The Marketer Show

In this episode, Jordan Crawford joined to chat about:

AI in Marketing and Sales

- AI will significantly change how we think about marketing and sales
- It is important to experiment with AI tools and find ways to work with them to expand capabilities
- Both the cost and time to perform tasks with AI have gone to near zero
- The challenge is asking AI questions that don't include nuance and can't be operationalized
- To work effectively with AI, we need to break down work into smaller chunks that it can do well

Using AI Effectively

- The focus is on how to use GPT Four in processes without worrying about its hallucination problem
- People who are not reinventing themselves are at risk of being replaced by someone using AI.
- To use AI effectively, one needs to provide creative constraint and context. Andy Grove's framework of "if the board fired me today, what would my replacement come in and do?" can be useful for thinking about how to use AI.
- AI is good for operationalizing, but one needs to work backwards and feed it the right inputs and training sets to make it work effectively.
- The speaker's company uses AI for programmatic outbound campaigns, feeding it website context to determine the top ten data points for targeting customers.
- Ideas generated by AI include employee turnover rates, gender diversity ratios, and presence of employee research groups.
- The speaker believes that in order to define prompts, one needs to invert their thinking and break down the process into discrete steps.
- By breaking down the process and having a clear understanding of the steps involved, the speaker is able to evaluate the AI's suggestions and identify which ideas are good or bad.
- The speaker recommends dissecting one's process by writing it out on a whiteboard to gain clarity.
- The speaker has campaigns using AI in a targeted way where they can trust the output.
- The challenge in automating this is that a wrong result can have high stakes and the pain is too high if it gets it wrong.
- The speaker is thinking about creatively constraining the automation to ensure that if it gets it wrong, it's a harmless mistake.

Using Creative Constraints with AI

- The speaker played around with an AI-generated sentence, which seemed like a human wrote it but was actually a lie, and warns against the weaponization of such technology.
- Many people consider AI as a productivity hack, but the author thinks that the real opportunity is using it to do things that would have been impossible before.
- The author has created a product that uses AI to identify the perfect moments for a company to sell their products and services.
- The author uses AI to write effective sales emails quickly and efficiently, using creative constraints and contextual understanding.
- This approach can be done at scale and can give companies a significant advantage over competitors who don't have sales teams of similar size.
- The speaker doesn't trust AI to write a complete email or to act on their behalf. 
- They believe that AI can't write a great email, except if it's trained on how to write five discrete sentences with specific steps and then summarize the email. 
- They don't trust AI to handle anything that involves requiring truth. 
- The speaker thinks the problem with AI is that people don't know how to use it or deploy it effectively, and there is a need to ask the right questions and to refine the process. 
- The speaker recommends clay.com as a useful tool for refining data by having conversations with ChatGPT on a per-row basis.
- The speaker has spent a lot of time tinkering with the tool to test its capabilities and come up with use cases.
- The speaker advises others to play around with the tool and think about what perfect thing involving language they could accomplish with unlimited time, and then consider how AI could help them achieve it.
- ChatGPT is a tool that can help you break down complex questions and tasks into smaller, discrete steps.
- The best approach is to ask the tool to give you the smallest possible steps to accomplish a goal, even if someone without context on your business could understand.
- By following these steps, you can audit them at every point and make sure they are correct.
- The possibility with ChatGPT is not just to save time but to have a person work on a creative task with nearly unlimited time, energy, and focus.
- The concept of constraint can help people go deeper and achieve higher output than they normally would.
- The guest gave an example of a big process they have built using constraint.
- The guest is asked if there are other examples of their process that have been automated and systematized due to constraint.
- The speaker has operationalized AI for a specific process, but only for pieces of it so far.
- They believe that the bar for successful AI implementation is high because it can begin "hallucinating" and making things up when given too much freedom.
- An example of this is when AI was tasked to build SEO pages for 10,000 companies without creative constraint and started making things up.
- The speaker believes that AI needs creative constraint and to be broken down into smaller processes to be successful.
- They are aggressive about killing processes that don't meet their standards and advocate for breaking down tasks into smaller, explicit steps similar to talking to a fifth-grader.

What is The Marketer Show?

Uncover hidden gems and insights from the world's greatest marketers.

Today. We're going to talk customer pain. We're going to talk

AI, we're going to talk b to b messaging sales. There's a

lot that's going to go on here today. Thankfully, we have Jordan Crawford here

to navigate it. Jordan, thanks for joining. And before we really dive

deeper, we'd love to just know kind of what you care about in

marketing right now. What what are the things that are keeping you up at night

or or that are exciting? You like, what what are those things that you really

deeply care about in marketing? Oh, boy. This is a

great question. And I would say that that answer to me is the

speed at which you need to be innovating now.

And specifically, AI is going to really change

how you should be thinking about marketing and sales.

And so I think it's absolutely important for you to be experimenting

with these tools, because AI is coming for my

job. Coming for your job, and we need to figure out ways to work

with it to drastically expand our capabilities.

And basically both the cost and time has

gone to near zero to do a lot of tasks.

And the challenge people have now is they're

asking AI questions that don't include

nuance and that they can't operationalize. And so

really, you need to be able to find ways to break down your work into

smaller chunks that AI can do exceptionally well.

So that's kind of my big focus right now, is how can

I take specifically GPT Four into a lot of my

process and do it in such a way that I don't have

to worry about its hallucination problem? And that's by

basically constraining the questions that I ask

of AI and the expected output.

Yeah, I think we're still at a point a place where even, like, early days

of search engines, it was really basic prompts or basic searches

weren't really harnessing the full power of this, but this is, like,

ten times more important, prompting an AI prompt versus a search.

Even so, I'd love to dive into that. You did talk about kind of

the risks that we might be heading into with replacement, things like that.

Where do you view right now in terms of marketers and AI,

that relationship? Who's at risk and what should we

be doing? Boy great question. Well, I think when

you talk about who's at risk here,

I really believe that you will be replaced by someone that uses AI.

Not an AI will replace you, certainly not in the next six months to

a year. And I think the people that are at risk are

not trying to reinvent themselves.

And this, unlike any time in the last ten years or at least five years,

is the time to reinvent yourself. Andy Grove, the previous CEO

of intel, had a really great framework with this, which is, if the

board fired me today, what would my replacement

come in and do? And that type of thinking

is the thinking that you need to be able to figure

out how should I be using AI,

what am I doing today that is inefficient and

how can AI help me? And to give you a sense of the

framework with which I use here is AI is

good with creative constraint,

with context. So you have to have those three things to

make AI really work for you. So it is amazingly good

at being creative as long as you constrain it and you provide

it the right inputs and training set to be able to do that.

Now, operationalizing, it is a different thing, different problem.

You can do all of this just inside of Chat GPT.

You can paste in the context, you can paste

in the sort of constraints. But to

do this well, you need to work backwards. And so the first

thing this weekend on Sunday, I'm giving a talk about how to use

AI for a company called Signals. And I

asked the question of AI, what questions would I ask if

I had AI replace me in the creative process?

And so my company does programmatic outbound,

so we'll help folks with large data sets, send programmatic messages,

and we've started to use AI for a lot of our internal campaigns.

And so my hypothesis is that really great targeting

data is the key differentiator between a good outbound and a

bad outbound campaign. And a lot of people have bad data. I'm not talking about

Apollo or zoom info. It's not good enough, it's not useful enough. So to

give you a concrete example here,

one of my customers is a company called Bonfire Women. They help women get

promoted at an organization. So I fed it the

company's web page and I've already done a lot of work for them, strategy work.

What should we target? Who should we target? When should we target them? So I

pasted in all of Bonfire's website and I

didn't ask GBT to write an email for me. I said,

here is the context of a website. If I had to score a company

from one to ten to determine if a company would be a good fit for

Bonfire services, what are the top ten data points I would use to

define who a perfect fit customer is? Pretend you had

perfect information on every company. So this is the lens that I

use when a client comes to me. And it came up

with ideas that I originally pitched. So I didn't provide any other context,

I didn't tell it who I was or other types of ideas.

I gave it context and that's all I said. And here are some answers to

give you some ideas. Gender diversity ratio. A lower ratio

of women to men in leadership positions would indicate a greater need for Bonfire services.

We actually ran this play. Employee turnover rate, especially for women.

High turnover rates, particularly among female employees. Which suggests the company could benefit from Bonfire's

talent and development retention services. There was a couple of bad ones in here

that were like, fine. They were just like a little too generic.

Public commitment to sorry. Presence of employee research group.

ERGs companies with existing or planned women focused ERGs

might be more interested in supporting. We ran this play, we look at historical job

descriptions for the keywords women at so we were able to determine

who are the companies that actively have an ERG employee research group?

So it was very creative. I constrained it and I gave

it context. Right. I did not say AI do my job for me.

So that was like one piece of the process that I pulled apart

so that I could have AI do different

pieces. And as I keep thinking about my workflows over

and over again, I am trying to dissect each piece and

inject AI with creative constraint,

with context. Is that helpful? Definitely, yeah. So if

we were to simplify that whole process just based on principle because I know that

every use case is so different in marketing totally just based on your

example, if you were to kind of dumb it down to a few core principles

of how to make a great prompt for that. How would

you simplify that in very clear, bullet point terms?

Boy, I'm going to answer the question I

wish you asked me, which is a politician's response.

But really the question I wish you asked me is like, what is the process

to start defining the prompts that

you should give? And I think that you need to invert

your thinking to get here, which is take what

you're doing today and write it out in as many

discrete steps as possible. So in my world,

it's like, understand who the company is, then once I understand who

the company is, I want to know who they target. Once I know who they

target, then I can think about my data sources.

Once I think about my data sources, then I can think about creative ideas

to come up with. Once we have creative ideas, then I can help with messaging.

And so I have those steps in my head and they

have never been sort of spoken out loud. And now that they are spoken

out loud and I have a process for them, I can zoom

into one particular part, which is take context on the company and

figure out what I would do. And now I have a small unit

of value that I can get GBT to help me with.

Right? And the more of those small units of

value I can string together, it helps me evaluate

what the AI is doing. So I know I look at these things and because

I know this customer really well, I'm like, bad idea, bad idea, good idea,

good idea, good idea. And there are enough good ideas because I have constrained

the problem well enough. So you need to dissect your

process. Just write it out on a whiteboard, just write down exactly what you're doing

and that should be enough. Yeah, I want to

get to the.

Operationalization I have the same problem. That's so hard.

It's so hard. Yeah, that's a big one. I want to get to that in

just a second. But really quickly, I would love to just kind of ask based

on everything that you have put together, process wise,

prompt wise, using AI for all your use cases you just mentioned, for those discrete

uses, do you feel like you're doing much

more work but much more efficiently? Or do you feel like you're doing

the same amount of work, but just ten times faster? So are you increasing the

amount of work you're doing, but you're still kind of like working the same amount

now and just doing more or working the same amount with far less

required of you to do that? I think actually

this is not a productivity hack,

which is weird because that's how a lot of people think about it.

They think about this like, oh well, I can take what I used to do

in 10 hours and do it in two. Actually, the huge

opportunity is what would you have done

with 100 hours that now you can do in one?

So something that you would have never done, you just would never have done.

Like, for example, one of the campaigns that we're running,

it takes in a company description and nothing else. It identifies

who their buyers are. And then in my case, it provides

basically a Google search for how you would find the exact

moments that those buyers would need you, which is like what my product does.

My product can sort entire markets worth of jobs data to find

the perfect moment for you for the companies that you target.

And so I basically created Google Searches

programmatically to thousands of companies

that I could have done probably 5

hours per company, right? But now I can send that because

I have creative constraint. And to give you an example, there's a good

friend of mine, David, that runs marketing at Kentech, which is a network observability

company. I know them pretty well. I have some understanding

about what they do. I did not write this sentence. I'm going to read you

a sentence that I did not write that AI wrote. But this is what

I would do if I had a limited time to write a sales email.

And this is an example of one of those search queries. So they do network

observability. So they tell you if your network is down and what pieces

are down and why it's down, et cetera. So show me

job descriptions for Network Operations Center NOC

managers that mention challenges in network performance monitoring,

maintaining network health, improving visibility across cloud and on

premise infrastructure, and ensuring fast troubleshooting for networks.

So the AI put together that search without

like it invented the title, which is like, right,

it invented the problems that exist in the title. And then it basically wrote

a Google search query for how you

would search inside of my system for the problems that they solve.

So amazingly great email that an account executive could

send one a day, I don't know, two or three a day.

So I think the greatest opportunity in AI is how do you apply

that creativity, constrain it, give it context,

but then do that at scale. And that's how you stay.

You just have a gigantic leap among the people that don't have

sales teams of 100.

So let's get into the operations side of things. You had mentioned you've

strung together some of these processes. Do you feel like you're pretty

far along there or still early days of building all the processes

you would still want to do? Oh boy. Every time I

log in to this tool, I think I'm

the problem. And I don't really feel that way about much software. I log into

much software. I'm like, damn, this dumpy thing doesn't do this, doesn't do this.

But this chat box is a particularly frustrating

thing because I think I'm

just not creative enough to be able to unlock its full potential.

And so I think that if I think about where my journey

has to go, it's way further.

But I do think I'm running far ahead of the pack.

But that is not because I have any particular insight.

I just spend so much time tinkering

with it to come up with these use cases, to test

it, to figure out what it does wrong,

what it does wrong at scale. And so I think

that really the best advice here is play with this stuff.

And you can play just with context inside of Chetty BT and

then figure out you should ask the inverse question, which is not how I should

use AI. It's like if I had unlimited time to do

the perfect thing, what perfect thing that involves language?

What perfect thing would I do? And then how can AI

help me get there? Yeah,

for sure. I love the concept of allowing it to help you

go much deeper than you ever possibly would because of the amount of effort

and constraint you would have normally. But now you can go so much

deeper. So it's not necessarily you're working 2 hours a

week, you're just going much deeper, putting much higher output

out. And I'm sure you're finding that across the board. You'd already given

one example of kind of a big process that you've kind of built now where

you've got this really impactful task you're able to accomplish. Are there other examples of

that in your process now where it's basically taken

over something that you would have been doing manually forever, and now you

just have this full system for doing it. Through AI not

the full system yet. I think I have just started a

piece apart. I just have plucked pieces of

the process and operationalized AI, because I

think that the bar for this thing

to successfully do something that I would hit send on

is high, which is that the second it starts hallucinating.

And let me give you an example. I took the three

most recent sales and marketing jobs in my database,

and I wanted to build SEO pages for every

company's ICP. So I wanted to say, well, marketer hire

sells to VPs of marketing at companies between

these sizes and that are trying to scale their paid ads or SEO

efforts, et cetera. It's not a crazy hard

thing for a person to do, but really hard at large scale.

So a perfect example of where AI would be good here,

but because I didn't give it creative constraint, and I let that

thing rip on about 10,000 companies, it just started making

shit up. To be fair,

if you gave a really smart person just the information I

give, they might like without I mean,

basically what you're trying to do is you have a fifth grader,

maybe a fourth grader, and then you give them the

Xanos glove or whatever, right? And you're like, use it responsibly,

and he's like and it's like, no, you're snapping too

much, way too much. And so you have to be like, okay,

no, do this very one thing, and test

it, and test it and test it, and then sort of scale it. And so

that's kind of how I'm approaching this process. And when I let it run amok

and I find two or three good examples, then I have to sort of pare

it back. At the end of the day, I have

to ship something that I'm proud of, and if I'm not proud of it,

I kill it. And so I'm pretty aggressive about killing stuff, but I

dip my toe, I scale it. If I review a

lot of these things, and it looks silly, the problem was that I gave it

too much creativity, and I didn't constrain it enough,

and I didn't break it down into smaller processes. So if

I had to do this again, for example, I would say, look at this job

description, and is there an ICP here? Do they talk about

who they target? Yes or no? If they do, where do

they talk about it? Great, now extract that information. Now with that information,

go do this. So really, you're talking with a fifth grader,

and you need to be explicit. And if you can break down something

that you think is one step into five steps, the AI will do much,

much better. Yeah, I think that's

a big downfall of a lot of results you see out there is just trying

to do too much with too little context. When you're looking at the scope of

if you're trying to actually build a process for this, trying to build something that's

going to be repeated many times for your business. How do you think about the

scope for a prompt or the scope for a specific query?

How do you know if it's too big or if it's too little?

How do you kind of land on that so that you kind of have some

level of certainty that you'll actually get what you wanted from

the results? I think that the

best thing to do might even be to have it impersonate

someone that knows nothing about you, your business or your context, and start

there and ask it a complex question and have it break down the

steps for you and then use those steps back in

this process. So it's like, what you might even and this is why this tool

messes me up. It's so meta. It's like you can ask it to help you

break down the process so that you can use it better. It's like, where does

it end? How far does this thing go? So I

think that the best approach is like, I'm trying to accomplish this goal.

Tell me the smallest possible discrete steps that

you would take to accomplish this goal. So that if I was explaining

this to someone that had no context on my business, my problem,

that even they could do it. Put that question into Chat

GBT, and then it will give you the discrete steps.

And then follow those I mean, audit those steps,

but follow those steps because then what you'll be able to do

is at every point you can check it.

Great. Is there an ICP? Yes or no? Okay, great. If yes, extract the ICP,

extract the audio customer profile, and print out a paragraph.

And then you can look at that paragraph and be like, that's wrong,

shoot. Let me go back and give it more like, let's break that down to

smaller steps, which is like and another way you can do is give it examples.

Here's an example. Here's an example. Here's an example. Now it's your turn. Now you

do it. And so every time, the more steps

that you have, every time along the way, you can

be clearer with it. And the possibility,

again, is not reducing time, right? This is something you would never do

with a human, right? The possibility is,

what are the gigantic possibilities if

you had a person to work near unlimited amounts of

time on a creative task? That's the possibility.

It's not saving. I mean, the low hanging fruit is saving time,

but the real opportunity is like,

what if you had unlimited dollars to spend on someone that's like,

okay, creative, and they had unlimited energy and they didn't get distracted?

That's the opportunity, I think. Do you have

any true automations going on for this yet? Because a lot of

it sounds like you're still kind of putting things in manually

to make sure you get really good outputs. But do you have plans to automate

or to have things just happen naturally? As you

have a process that's triggered in certain software, you got to start writing things.

Do you have any of that set up yet or is that in your plans?

I do. I think about this is I have campaigns that are going out right

now with AI, which is this like search campaign, for example.

So I have a handful of campaigns that I have used

that in very targeted ways where I can trust the output,

where I'm not saying asking it like truth questions.

And I think that the challenge here is that to automate

this, you need to make sure that a wrong result

won't kill you. And what I mean by that is that

the other day, sales reps like to use this for Icebreakers. And this

is an email. This is a really email someone sent. Hey, did you know that

Fiona also means white? That must be why you so pale.

Ha, just kidding. And so it's like, what? What are

you doing? That's a terrible idea. And so

that's the kind of and just the pain is

so high if it gets it wrong or it's like if

I start talking about case studies that marketer hire doesn't have,

or if it lies one times out of 100,

it's too many. And so most of my

automations are not discreet.

They're like campaigns, they're not push

things through. But I will get there. But I need to

think more about how to creatively constrain it so that if

it gets it wrong and it'll get it wrong, like, it's not going to get

all these titles right, for example. But if it gets it wrong,

it's like someone will see it and be like, well, that's not exactly right,

but pretty close. As opposed to like I

see. I asked it, I was playing around. I said write

a sentence as if my wife went

shopping at this company's Brand,

and I was just like casually encountering them. And it's like invented

children and stuff. It's like I was buying some diapers for my kid. And it's

like that's where the message is probably pretty

compelling. People respond to it. It seems like a human wrote it,

but it's like weaponization in the worst possible way

because then how far do you follow that lie down the path, right?

And so that's what you need to make sure that it's

not doing is that if it's lying or if

it's not getting it perfectly right, that it's a harmless mistake.

Last one for you here. And then we'll call it quits here for today.

But appreciate your time and would love to chat again for sure.

There's a lot to dive in here. This is just one quick little foray into

AI, but my last question would be around the knots of

AI, because you have talked about a lot of things that you're trying to build,

keeping it really positive with, like, this is cool, this is cool. Here's how

to fix certain things. What are the things that you just absolutely will not

trust AI with right now, with your business?

Most things. So I think that I don't trust

it to write a complete email. I don't

trust it to act on my behalf. Like,

I would never send messages as me that

are fully written by AI. I don't believe that it can write a

great email. Maybe the only way

that I think we could write a great email is if you basically trained it

on how to write five discrete sentences with like,

six to 15 steps per sentence, and then you took

another step to summarize that email. But if anything

involves requiring of truth,

I don't trust it. And so,

yeah, there's a lot of things I don't trust with AI,

but I do think that the problem is us.

We don't know enough about how to use it, about how best to deploy it.

And so there's a question problem first, which is how do you get

the right questions? And then there's a process problem. And the

best possible tool that I've found for the process problem is clay.com.

And you basically can have conversations with Chat

GBT on a per row basis, which means that for every row

of data, you can have four 5610 chats

with chat GPT to further refine, further refine,

further refine. And that has been really helpful

to hone the results to be exactly what I want.

Awesome. Well, this has been fantastic. There's so much more

to cover here, but any parting words or last little thoughts before

we hit the stop record button? Totally play

around with it. Use it. Use AI. It doesn't matter what you use it for.

Ask you questions about distances, about just the more time

that you spend with the tool, the more you

think, I should try this with chat GBT first, you're going

to get better and more creative ideas about how to start replacing

pieces of what you do today. And think about firing yourself.

If I had to fire me today and I replaced myself

today, what would new Jordan come in to do with

AI and how would he rebuild this from the ground up?

Amazing. Jordan, thanks for coming on.