200, 000, 200, 000 to 500

Speaker:

a million, and when Y axis and number of interactions with marketing on the X

Speaker:

axis, and I want to scatter plot of all of the deals or, where they landed, and

Speaker:

I want to see a thing that correlates them that says the more, the better.

Speaker:

And if I have that, I could then fade everything into the background and have

Speaker:

my correlation line that says there is a connection and here's what it is for

Speaker:

every additional marketing interactions, you get 50, 000 bigger deal or something

Speaker:

and then you write that in big letters.

Speaker:

Everything else goes away, and you say, here's the math, you give me

Speaker:

more money for more interactions, I give you more money back, that's

Speaker:

the investment math, and therefore, ladies and gentlemen of the corporate

Speaker:

jury, you should award me more money.

Speaker:

you've come back to the story and so I think it's an elegant example of how

Speaker:

you start with the story into the data, all the difficult engineering, and

Speaker:

then back with, with the money slide.

Speaker:

And it's, it's funny that we chose this example because I was doing

Speaker:

some similar analysis, but you also mentioned the word correlation line

Speaker:

and it's called correlation line.

Speaker:

It's not called causation line.

Speaker:

And this is the.

Speaker:

I was careful to kind of asterisk my own analysis in this way of like, correlation

Speaker:

doesn't always equal causation.

Speaker:

How do you kind of address that problem

Speaker:

you

Speaker:

addressed it with executive trust.

Speaker:

That's how with business acumen.

Speaker:

The problem is, is that the executive who's asking for this information

Speaker:

feels under siege.

Speaker:

There is fear in the air and these numbers are there to assuage whatever

Speaker:

doubts and concerns some other executive has about this money.

Speaker:

And so they're in this struggle for power.

Speaker:

if.

Speaker:

they had the full trust and understanding of their colleagues and the CEO of

Speaker:

how marketing works and that they believe that this person is a competent

Speaker:

marketer and they know what they're doing, none of this would be necessary.

Speaker:

This is like, we're jumping through hoops just to satisfy this, like.

Speaker:

Uncertainty that exists that is unnecessary ultimately, but in my

Speaker:

opinion, in the context of that company, in the context of that decision making

Speaker:

process, they find it necessary.

Speaker:

So you say, oh, it's not correlation.

Speaker:

It's causation.

Speaker:

Okay.

Speaker:

Well, now they're now we're nitpicking this analysis.

Speaker:

And we could, the marketer could come back and say, we can spend another 30, 000 and

Speaker:

go and run a regression model, with some analysts, we can hire data scientists.

Speaker:

And they will go and they'll run correlation, we'll find some coefficients

Speaker:

that are technically statistically valid within the bounds of the

Speaker:

P's and the S's and the whatevers.

Speaker:

does anybody care so that we switch the word from correlation to causation?

Speaker:

It's the same chart.

Speaker:

maybe.

Speaker:

The outcome is the same.

Speaker:

We just wasted more and more money chasing false precision.

Speaker:

So it's a political exercise in other words,

Speaker:

So that's why I go back to the story and I'm like, how much fear is in your

Speaker:

voice when you tell me that story?

Speaker:

which also correlates to how many asterisks and slides will this need to

Speaker:

have to really show, you know, the results because the data is already inaccurate.

Speaker:

There's so many air terms in there that flipping it from correlated

Speaker:

to is not going to overcome them.

Speaker:

it's an imperfect data set based on what we talked about before.

Speaker:

And this is maybe the most common, type of analysis, especially in,

Speaker:

today's climate, when many marketing teams do feel under siege, there's

Speaker:

another flavor of analysis where.

Speaker:

It is more about how do I understand what's working and I know that you're

Speaker:

playing, these days Moji is playing in a more tactical space at the moment.

Speaker:

Like you're looking specifically at the email channel and trying to

Speaker:

understand what is working there.

Speaker:

Tell me about that type of analysis because there it's, it's less about

Speaker:

like, should we stop sending email?

Speaker:

I don't think anybody's thinking that it's more like, which emails should we do more?

Speaker:

What's working?

Speaker:

how are those stories being told?

Speaker:

you're right.

Speaker:

So that was just one example.

Speaker:

That was like the existential, should we even exist example?

Speaker:

Really scary for everyone involved.

Speaker:

And that's

Speaker:

why trust is so important.

Speaker:

That's why that really big comes down to relationships.

Speaker:

Ultimately, no PowerPoint is going to save you if the relationships are broken.

Speaker:

but there's many different analysis and many different

Speaker:

types of data or stories to tell.

Speaker:

And so the one that we talked about is kind of the highest level and then there,

Speaker:

it kind of goes down to more tactical, like the teams that are executing the

Speaker:

marketing itself need some sort of feedback about what are they doing?

Speaker:

Let's.

Speaker:

Effective.

Speaker:

How do we measure effective?

Speaker:

Okay.

Speaker:

Well, many different ways, but we need some way to measure so that we

Speaker:

can give them feedback about, Hey, this campaign worked better than

Speaker:

this other one, whether that's email or a trade show or advertising.

Speaker:

We need some way to measure that.

Speaker:

And there is a huge amount of resources being spent.

Speaker:

And addressing that and it is a chaotic wild wild west world because

Speaker:

we have so many different channels ever changing channels, and lots of

Speaker:

tricky vendors that want to show that spending money with us was so good.

Speaker:

And so as a company, then you need to have enough proactive analysis and

Speaker:

proactive data to do your own assessment of what was effective and what was not.

Speaker:

so yes, when we started Moji, we said, let's try to address

Speaker:

this, like one area, which was.

Speaker:

all about email.

Speaker:

Uh, we have this summer expanded since then, so I can't wait to tell you about

Speaker:

but the way that I think about that is.

Speaker:

It's much more routine.

Speaker:

And there, the stories are stories of success and failure,

Speaker:

basically about specific campaigns.

Speaker:

And so I want to be able to, on a timely basis, go back to marketers and say,

Speaker:

Hey, remember that campaign that you did?

Speaker:

Here's the good of it.

Speaker:

Here's the bad of it.

Speaker:

And here's net net, some sort of feedback and based on that, here's some

Speaker:

suggestions on future campaigns and you can cut it many different ways.

Speaker:

You can go as deep as you want, but that's basically the underlying story.

Speaker:

and just, although it can be depends on how high stake that event is I'll give

Speaker:

you an example, these days, it's very popular for, uh, Big software companies to

Speaker:

throw these like big annual shows, think Marketo summit or dreamforce or stuff.

Speaker:

Everyone wants to have one of those for themselves.

Speaker:

And that's a huge high stakes of it.

Speaker:

They spend hundreds of thousands of dollars on those events.

Speaker:

And so for sure they want to.

Speaker:

Tell a story of some sort of success.

Speaker:

so that's where everyone gets involved.

Speaker:

Lots of visibility into that story.

Speaker:

And it's a very tactical level.

Speaker:

How many customers came?

Speaker:

How many prospects came?

Speaker:

Did they then later sign?

Speaker:

How many new prospects that we generate, but usually it's a lot more mundane.

Speaker:

It's a lot more like, let's make a list of all the campaigns.

Speaker:

Let's see which ones had were executed well.

Speaker:

So, you know, there were participants.

Speaker:

Not everybody unsubscribed.

Speaker:

There were some decent engagement.

Speaker:

Like, we can talk about those metrics also.

Speaker:

And then beyond that, move on to, well, but did it make a difference?

Speaker:

Was there an op?

Speaker:

Was there a meeting?

Speaker:

Was there free trial?

Speaker:

And then ultimately, did it lead to some sort of sale?

Speaker:

That's how I think about it.

Speaker:

It's like a layered cake.

Speaker:

We start with the basics and then we go deeper and deeper and deeper

Speaker:

and, on the one hand, we go deeper and deeper in a sense of like,

Speaker:

was it the content that mattered?

Speaker:

Was it the title of the person?

Speaker:

you can go deeper into each channel, or you can go up the strategic chain and

Speaker:

talk about the impact of that, which then takes us right back to the beginning

Speaker:

of the conversation where we started.

Speaker:

when thinking about the need for this type of platform, do you see the problem

Speaker:

as, you know, marketers just don't have time to circle back in this way.

Speaker:

In other words, an efficiency and process problem.

Speaker:

Or do you see it as marketers actually are lacking.

Speaker:

The analytical skills and acumen to analyze this data

Speaker:

and come to these conclusions.

Speaker:

And so we're solving for that.

Speaker:

Or is it a mixture of the two all at

Speaker:

For sure, mixture of the two, I would say it's a balance.

Speaker:

If you had, you know, out of the 100 percent of your time,

Speaker:

would you rather spend that time executing on the marketing?

Speaker:

Or would you take out some of the marketing time or some of the sales

Speaker:

time to look back at the stuff that you already did, to see what worked

Speaker:

or what didn't, it's literally, you're taking money away from additional

Speaker:

advertising spend and giving it to the grants of the world to do what you want.

Speaker:

More Excel work, hoping that the, whatever insights, whatever ROI they find there

Speaker:

makes its way back smarter, not harder.

Speaker:

So they have to have some sort of a.

Speaker:

Understanding that looking at the data will make them somehow better marketers.

Speaker:

If they feel like it won't, when they look at those dashboards or when they

Speaker:

have the conversation with the analyst and they start to feel like, I already knew

Speaker:

that this isn't giving me any additional things, It starts to become really painful

Speaker:

to try to do analysis because I already knew all of this and I should just be

Speaker:

spending my time doing more marketing.

Speaker:

that that interaction has to be net return to marketers.

Speaker:

100%.

Speaker:

If it's not, they will just go do marketing all of the time without

Speaker:

the nudging and the feedback that would make them hopefully.

Speaker:

Do better work if that makes sense?

Speaker:

I know that it's relatively early days for Mojo.

Speaker:

You have been iterating on it for some time now.

Speaker:

What's the feedback that you've heard from marketers what's been encouraging

Speaker:

in terms of showing you that you're heading in the right direction here.

Speaker:

so I'll explain a little bit about what Moji does so with Moji the

Speaker:

problem that we're trying to solve is What's called the last mile of

Speaker:

analytics problem for most of my career?

Speaker:

I spent working on getting the data cleaning the data Putting the data

Speaker:

on the dashboard, or into, some sort of a spreadsheet or slides

Speaker:

and then giving it to marketers.

Speaker:

And what happens is the, in the last mile of analytics is what

Speaker:

happens when the marketer is looking at the dashboard, right?

Speaker:

Is the information going into their brain or is it getting lost in

Speaker:

transmission, somewhere, cognitively speaking, literally, are they, do they

Speaker:

even look at the thing that you made?

Speaker:

And, after doing this for over 10 years, I found incredibly that so

Speaker:

many of my colleagues in marketing and in sales don't actually spend

Speaker:

the time looking at the dashboards as much as I thought that they would be.

Speaker:

And they need a lot more handholding and they need a lot more explaining

Speaker:

and they need a lot more help for many different reasons that we can get into.

Speaker:

both because Let me tell you just a quick story.

Speaker:

So when I was an analyst, this is a different company.

Speaker:

I'm an analyst there and I'm producing these dashboards and

Speaker:

I'm working with the CMO directly.

Speaker:

she has full trust of her executive team.

Speaker:

So she's like, I'm, I gotta go to an executive meeting.

Speaker:

Can you help me prepare?

Speaker:

I'm like, great.

Speaker:

And I did all the things, because I already know about the data or,

Speaker:

you know, about the dashboard.

Speaker:

So I've created her a beautiful prep for executive meeting dashboard.

Speaker:

And I highlighted the link.

Speaker:

I said, here you go.

Speaker:

And she was like, yeah, that's great.

Speaker:

Grant, If I had the Saturday afternoon off, if I had the Saturday afternoon,

Speaker:

instead of spending time with my children, if I engage with this dashboard

Speaker:

and I looked at each chart and I clicked into it and I made notes about

Speaker:

it, and then I extracted the things that I care about, I could take your

Speaker:

dashboard and turn it into something useful for the meeting, but I don't.

Speaker:

I'm a busy executive.

Speaker:

I'm smart enough to engage with this dashboard.

Speaker:

But that's literally your job.

Speaker:

You should engage with the dashboard on my behalf.

Speaker:

That's like why you're here.

Speaker:

It's not just to make the dashboard.

Speaker:

You should spend time stare at the actual numbers to come up with some

Speaker:

bullet points, some talking points for me, for my meeting, and if I have

Speaker:

any questions, I can just drill in and look at the dashboard so I can, we

Speaker:

can, you and I can then have a debate about the business value of these.

Speaker:

talking points, much better conversation than just a technical one, right?

Speaker:

I'm happy because I got the thing prepped for me for the meeting and you're happy

Speaker:

because you're having a conversation at an executive level about all of the metrics

Speaker:

instead of just being a reporting robot.

Speaker:

And so that's the interaction that I was like, I was seeing over

Speaker:

and over and over and over again.

Speaker:

And I really wanted a way for.

Speaker:

Us analyst class, the robots, the reporting robots to elevate our careers

Speaker:

and to start having these more strategic conversations with the executives

Speaker:

using their language and their language is all about these business stories.

Speaker:

they articulate some sort of a hypothesis that we should do this.

Speaker:

And it's because of X, Y, Z, and da, da, da, da, da it's very much like

Speaker:

business school slideshows and much less.

Speaker:

You know, spreadsheets and dashboards.

Speaker:

Those are all in the appendix.

Speaker:

If somebody really wants to, they can go check it out.

Speaker:

But you need to be able to pull out and tease.

Speaker:

the

Speaker:

idea of Moji was like, can we use technology to do that?

Speaker:

can we spend more time analyzing the data?

Speaker:

And helping the users, the business users analyze the data and tease out

Speaker:

the stories from the data, full stop.

Speaker:

Can we do that?

Speaker:

And I was working on this problem and simultaneously.

Speaker:

Oh, the LLM AI revolution kind of happened a couple of years ago,

Speaker:

and I was like, this is my moment.

Speaker:

This is what I've been waiting for.

Speaker:

Let's just put some of these dashboards straight into LLMs and off we go.

Speaker:

And so we've been working on that problem since.

Speaker:

And turns out, it is in fact a very powerful way to help

Speaker:

articulate the insights that might be hidden in the dashboard,

Speaker:

but you have to do it correctly.

Speaker:

it's also a very burgeoning new field.

Speaker:

And so we've been experimenting and playing with that idea.

Speaker:

in my opinion, is inevitable.

Speaker:

It is inevitable that in the future robots interpret the data for us humans.

Speaker:

There's way too much data.

Speaker:

It's way too complicated.

Speaker:

Imagine running a regression coefficient analysis.

Speaker:

how are you going to explain it to the executive team?

Speaker:

Who's going to explain it, Somebody will need to do that work and put

Speaker:

it in the context, the business context of what's being discussed.

Speaker:

And so they are going to be great at it.

Speaker:

I imagine you would come to work and you would say, Hey, robot, give me an update.

Speaker:

And it was like, well, Grant, the stocks are up and the Dow is two

Speaker:

points and you know, whatever the equivalent is for your work, well, the

Speaker:

MQL numbers are looking good, but this advertising campaign was doing this.

Speaker:

So I adjusted the things.

Speaker:

I hope that was okay.

Speaker:

And, you know, Nancy over here, she requested this sort of data.

Speaker:

So I sent her this, I hope that was okay.

Speaker:

And I prepared for you some, you know what I mean?

Speaker:

Like it's going to be proactive.

Speaker:

It's going to be doing all this work for you and analyzing all of this data.

Speaker:

And explaining it to everyone in the context that they need.

Speaker:

Oh, I sent this person a spreadsheet and this person got a little

Speaker:

summary update and, um, and for you, I put in the long report.

Speaker:

Cause I know how, like, how much you like to read long reports.

Speaker:

Here you go, Grant.

Speaker:

So I was like, okay,

Speaker:

let's go make that.

Speaker:

I believe you completely.

Speaker:

I think, that probably is inevitable.

Speaker:

I'm curious.

Speaker:

Here's the thing, you know, when you actually start to work with these tools,

Speaker:

you start to see that like, yes, it's there, but there's also, limitations.

Speaker:

There's also challenges.

Speaker:

Where's the gap between that ideal sort of Jetsons like

Speaker:

future and where we are today?

Speaker:

Absolutely.

Speaker:

So the thing that I found to be very helpful as any, anyone who's, you know,

Speaker:

there's a lot of initiatives now in companies to like, Hey guys, like we've

Speaker:

got to use AI We can't be left behind.

Speaker:

So I would say this is one

Speaker:

of those spaces that you should try to use AI and start playing with it.

Speaker:

And so the thing that helped me a lot to think about is to break the word AI.

Speaker:

we just use it as a Stand in for the new set of technologies that

Speaker:

are all, and all the vendors just called all of their things AI.

Speaker:

And so now it's super confusing.

Speaker:

What is it that you're talking about?

Speaker:

So I almost never use the word AI in marketing and sales, cause you

Speaker:

have to, what was really innovative and that would really push this

Speaker:

forward was the LLMs, the LLM models.

Speaker:

Surprising even to the vendors who made them suddenly sounded really smart

Speaker:

and they sounded really convincing.

Speaker:

You can chat with it.

Speaker:

You can have like a normal seeming conversation, but you

Speaker:

have to just remember on the other side is like a weird thing.

Speaker:

That's just trying to predict what the next right word is.

Speaker:

It has.

Speaker:

Zero agency about anything.

Speaker:

It's just like a Vorg vomit machine.

Speaker:

it is a language generator.

Speaker:

What I found is that if you attempt to give it any sort of numbers to

Speaker:

interpret, it will be terrible at it.

Speaker:

Yes.

Speaker:

There are vendors who are working on this problem and open AI and

Speaker:

Gemini and all these people.

Speaker:

They really want to make it into a coherent, a system that will be

Speaker:

reliable and accurate, but today it is just a language generator.

Speaker:

you know what the problem is?

Speaker:

There's not enough language here.

Speaker:

That's when the LLM comes in, not enough language.

Speaker:

And so there are actually many cases like this.

Speaker:

I can, for example, say here's a paragraph turns into bullet points.

Speaker:

Amazing.

Speaker:

Does a really good job at that.

Speaker:

Here's a page turned into a PowerPoint.

Speaker:

Incredible.

Speaker:

Because it just takes language and just puts it into different places.

Speaker:

But it is terrible at math.

Speaker:

It is terrible at doing the analysis.

Speaker:

Except the very rudimentary.

Speaker:

So what we found is that the problem that we're working on is

Speaker:

how do we bridge that last mile?

Speaker:

How do we personalize analytics?

Speaker:

How do we tell the CMO a story in a way that is relevant to them while at the

Speaker:

same time using those same numbers to tell a slightly different story to the

Speaker:

campaign manager and a slightly different story to the rest of the company.

Speaker:

we need more language, what's important is that the underlying

Speaker:

data has to be the same.

Speaker:

All of those people need to get the same data, just phrased

Speaker:

differently, dressed differently.

Speaker:

And so, that's the thing that I think is going to be really powerful with it.

Speaker:

What we found is if you give it context, if you say this is a team of, event

Speaker:

marketers, and they meet every Thursday and you're creating a communication for

Speaker:

them for to prepare for this meeting.

Speaker:

And so therefore you should go look at, I'm going to give you tables that

Speaker:

have the best events from the past.

Speaker:

90 days and some of the highlights from the event.

Speaker:

And so I'm going to give you tables with actual real data in your job.

Speaker:

LLM is to describe the table and point things out from the table that I give you.

Speaker:

Do not, under any circumstance, create your own table or your own numbers.

Speaker:

Just take the data that I have and talk about it.

Speaker:

It does a really good job at it.

Speaker:

If you give it context, if you make sure to give it data to talk about, because

Speaker:

it will make up data if you don't give it data, and if you give it the specific

Speaker:

things that you care about, the story.

Speaker:

What's the story?

Speaker:

You literally tell it what the story is and it will say that story.

Speaker:

So, to prepare for that meeting.

Speaker:

Right.

Speaker:

You know, maybe the day before everybody receives a PowerPoint or

Speaker:

a document that says, Hey, everyone, I know we're meeting tomorrow.

Speaker:

Here's the data that we need for tomorrow's meeting.

Speaker:

Here are the best things here, the campaigns and here's what went well.

Speaker:

And here's what all the things that you would have said it will do it because.

Speaker:

Underneath of it, you already did the work of having, you have a dashboard

Speaker:

for where the data comes from.

Speaker:

It's already been cleaned up.

Speaker:

It's already ready to go.

Speaker:

What it's doing is just doing the communicating within the context.

Speaker:

And, it does it automatically.

Speaker:

It will do it every Thursday from that point forward when

Speaker:

it's coupled with that event.

Speaker:

Campaign ROI dashboard.

Speaker:

And that we found actually works well because we never

Speaker:

use it to do any analysis.

Speaker:

We just use it to mess with the format.

Speaker:

Is it a PowerPoint slide?

Speaker:

Is it a document?

Speaker:

And we also use it to contextualize the communication.

Speaker:

It's a communication tool, not an analysis tool.

Speaker:

there you go.

Speaker:

That's how you can play with it.

Speaker:

let me just dig deeper to make sure I understand.

Speaker:

You're bringing in raw data and you're producing a summary table, which is all

Speaker:

existing technology We can we can do that with the sales force report for

Speaker:

you know, 20 30 years, whatever fine You The LLM is able to take analysis

Speaker:

and communicate it in different ways.

Speaker:

Got that.

Speaker:

What I'm missing, the gap for me, and you're going to fill

Speaker:

it in I'm sure, is Um, what's actually producing the analysis?

Speaker:

Because you said it's bad at doing the analysis, but Someone has to

Speaker:

look at that table and be like, This is high and this is low, like the

Speaker:

thinking about the numbers, which you said that the AI isn't that good at.

Speaker:

Where does

Speaker:

Yes.

Speaker:

Very good.

Speaker:

Very good question.

Speaker:

What we found is that it's.

Speaker:

Another thing that's helpful to do is to tear out types of analysis.

Speaker:

it turns out that when you as an analyst are confronted with a dashboard or a

Speaker:

table, there is actually multiple things that you're doing in terms of analysis.

Speaker:

The first thing that you're doing is you're scanning.

Speaker:

You're scanning and you're kind of looking at low lights, highlights, things

Speaker:

that you should focus more on, less on, So really it's an editorializing job.

Speaker:

You need to pick out things that stand out, things that you're

Speaker:

going to highlight, or things that you're not going to talk about.

Speaker:

Because you can't show the whole thing.

Speaker:

You can't just shove the whole dashboard into somebody's brain.

Speaker:

until we get the matrix plug in and the USB connections, we can't do that, right?

Speaker:

that's the whole point.

Speaker:

Human communication is sequential, so you can't do it all at once.

Speaker:

It has to come in order, this is important, it's

Speaker:

like a slideshow or a movie.

Speaker:

Right now I'm talking to you sequentially, through time.

Speaker:

And so we are designed to receive information bit by bit by bit.

Speaker:

So what's going to come first, what's second, and what should we highlight

Speaker:

or not about each of the things.

Speaker:

So that's like level one.

Speaker:

Level two is like, Why is this happening?

Speaker:

Okay, the numbers are down.

Speaker:

I'm glad you caught that.

Speaker:

I'm glad you caught that the numbers used to be high, but now they're low.

Speaker:

Why?

Speaker:

That's level two.

Speaker:

That's much harder.

Speaker:

Because very often we need additional information.

Speaker:

It's no longer on that same dashboard.

Speaker:

You gotta go talk to Frank.

Speaker:

You gotta go talk to this other person.

Speaker:

You gotta go talk run more reporting.

Speaker:

Maybe you need to run some regression or something.

Speaker:

You know, it goes into some other direction.

Speaker:

so what LLMs are decent at is level one.

Speaker:

It will scan it and it will say you human dedicate your human resource,

Speaker:

literally your human brainpower to coordinate and do triage on this.

Speaker:

table here because this number is lower than what you told me.

Speaker:

You told me to tell you.

Speaker:

Anytime this number is lower than this other one you told me to tell you.

Speaker:

So here it is.

Speaker:

And so it's lower.

Speaker:

I don't know why, but it is.

Speaker:

So someone should look into it, so it will generate additional deeper questions

Speaker:

and it will to guide further analysis.

Speaker:

And the whole idea is that the marketing analytics team does not

Speaker:

have to do a level one analysis that looking at the dashboard, scanning it,

Speaker:

pointing out things that are either really good or really bad or normal,

Speaker:

and then sending that communication to whoever needs to receive it on a

Speaker:

timely basis, every single time that we can do, that's possible today.

Speaker:

We can execute that.

Speaker:

We're not yet at level two.

Speaker:

I think we're all working on.

Speaker:

Okay, so if it didn't work, why, why?

Speaker:

Send a survey to Frank.

Speaker:

I don't know.

Speaker:

I mean, I don't know how to, I haven't gotten to level two yet

Speaker:

and level three could be, you know, what should we do about it?

Speaker:

What, what does this mean?

Speaker:

these deeper questions are still going to be up to us humans to figure

Speaker:

out, but this level one is very much possible to take what was on a debt.

Speaker:

So that's the, that's the new kind of direction for emojis,

Speaker:

not just the email data.

Speaker:

But what we've found is that it's the general dashboard communication.

Speaker:

So what we're doing is.

Speaker:

We're ingesting your dashboard.

Speaker:

And so we take the whole dashboard that you built with your data on it.

Speaker:

And then we're saying, here's a tool to help you tell a

Speaker:

story about this dashboard.

Speaker:

What should we talk about first?

Speaker:

And you would literally say, first talk about this component because usually

Speaker:

they're designed to be scanned by.

Speaker:

Human in a certain order, and so first notice the overall

Speaker:

revenue numbers We're behind.

Speaker:

Oh, God.

Speaker:

Okay.

Speaker:

All right.

Speaker:

So in which region are we behind?

Speaker:

Okay.

Speaker:

It's this one, so say something about the region.

Speaker:

And then there's expectations and business context.

Speaker:

What does it mean to be behind?

Speaker:

What are we behind in?

Speaker:

Is it the MQLs?

Speaker:

Is it the ops?

Speaker:

Is it the revenue?

Speaker:

What are we talking about?

Speaker:

And so you, it is kind of an interactive story creation aspect that then

Speaker:

contextualizes the communication from that table and the LLM is able

Speaker:

to then ingest it, the table, the context that you gave it and produce

Speaker:

an output that's relevant for that moment in time or whatever that was for.

Speaker:

So that makes perfect sense, breaking it out into those levels.

Speaker:

so level one, more like a pattern recognition.

Speaker:

What's high?

Speaker:

What's low?

Speaker:

Makes total sense that it could be able to do that to get to level two to

Speaker:

level three Is it a limitation of the underlying language models today in other

Speaker:

words like open AI or whatever like?

Speaker:

Lower level of granularity tools, but beneath that need to up their

Speaker:

game or is it more folks like you?

Speaker:

Need to be building smarter software to address that Allow,

Speaker:

those models to do that next level or those next levels of analysis

Speaker:

it's a language generator.

Speaker:

It's just a word creation machine.

Speaker:

It has no additional context.

Speaker:

On how to solve a problem, a business problem at your company,

Speaker:

at the event that you just did, it has no idea what's going on.

Speaker:

So the only thing it's good at is like communicating what you already told it.

Speaker:

So if we, for example, go back to the analysis about the number of interactions

Speaker:

with the accounts, how would it do that?

Speaker:

could we make an automated way to tell that story?

Speaker:

That'd be cool.

Speaker:

So that the CMO doesn't have to ask me.

Speaker:

He can just schedule every Monday.

Speaker:

Tell me that story again.

Speaker:

It's the most beautiful story ever told at this company.

Speaker:

Tell me again, LLM, how I contribute to deal size, right?

Speaker:

Every night before bed.

Speaker:

I want you to tell it to me in a haiku form.

Speaker:

Go ahead, LLM.

Speaker:

It will do it.

Speaker:

It will literally do it.

Speaker:

give it to me in a sonnet.

Speaker:

And so, The way to execute on that is we'll need to give it that table

Speaker:

that that chart is made out of.

Speaker:

So you will have to have.

Speaker:

Interactions already pre calculated, deal sizes already pre categorized, and even

Speaker:

that line pre made, and even some talking points about what the slope of that line,

Speaker:

if that slope is positive, that's good.

Speaker:

Say positive things about it.

Speaker:

If that slope is negative, that's bad.

Speaker:

We want more interactions to mean bigger deal sizes.

Speaker:

Okay, so what if the additional questions asked are like, well, tell

Speaker:

me more about which type of titles better have a stronger connection?

Speaker:

Or is there certain companies that produce a better correlation, certain

Speaker:

industries where this resonates even more or even the reverse?

Speaker:

Are there situations where can you pick out accounts that

Speaker:

don't produce this result?

Speaker:

That somehow we're immune to marketing.

Speaker:

we could ask that question, but we need to have the answers ready to give it.

Speaker:

We need to do that analysis up front and say, here's an example of when it doesn't

Speaker:

work and have separate tables for that.

Speaker:

So the AI, the language model has no ability to do any of those things.

Speaker:

Like very custom analysis.

Speaker:

All it does is just verbalize what you tell it.

Speaker:

that's why I would never rely on it to do anything.

Speaker:

Do you know what I mean?

Speaker:

But you can orchestrate that kind of a conversation, if you will.

Speaker:

Like if you were chatting with an AI.

Speaker:

if you said, if someone asks you this question, I want you to go and

Speaker:

look at this other set of tables.

Speaker:

If it's within the context of that conversation window, maybe the

Speaker:

context is we're going to talk about marketing interactions and deal size.

Speaker:

And in that window, it has access to all of that data that you

Speaker:

already pre selected for it.

Speaker:

It can definitely tell that story.

Speaker:

Absolutely.

Speaker:

And it can go and say, Oh, also I wanted to highlight companies

Speaker:

where they're immune to marketing.

Speaker:

Here they are based on this table over here that you can go look at yourself.

Speaker:

Human.

Speaker:

Taking a slightly different example to explore this further.

Speaker:

Let's say every week I do a report out of marketing performance

Speaker:

I look at my overall number.

Speaker:

We, we look at SQLs and I say, how are we doing?

Speaker:

Are we on pace?

Speaker:

If we're not on pace, you know, we break it down by market.

Speaker:

And we say, this market is behind pace.

Speaker:

And if this market is behind pace, then I break it down by channel.

Speaker:

And then we look at conversion rates and you know, I don't need to tell you this,

Speaker:

but there's like a series of things.

Speaker:

It would seem quite possible, assuming we provided a persistent set of data to

Speaker:

AI that it could perform that and that we could train it to perform that analysis

Speaker:

up and down the chain as many times

Speaker:

as it

Speaker:

absolutely.

Speaker:

Will you describe so far?

Speaker:

No problem.

Speaker:

Keep going.

Speaker:

What else are you,

Speaker:

that task has now been automated.

Speaker:

I would like that by the way, that would, be very helpful because

Speaker:

it's, it's very time consuming

Speaker:

and you need to remember to do it, et cetera.

Speaker:

Now let's say, Oh, so getting into the why, Hey, it's, it's labor day.

Speaker:

We, you and I are having this conversation on a labor day.

Speaker:

So we're spending part of our vacation in North America.

Speaker:

This is a holiday.

Speaker:

So we would expect that performance for this week in North American markets would

Speaker:

be Lower proportionally by, by one day.

Speaker:

So 20 percent of the week could AI figured that out if it was aware of the

Speaker:

calendar and when the holidays were.

Speaker:

That's a good question.

Speaker:

That's where, probably not.

Speaker:

unless you asked it to, probably not.

Speaker:

So that's crazy to me because like, that seems like a pretty basic and I'm,

Speaker:

I don't consider myself the world's greatest analyst at all, but I could

Speaker:

arrive at like, Oh, we had one less day.

Speaker:

or would we have to tell it like check the days of the you know Which all

Speaker:

of a sudden starts feeling very rules based again Which was the whole the

Speaker:

magic that valance was supposed to be that it doesn't have to be rules based.

Speaker:

It learns it trains itself, etc

Speaker:

that's not true.

Speaker:

The magic of LLMs is that it has more and more context to generate

Speaker:

more, plausible sounding words.

Speaker:

I wouldn't conflate LLMs with some sort of a statistical machine learning algorithm,

Speaker:

which is a different type of AI.

Speaker:

That basically generates some sort of coefficients that connect some

Speaker:

things to others, that whether Labor Day is correlated to performance.

Speaker:

And over time, over many years of Labor Days, it will find that it

Speaker:

does, and therefore definitively have a positive coefficient that

Speaker:

says Labor Day impacts performance.

Speaker:

And it is that coefficient that you would feed into an LLM to then say, When

Speaker:

that coefficient is positive, say that Labor Day contributes to, so what I've

Speaker:

seen cleverly done with some vendors that I think we have an opportunity to

Speaker:

do and anybody who's doing this sort of, like, for example, uh, you, I remember

Speaker:

when we talked about email, you asked Grant, this email statistics stuff is

Speaker:

great, open rates, click rates, but like, Can you go deeper into why these

Speaker:

emails are performing well or not?

Speaker:

Can you like analyze the subject line and the content length?

Speaker:

And was it sent on a Tuesday and who was it sent to?

Speaker:

What are their titles?

Speaker:

Where do they live?

Speaker:

Do you know what I mean?

Speaker:

I was like writing it down, being like, I can't do all that.

Speaker:

What are you talking about, Justin?

Speaker:

but well, you could do that.

Speaker:

You could take all of those as causal factors, put it in a

Speaker:

regression and just run that.

Speaker:

as often as you like, build an machine learning algorithm.

Speaker:

LLM is not that, that is not the kind of AI that you're looking for.

Speaker:

That's a different type of statistical modeling software that would take causal

Speaker:

factors and arrange them by coefficient of influence, which one has the highest

Speaker:

and which one has the lowest, and it will spit out a table, a very confusing

Speaker:

looking table with lots of decimal points.

Speaker:

That someone has to explain to us mere mortals, what are

Speaker:

we supposed to make of this?

Speaker:

Well, it turns out that a big contributor to email success is

Speaker:

the length of the subject line.

Speaker:

How many characters it has.

Speaker:

And so that is a big determining factor.

Speaker:

I see that your specific email had a really long subject line, so perhaps

Speaker:

that's why it didn't do as well.

Speaker:

So you would have those coefficients to talk more deeply about the performance.

Speaker:

So you would have to give it the coefficients.

Speaker:

And then it

Speaker:

would have that in it's yet another table to generate more words about.

Speaker:

what i'm hearing is that it's almost like multiple different

Speaker:

types of ai working together

Speaker:

that you would need to orchestrate somehow at the software level

Speaker:

with like You know, math, ai, like machine learning, AI generating

Speaker:

Yes.

Speaker:

just think about like open AI their thing is all about making a thing

Speaker:

that is roughly like human like.

Speaker:

It lacks business context.

Speaker:

So it doesn't know anything about your company.

Speaker:

So you have to teach it about your company and the language that you speak

Speaker:

there, what widgets are we selling?

Speaker:

Are we counting MQLs?

Speaker:

What's an MQL, all that kind of stuff.

Speaker:

You have to give it that context.

Speaker:

You have to give it access to data and then you have to tell it like what

Speaker:

sort of things to do with the data.

Speaker:

And they're not going to make very niche things.

Speaker:

so underneath that machinery, when they get a table, they're going to run actual.

Speaker:

Run of the mill statistical analysis on it, they're going to find the mean,

Speaker:

they're going to sort it, they're going to do all sorts of things to the table that

Speaker:

they think is normal to do to a data, to any data table that any user gives us.

Speaker:

Generate those tables that you can then talk more about so they will guess

Speaker:

at what you are probably trying to do and they will probably get it wrong

Speaker:

because it's just a generalized smart system, you know, it wasn't made for

Speaker:

your marketing team and your company.

Speaker:

So that's why I'm not relying on them getting that right.

Speaker:

Not yet, maybe in five years to 10 years, that, depth of knowledge of

Speaker:

statistics and domain knowledge will be deep enough that it's like, just hook

Speaker:

me up to GA for Marketo and Salesforce.

Speaker:

And I'll just generate insights.

Speaker:

You know what I mean?

Speaker:

Maybe it will get there.

Speaker:

but today you can approximate a very intelligent sounding responses

Speaker:

as an by feeding it what you want it to talk about and telling it.

Speaker:

Who you want to say what, and having that automate some of your analytics function,

Speaker:

the most basic ones, the ones that are most time consuming and boring, the

Speaker:

scanning, the routine updates that can be automated, you will become the tier two

Speaker:

analyst, you will proactively see where the things are up and where the things

Speaker:

are down, and then maybe chime in and be like, Oh, I already did the analysis

Speaker:

of why I talked to the right people and it turns out it's Labor Day, guys.

Speaker:

It was Labor Day all along and that's why.

Speaker:

Everybody calm down!

Speaker:

The numbers are down because of Labor Day.

Speaker:

And they'll be like, oh wow, that was the context that we were missing.

Speaker:

by Already we have apps where you can upload your own docs.

Speaker:

So I know like some writing apps, like copy AI, you can upload your own brand

Speaker:

guidelines and style guidelines and stuff,

Speaker:

and then it will generate.

Speaker:

Things with that context in mind, how far away are you?

Speaker:

This sounds kind of crazy, but it, as we're talking about, like the,

Speaker:

model is only as good as the context.

Speaker:

You know, the way that you onboard a human analyst into a business, you

Speaker:

could onboard an AI agent, have it read every email, attend every meeting,

Speaker:

read every post and every system and every slack message, it would generally

Speaker:

or gradually, generate that context for itself, absorb that context.

Speaker:

would it then be able to use that to provide

Speaker:

Yes.

Speaker:

I think all of the big vendors, they are racing us.

Speaker:

Fast as they can to create this agent future.

Speaker:

I mean, Microsoft tried to include the agent as part of

Speaker:

the operating system itself.

Speaker:

So you would just watch all of your windows and everything that you do.

Speaker:

And the users were like, no, don't do that.

Speaker:

Uh, you know, I need a way to like turn you off some of the contacts.

Speaker:

I don't want you to know about.

Speaker:

And absolutely.

Speaker:

I mean, if you could manipulate your screen, if you could just click

Speaker:

into the dashboard, it could click, you could just watch you work.

Speaker:

We're a little bit, um, watch everyone at the company work and it's the same

Speaker:

agent really, you know what I mean?

Speaker:

Um, that knows

Speaker:

what everyone, it's kind of creepy.

Speaker:

I don't know,

Speaker:

it's incredibly creepy.

Speaker:

I mean, it's, it's like where your brain goes, like, how

Speaker:

would you solve that problem?

Speaker:

And then you sort of stand back and kind of horror at the dystopian aspect of it.

Speaker:

But,

Speaker:

the white collar job is this level one.

Speaker:

So that's why I Tell the kids is that, guys, level one

Speaker:

stuff, that's not gonna cut it.

Speaker:

Scanning, oh this number is high, this number is low, here you go boss.

Speaker:

We can't do that.

Speaker:

it's a matter of time that that is automated and you want it to be, and so it

Speaker:

will become more like, what does it mean?

Speaker:

What are we supposed to do?

Speaker:

But it's going to do that too.

Speaker:

a better job than we do.

Speaker:

In many ways, because it will be more reliable.

Speaker:

It will not get tired.

Speaker:

It'll not just start checking Facebook all of a sudden as we tend to.

Speaker:

So, absolutely.

Speaker:

That's so I try to squint and I'm like, Where is this going?

Speaker:

What's it going to be like?

Speaker:

So that's why that story I told about this agent future, where there is some

Speaker:

system, that has the business context for work, I'm hoping it's separate from home

Speaker:

and work, whatever, but it's like you join a company and there's this guide and

Speaker:

says, Oh, hi, you've joined this company.

Speaker:

Let me show you what's important.

Speaker:

What's not important.

Speaker:

I'm going to tell you, I'm going to be here every day with you.

Speaker:

And I'm going to.

Speaker:

help on board you.

Speaker:

I'm going to help, with communications with the rest of the team.

Speaker:

Oh, this person over here, you know, this is what they need now.

Speaker:

And this person needs this.

Speaker:

And this is what you're supposed to read and give your feedback here's my feedback.

Speaker:

let's have a discussion about it.

Speaker:

so I try to play those games in my head, like, gosh, what's it going to be?

Speaker:

Like, and very quickly, I'm like, We don't even need that person.

Speaker:

Do we?

Speaker:

That's, uh, yeah, the, the, the benign fantasy is that like, we're just

Speaker:

going to be there and AI is helping.

Speaker:

And then the, the reality is at least there would, there's no

Speaker:

way that there isn't many fewer people doing these jobs than

Speaker:

They're doing a very different job.

Speaker:

so that's why I'm like, okay, I'll let everybody else do these like

Speaker:

futuristic, big picture predictions.

Speaker:

There's many podcasts you can go listen to about all sorts of revisions of

Speaker:

what that future is going to look like.

Speaker:

I'm like, can I just work on my analytics things?

Speaker:

And what does that look like today?

Speaker:

The reality is, is that the amount of infrastructure and effort it

Speaker:

takes to produce a decent data set to even consider looking at is so vast.

Speaker:

Like I think about all of the integrations that need to go into

Speaker:

some sort of a data warehouse, all of the data engineering of connecting

Speaker:

everything, defining everything.

Speaker:

And every company has very unique things.

Speaker:

From products to, you know, divisions and industries and the way that they sell,

Speaker:

all of that is custom made by humans.

Speaker:

They will have their little helpers to help them along the way, but

Speaker:

that custom bringing together, I can't see an AI just taking over.

Speaker:

So there is lots of data engineering stuff that will still be done

Speaker:

very manually, very deliberately.

Speaker:

And then there is, the creation of these, like, dashboards.

Speaker:

They're human.

Speaker:

What's not changing?

Speaker:

I guess with, like, Neuralink, I guess it is changing.

Speaker:

But our ability to consume information is still, we're still humans.

Speaker:

That we were, thousands of years ago, and so someone will need to

Speaker:

explain to us what the hell is going on to the ones that are left.

Speaker:

so that's why I picked the last mile.

Speaker:

That was the most interesting to me.

Speaker:

I was like, can we be better at this?

Speaker:

Can we, instead of doing a slideshow, can I do a little movie?

Speaker:

Why can't I make a little animation about, uh, when you log in, you know,

Speaker:

maybe it's like a little animated thing about how your campaigns are performing.

Speaker:

Maybe it's like a TikTok video about, marketing attribution.

Speaker:

If that's what's engaging.

Speaker:

So I was thinking of more of What content captures attention because what

Speaker:

we have limited amount of is time spent and what are we paying attention to?

Speaker:

Who cares about this?

Speaker:

And so I want to make analytics content that's, engaging, useful,

Speaker:

that captures attention, that then drives some decision making.

Speaker:

And I was like, okay, the AI tools that we have today can help

Speaker:

tremendously with this effort.

Speaker:

let's get on with it.

Speaker:

that's why I'm so excited about the future and not really down on it because at best,

Speaker:

before we're all fired, we will have a few decades where We're interacting with this

Speaker:

AI and it's helping us with work and it's driving us to more strategic conversations

Speaker:

and the work is less mundane.

Speaker:

and then when, you know, what happens 50 years from now?

Speaker:

I have no idea.

Speaker:

mean, by the time we're all fired, maybe we're in a world with no

Speaker:

resource constraints and the notion of needing to work for, for

Speaker:

money becomes obsolete as well.

Speaker:

And so

Speaker:

Yes, before they turn on humanity, it will be like paradise.

Speaker:

but then it's back to Terminator,

Speaker:

right?

Speaker:

depends which genre of science fiction you like to watch.

Speaker:

I want to be, I want to be in a trolley with the Slurpee.

Speaker:

That's what I want.

Speaker:

the Wally versus Star Trek and I, you know, 'cause in Star Trek it's

Speaker:

just like, oh, we, we don't have any constraints in, people are sort

Speaker:

of more, fully self-actualized.

Speaker:

And then Wally, you're in the

Speaker:

In the Matrix.

Speaker:

in.

Speaker:

The color of the day is red and you're sure it changes.

Speaker:

that's a fascinating, thought, journey there.

Speaker:

So thank you for riding with me, along on that grant.

Speaker:

Maybe just in closing, I want to ask a little bit about, uh, about you,

Speaker:

your motivations, your repeat founder.

Speaker:

I think as everyone can probably see, you're just like a guy that likes

Speaker:

thinking about these innovative problems.

Speaker:

You're sort of drawn to these things, moth to a flame.

Speaker:

Um, what's the motivation for you?

Speaker:

Certainly not the most like stable and like relaxing life as a startup founder.

Speaker:

What, what keeps drawing you back in this direction?

Speaker:

Yeah.

Speaker:

tech.

Speaker:

when I graduated college, I had like a normal job in a cubicle.

Speaker:

wear shoes every day and I had to go there in person and ride the elevator.

Speaker:

Yeah.

Speaker:

I say that as I sit here without any shoes.

Speaker:

Oh, my God.

Speaker:

The dream has come true.

Speaker:

Yeah.

Speaker:

and I found myself just, I wanted more.

Speaker:

So I remember 1 of the 1st things is I started a tutoring company and I'm

Speaker:

so we were doing hired some tutors.

Speaker:

We did that.

Speaker:

then I was part of a beekeeping company.

Speaker:

I did a marketing and selling of honey.

Speaker:

It was really amazing, probably a different podcast, but I found this

Speaker:

amazing beekeeper that produced honey of different colors because the

Speaker:

color of the honey changes depending on what pollen source the bees went

Speaker:

to.

Speaker:

And he was a migratory beekeeper.

Speaker:

And so all these different honeys were all the different flavors and

Speaker:

we like paired them with cheese.

Speaker:

And I would cold call like Kroger and cheese shops around the country and, sell.

Speaker:

Honey.

Speaker:

Anyway, I loved it.

Speaker:

I loved all of that.

Speaker:

For some reason, I was just always drawn to that.

Speaker:

And so when I found tech and combining that with, uh, entrepreneurship

Speaker:

off, I I was like, okay, there's ability to innovate here.

Speaker:

that's what I want to be doing.

Speaker:

And when I found these thorny Salesforce Marketo analytics questions, I immediately

Speaker:

was hooked because I was like, Oh, this is, first of all, everybody wants to know

Speaker:

the answers to some of these questions.

Speaker:

Questions and everybody's doing it in their own way.

Speaker:

but they're all kind of asking.

Speaker:

So when I became a consultant, that's when it became really crystal clear for tech.

Speaker:

Because as a consultant, you get a broad but shallow view into many

Speaker:

companies, which is incredible.

Speaker:

If you want to build products.

Speaker:

Because you suddenly see the commonalities across many different companies and you

Speaker:

say, wait a minute, these people all hired me, that is, they're willing to

Speaker:

pay to solve this same set of problems.

Speaker:

Can I like scale myself in some way?

Speaker:

And so then you start thinking about carving out what it is that you

Speaker:

did and finding a solution for that one specific thing and off you go.

Speaker:

so after the first company I was hooked, I was like, okay, let's do it again.

Speaker:

Step one, become a consultant.

Speaker:

When you become a consultant and I still consult.

Speaker:

You experience, first of all, you're on the ground and the person across from

Speaker:

you says, Grant, we got to tell a story.

Speaker:

What's the story with this data?

Speaker:

Can you help me?

Speaker:

And you're like, okay, you know, roll up your sleeves.

Speaker:

Show me what you got.

Speaker:

where's all those interactions.

Speaker:

How big are your deals?

Speaker:

Let's go do it.

Speaker:

And then in the back of my head, I'm constantly like, okay, okay.

Speaker:

This is what they're, this company is doing this other company.

Speaker:

This is what they're doing.

Speaker:

And with Moji, that was the thing that was the frustrating thing is like.

Speaker:

We made this amazing dashboard, we're onboarding everybody on it.

Speaker:

We're trying to explain what it means.

Speaker:

And inevitably somebody asks, Oh, Grant, thank you.

Speaker:

Like the colors of the dashboards look great.

Speaker:

I see that you spend a lot of time, but like, what are

Speaker:

we supposed to do with this?

Speaker:

And I was like, ah, I failed.

Speaker:

And you know, they don't know how to use this data.

Speaker:

That's the part that's missing.

Speaker:

So then it's when I found myself doing that over and over again,

Speaker:

and I was like, okay, maybe we need to focus on that part as well.

Speaker:

Well, it is a fascinating problem, and I agree with you, a very,

Speaker:

valid and important problem.

Speaker:

So, excited to see what you continue to cook up over at Moji for folks

Speaker:

that want to, uh, Check out more.

Speaker:

It is M O G I dot A I on the web.

Speaker:

Grant, correct me if I got that

Speaker:

wrong.

Speaker:

go over, take a look.

Speaker:

You can request a demo if you want to see a bit more in person.

Speaker:

You can also follow Grant on LinkedIn.

Speaker:

Just look up his name, Grant Gregorian.

Speaker:

Uh, Grant, always, uh, always a blast, as I said.

Speaker:

Had a lot of fun during this chat, so thank you for coming on.

Speaker:

I had so much fun too.

Speaker:

Thanks for having me on.

Links

Chapters

Video

More from YouTube