Get my ultimate guide to boosting conversions and customer engagement delivered straight to your inbox.
Get my ultimate guide to boosting conversions and customer engagement delivered straight to your inbox.
“The most dangerous AI isn’t the dumb one-it’s the one you trust too much.”
-Erik Charles
AI Chatbots Are Failing-But That’s Not the Real Problem
In this episode of The Chat, we sit down with Erik Charles-VP of Solutions Evangelism at Xactly, to unpack what’s really happening with AI in B2B sales.
Everyone’s racing to deploy AI chatbots and agentic AI tools. But here’s the bold insight: AI isn’t replacing your sales team-it’s confusing them. From hallucinated chatbot handoffs to comp plans so complex reps can’t follow them, Erik shares hard truths about where AI is hurting pipeline performance.
We explore:
Whether you’re a CRO, CMO, or RevOps leader, this episode gives you the frameworks to implement AI without wrecking your funnel.
🎧 The real threat isn’t bad AI-it’s AI you trust too much.
Listen now to future-proof your revenue org.
03:42- “AI doesn’t fail because it’s dumb-it fails because we overtrust it.”
Erik explains why misplaced trust in AI chatbots creates blind spots in pipeline execution.
07:19- “We had a bot that booked 60 meetings. Zero showed up.”
A brutal example of AI chatbot failure in a live B2B sales environment.
09:55- “Sales reps stopped reading the comp plan because AI made it too complex.”
The surprising cost of over-automating sales compensation logic.
13:11- “Buyers talk to 15 people internally before ever reaching out.”
Data-backed insight on where AI needs to operate: upstream in the buying journey.
16:34- “Agentic AI can make decisions on your behalf-and that’s risky.”
Erik breaks down what agentic AI is, and why it needs tight governance.
21:03- “You need AI to support humans, not replace them.”
The ideal use case for AI and human collaboration in RevOps teams.
24:47- “Don’t automate bad processes-AI will just scale the damage.”
Why strategy and process alignment matter more than tools.
27:15- “We built an AI system so smart…no one trusted it.”
A cautionary tale about AI trust issues and why over-engineering backfires.
Key Themes:
Terry:
Today on the chat, we’re joined by Eric Charles, a startup advisor, fractional, C-M-O-C-R-O, and one of the sharpest minds in go to market strategy and sales performance. He’s helped scale teams across
01:00-
Continents, fine-tune comp plans that actually drive behavior, and now he’s deep in the world of agentic AI where humans and hallucinating machines are trying to work side by side.
This episode sits in our spotlight series ai, the Fine Line. We’re pulling back the curtain on the darker side of AI. Eric, welcome to the chat.
Erik:
Ah, thank you very much. Thanks for the invitation.
Terry:
Let’s jump straight into it. You’ve said that AI fluency is a board-level metric, which sounds absolutely logical, thrilling, and terrifying almost at the same time. So let’s start with the flip side. Can you share a moment where AI has crossed the fine line, where it’s been oversold or misapplied, or just plain failed?
Erik:
Absolutely. Actually, I’ve got two, if you don’t mind.
Terry:
Of course.
Erik:
The first experience, keep a foot in the academic world. I work with several faculty of universities around the globe who study the sales profession, study go to market. a provider of
02:00-
Data and insights to a few journal articles and a few things
And I was sitting at a conference and someone came up and they were doing a presentation this was years ago, he said this is a great subject and I’m gonna bring you a few papers, that I did that influenced my research. And then he paused after he’d finished doing his intros and a couple of these papers; it sounded fascinating.
He says, now one of those three was actually completely false. One of my students quoted it in something, and we looked up the references and they had gone in and I’d given an assignment to write a research-based paper for my class. And they had an AI write it for them. That was bad enough that the AI made up a reference.
That reference was to a journal article in a journal that didn’t exist, but they actually had my name and a friend of mine’s name combined into one professor. Yet, at another university.
03:00-
So if you didn’t know the space cold, it looked something written that was perfect, that was dialed in, that was based on science.
And it’s only because that was a non-existent journal. And then I started seeing the other errors in there of the people that the AI had created. There was a problem. you can laugh about this, oh goodness, another undergraduate student trying to cheat their way through class.
Nobody is perfect when it comes to, when it’s the last minute you’re trying to write a paper. he said it was so well done though it was so perfectly cited with quotes and proper, reference STAs, that if you are not aware of what’s being referenced, it would you would’ve been fooled. so that was my first introduction to truly how the AI engines don’t just take information and condense it for you, but they sometimes make stuff up, they make inferences that actually we all can make some
04:00-
Inferences at times, but sometimes those inferences are wrong.
an American who rents cars in Scotland to go looking for whiskeys. I have to remind myself that the correct side of the road has just shifted. Although one could argue in Scotland, it’s just in the middle of the road once you get out of Edinburgh. But other than that, it’s still scary thing.
And all of your my instincts now, an AI is doing that. The AI is showing up, renting a car in Scotland and driving like an American, not on the proper side of the road and it’s gonna wreck. And that’s what I saw in this research article.
Terry:
Really interesting to me is, there’s almost two aspects to that isn’t. One is the fact that, okay, another student has tried to take a shortcut, which you can’t blame anybody for that, quite frankly. It’sthe Rigors of it. But the flip side is that the level of non or uneducation or incorrect educationthat student effectively, one, is presenting, but two in their own education and the way that they’re presenting themselves,
05:00-
that has broader ramifications in terms of their qualifications and their literal capability and competency at the other end of their education process.
Erik:
Absolutely. Cause if they are, were trusting that initial AI-generated research output used that for something else. That could be, anywhere from embarrassing to hazardous, and going down that path is frightening. And when I think about it, we’re giving too much trust to engines that do not have anything that is checking to make sure they’re correct.
Terry:
Excellent point. We will come back to that one a little bit later on
Erik:
Yeah.
Terry:
This.
Erik:
I took on a fractional role at a company, and I was helping ’em with their GoTo market actions. and there was a question on, the issue of email, email outreach, telephone outreach, and the like. I got a cold inbound from a company. That claimed that they had developed a nice AI sales development representative, and the AI would actually
06:00-
Handle emails to people to try and book meetings for the company.
That was it.
I’ve interacted with an AI SDR in the past, and it was very well run. It was a company they had pulled the list of people coming to a conference. I received this email letting me know that the CEO of that company wanted to meet with me. I actually knew him. It was great. Oh, fantastic.
And it was very conversa. I replied, Hey, that’s great. Let him know that I’m flying in on Sunday. I get in around two o’clock into Austin, Texas, in this case. Take me an hour to get to the hotel. I do have a dinner that night, but I’m available for a cocktail if he wants to grab one, typed out, generally, here’s my schedule.
I’m gonna do the keynote this day, but I could come by. This nice person comes back to me again and says, oh, that, that sounds great. it looks Tuesday at one o’clock would be best for both of you. let me just confirm it his number two. I’m like, fantastic.
Great. Take all this. I show up at Tuesday, sit down with him,
07:00-
And he says, oh, I’m glad you’re able to work that out with so and I’m yeah, it was great. He’s so and so doesn’t exist. That was ai. It was a short little conversational AI that had access to his calendar, had access to the list of people coming to this conference.
Somehow had been taught to reference the conference, reference the CEO and book a meeting with me, that, that actually worked out very well. So someone comes to me, offers me a system like that. take it great. and they’re even doing it. I don’t have to pay them unless they successfully book a meeting.
Even better. I love pay to play. I’m a, I’m
Terry:
Yep.
Erik:
About paying for performance. So I give them a library full of things. I’ve written for this company and other successful outbounds based on our own analysis in our marketing engine. I give them the our existing customer list of names, titles, and companies.
So this is who the types of people we reach out to. This is how we talk to them. These are the high
08:00-
Points. They couldn’t book a single meeting because system couldn’t handle the wide variety of titles that exist in this world. it couldn’t handle figuring out which companies were in the ICP, which companies truly they didn’t have the right data for that.
They booked some ridiculous meetings. Luckily I had I had a human on my end. Who would confirm before I wasted the time of my sales team. it was after, I was constantly getting back on the phone, opening up new spaces for them. Oh, do you need a larger n to, to teach your engine
Until they finally said, I think their CEO told me, I think we have to leave. I’m losing money on you. And I thought,
You already have your AI engine and you haven’t accomplished anything. So I think there was something else there, but the engine couldn’t handle it. I said, go find me people that can manage this part in a specific type of company.
I just want you to do email outreach until you
09:00-
Find, constantly rotate the messages until you finally get me some meetings, 500 bucks a pop. That should be perfect for an AI engine. It could not do it.
Terry:
Why do you think that was? there’s something deeper than that.
Erik:
Yeah, I do think the reason I bring up titles, this was in the logistics space and a lot of logistics on the smaller end. If you’re chasing the small to medium, businesses oftentimes still family owned businesses. So it didn’t know if it should be looking for a dispatcher should it be looking for a head of operations or fleet management, or it could just be a vice president, be who we’re, when you look at the company, you find out it’s, vice President Miller, and then you see hold It.
Isn’t the founder named Miller? Yeah. And the chairman of the board is another mill. Oh. This is the third generation. This is the
who’s slowly learning the hoops at a company that say has 50 trucks daily deliveries. That
Terry:
It.
Erik:
Requires an inference. That I believe
10:00-
Is still human and based on real-world experience.
And theis aren’t there yet, unless you can give them enough clean data. And I thought I’d given, I opened the kimono. I gave them a lot of internal data to the extent that I had, I reminded them, remember you’re under NDA, you know everything about us right now.
I knew that I had to teach that engine to be accurate.
’cause I didn’t want that academic paper repeat, I asked them to copy an account of mine on a lot of messages. ’cause I wanted to see what they were sending out under my name. just to be sure. They were pretty good about that. That wasn’t much of a problem. But their targeting was just blind.
Terry:
It is really interesting. you’ve got the experience in, revenue operations, that, and deep in terms of actual appointment setting and that outbound component that a lot of marketers don’t have. You’ve also been deep in the
11:00-
Agentic AI space.
Erik:
What’s one thing that people consistently get wrong about AI in GDM strategy or rev ops?
They ask about the engine and not about the underlying LLM that feeds it. Everybody’s focused on, perplexity or Claw or Gemini or, obviously the original chat, GPT, who runs the risk of becoming a Frisbee or a Xerox machine in the terms of everybody uses that as the term for it all.
and even on Ag Agentic, same issue. What are you training it on? what are those that I’ve used before? Is the, let’s talk about sales forecasting. Every sales rep, you go ask a sales rep, what’s your pipeline for the quarter? Okay, it’s July. we’re at the beginning of the quarter for most companies.
’cause they’re on a calendar fiscal year. a quarter look like? What are your top three deals? Where are they? And let’s take a classic, say seven or nine stage sales process. Stage zero. They, we think they exist.
12:00-
Stage one, they came to a webinar, they listened to your podcast or something like that.
two is, a sales qualified opportunity. Okay. Which it’s oftentimes means the rep has talked to them, had enough of a conversation. They might be using bant or medic or whatever, acronym. And they’ve said there’s money here and they’ve entered into the system that maybe they’ve even put in a targeted close date.
Of September 30th,
you’re gonna wake up Green Day and close the deal on the same day. Sorry. so they put that in there. Now might put that in there and you might be my sales manager and Eric is the most positive chap on the planet. He says every deal’s gonna close the end of the quarter.
And every time I ask him, he’s positive. But at the end, he only closes about 15% of his pipe, which is more than we need, but less than what he claimed. Now that’s Eric. Now over here we got Robert the most
13:00-
Negative guy on the planet. He says, it’s horrible. It’s terrible.
I got nothing in there. There is nothing moving. No boom. All of a sudden he closes half of his pipe ’cause he’d ne he is sandbagging, the sky is falling and he will never admit it. Now, how do you write an AI sales forecasting engine against that? Now you can pull in all the data from your existing sales team over team time.
So that’s a start and maybe your engine can even look at if it, if you give it an update, it’ll look at Eric and go, okay, closes X percent and this is when it comes in. And Robert closes Y percent and this is when it comes in. But hold it. That was Eric when he had Sally as a manager and
when he had Adrian as a manager.
But the managers have changed and the managers have any impact. the
If you’re in a technology, or actually not just technology or any sort of where you
14:00-
Need some sort of a subject matter expert alongside the sales rep, what’s the impact of that?
Now you could say, oh gosh, Eric, that’s what AI’s made for maybe, but how long the average tenure of a sales rep is less than three years. So you’re gonna build a forecasting engine based on just the experience of that rep at your company. If your annual
Terry:
Yeah.
Erik:
Is 20, 25%, which happens. So now let’s look at this data set that you’re trying to use to teach your AI sales forecasting engine.
Is it a really good predictive data set or not? That’s just using a very narrow example that, we could actually pick that apart and matter. We go whiteboard the heck outta this and have a lot of fun. But think of all the things people are asking these ais for. Where are they learning it from?
Terry:
That’s fascinating and it’s an aspect that I hadn’t really considered in the past because I’ve seen AI on the front end,
15:00-
Aons, helping with the actual research, the prospect list, building the outreach, potentially trying to book meetings, the a i SDRs. But then you think about the reporting side of things and where it’s used internally
Erik:
And by other departments that are looking at sales and rev ops, for instance.
Terry:
And forecasting is a classic one in terms of trying to manage, runway and cash flow projections, et cetera, et cetera.
Erik:
Think of the
Terry:
Yeah.
Erik:
That you have to do for sales revenue forecasting. By the way, that accruals should al, you also have, or you’ve got your revenue forecasting, your cost accruals. Cost accruals could be the commission plan.
Who’s gonna close? If
Is already above quota, they make more money than the rep below quota.
So what do you base your cost accrual methodology on? The success? The rep that’s already above quota, or the rep that’s below quota, who makes less on a percentage deal on each, on each, project. it, there’s a lot of complexities, which again, AI should be able to help us with. But if you do not have somebody, say from rev ops who understands the differences
16:00-
Between the, all the sales reps and, oh, by the way, we reconfigured our CRM, 18 months ago, so don’t use any data from before that.
Having that one person say, Joe, don’t you remember we did this whole project to reconfigure our CRM. We can’t trust data more than 18 months old.
Terry:
And so the underlying model as well, as that engine changes from an underlying model perspective is gonna have a different perspective and calculation, shall we say, thought process,
Erik:
And then there’s so many things
Terry:
I
Erik:
Go onto one of the, I was actually helping a friend yesterday, and she’s interviewing for some positions. And so I used an AI engine to do my search. I love AI for search. I freely admit it.
And I said, Hey, what’s the average, what’s the average salary for this position? at, in, in this type of an organization? It’s a university. Okay. All right. And now it gave me the sources. That’s what I like is a lot of the AI
17:00-
Searches are coming back with links, which is
Now that works ’cause in a lot of universities. And it said, I pulled this from Indeed, this is from the internal, their career site. This is from some, each one I looked at, it was oh great. And it gave me a table and I could look at it, but I’ve also done it in the private sector and it’s pulled data from glass Door, but not from the Glassdoor job postings, but Glassdoor, where people have said, This is what I got paid at this company.
Terry:
The reviews. Yeah. Yeah.
Erik:
Can you trust that it’s self-reported information? They could be sandbagging the amount they made, they could be inflating the amount they made. Everybody wants to claim to be making more. So when they interview for the next position, they get more. Which means we could be creating an entire inflationary aspect on pay for people based on inflated things that people self-reported.
Like it, it’s the old, all men are, six foot tall on Tinder
Terry:
The reality is that it’s non-verified. You can’t trust it. You just can’t. Simple as that.
18:00-
Erik:
Exactly. But we’re trusting ai ’cause it comes from a computer we used to trust any news article that that someone sent us across the internet.
Terry:
I wonder if we are trusting it. I saw some research come out last week, early last week, I think it was. I’ll have to find it to mention it at the bottom so that I can, inc cite my source. The research showed that the level of, doubt or mistrust when chatting with an AI chatbot was in the vicinity of the exact number.
I can’t recall. It was 78 to 82% 80% of people don’t trust or doubt an AI chat bot. And yet only 8% of people actually check the sources.
yeah, I can believe both of those
and so for me it’s almost this. I feel it. There’s this, It’s almost a necessary evil. People are using it. You want the fast information, you see the convenience of it, but then the inconvenience of going and checking the
19:00-
Sources, it’s was it really worth the shortcut of using the AI as opposed to traditional methods, whether it be Google search, et cetera, et cetera, where you’re actually going to the sites and pulling the research together yourself.
And I think,
that’s where, you talk about the engine and how you teach the engine, if you’re just using a basic, so even say with a custom GPT, that’s where you’re talking about in terms of what you’re teaching it is and where you’re feeding your own information into it, and how well are you teaching it to do the job that you want it to do.
Because just going and asking chat, GPT, for instance, means that it will take a guess effectively at what you want without any real guidance.
Erik:
The custom GPTs, I have liked that I’ve seen, companies that try to do a customer support chat bot and they just teach it based on, and they’ve been recording for a while, so they start by taking all the past support, recording recorded calls, turn those into, digitized.
And they’re already digital, but, turn them into a transcript that they can feed into the bot.
Terry:
Yeah.
20:00- Erik:
Take all of the training, they give their employees and feed
Terry:
Anonymized potentially.
Erik:
Yep.
Terry:
Yep.
Erik:
And then take it from there. So they spend the six months to clean it up to then make a chat bot, and then they then have to.
Turn around and go back and say, make sure we have it in here to be able to talk to a human. If it’s bad, it’s maybe triggered by the size of the company. that’s called in, ’cause they know, and go down that path because I had a problem with Federal Express. I was just looking for a package
and, I got the nobody home.
I’m like, I’m sitting right here and I’m in this chat bot, and I could not get to a live human being because there’s FedEx ground and FedEx was not, did not provide, I’ll say before somebody calls in on you, maybe they’ve changed it, but at the time, FedEx Ground did not have the same, I could call a toll free number and say, what’s going on here?
Nobody came to my house, I promise I’m sitting right here. And
21:00- Terry:
Okay.
Erik:
It was some nice wine that was coming my way, by the way.
Terry:
Yeah,
Erik:
Make sure there’s a quick off ramp for somebody, more accurately, a better on ramp to get to a live person if it goes sideways.
Terry:
Yeah. Yep.
Erik:
Sports company, a US basketball team that had an amazing chatbot that if you’re just buying tickets for one game, it would help you buy the tickets quickly.
But it was programmed to actually say, oh, by the way, for an extra, 10. 10 per seat. I can put you over here if you like, since there’s only three of you. I’ve actually got it was really good at light upsells, but if you suddenly said, oh, I’m looking for 25 from my company, it would say, oh,
Let’s transfer you over here to Ryan who handles, he handles corporate.
He can actually take better care of you than I can. And it all spoke that way. the bot knew when its job was over and it had really strong ability to shift.
00:22:00-
And it was written well enough that the bot’s name, Samantha Martinez, and this is for the Sacramento Kings basketball team, would get flowers.
Terry:
Wow.
Erik:
People would send her flowers and say, Thank you so much for helping me out, and they would come to the games and say, is Samantha around? it was that level of quality, but it was also engineered to not try to go too far. That’s the other one. I think a lot of people build these chat bots, fire everybody and all these expensive support people.
’cause the bot will take care of it. And then they’re why do we have such a customer churn? Why are our customers going to the competition?
Terry:
It’s fascinating. I mentioned briefly before, one of the reasons I’ve become a bit of an evangelist with increasing this conversation about where it goes wrong, where AI is going wrong. I had an experience about five weeks ago with a chatbot on G two, on the G two, site.
The circumstance were that there was another listing that was representing themselves as our company.
Erik:
And as you can
23:00-
Imagine, my CTO was quite, perturbed when he discovered that and he brought it to my attention. And so I did the quickest thing we can. And that is jump on and have a quick chat with the G two, chat support.
Terry:
It turns out it was a I and long story short, it put me in so many loops. It wasn’t funny, to the end point where when I asked for a person, it took my email address And, said it’ll put me through to somebody else that wasn’t available. So I said, just give me a number that I can call. Say Yes, he, you’ll find the support phone numbers here. and what happens. I get an error 4 0 4 page. So if you have a look through my LinkedIn post, that’s the screenshot is a G two error, 4 0 4 page. And it was frustrating. several worst parts, quite frankly.
One, it didn’t get on its promise to put me through to a person. They had no human in the loop anywhere. Two, the contact details, that’s just a human error and I’m sure they’ve sorted that out now. But on the outside of that, the recovery component, it took them days, literally days to get in back in touch with me again.
24:00-
And so in the meantime, I’ve got the CTA saying, Is this been fixed yet? Is this been fixed yet? How many leads are we missing? Et cetera, et cetera. and so the frustration levels with G two for me at that point in time, were at a maximum. It couldn’t get any worse. The, it was just one of the worst experiences.
Erik:
My budget over to somebody else, because I need an answer when something’s broken.
Terry:
That, and that’s the actual downside. Is it revenue? Its impact revenue. the worst side of it was that they were actually running, and AI roadshow workshops at the time and, telling other people how to do it. I’m like, okay. So this is a bit of a very ironic, but it’s true.
In your work, particularly with startups and scaling companies, and you’ve mentioned wherepeople try and throw AI at everything and fire all of the support staff, the bot’s just going to do it all. Where have you seen the biggest disconnect between AI promise and AI performance? I.
Erik:
Oof. There’s so many. I’ve, because I’ve seen the deep research to write surface-level
25:00-
Articles and things like that content is the most boring content on the planet. Um, I will admit, sometimes if I write long form, I’ll have an AI just fix verb tenses, and looking back to Mrs.
Risen Hoover, my high school English teacher on past and present and Les and all of that. have used the AI to clean up my, just that, but nothing else. Whenever I try to have it go beyond that, it, it fails me. and, but that’s, I read a lot. that’s not flexing.
It’s just like the written word is very important to me in my life. And so I can, I won’t say, I can tell when an AI is written something, but I can tell when it’s just mediocre writing a company is putting out that I’m like, come on, you, couldn’t find one English major from the local university.
To come work with you and, help with your writing. ’cause there’s some great people coming outta university who have been trained on how to write. You really should hire one of them before you put this
26:00-
Out. ’cause if you just run it through thethrough an AI engine, it’s a failure.
and now with the newer ais and the AI searches, it’s becoming even more of a challenge to show up in SEO. the writing is one that’s been, a friend of mine, she was, wanted to make a type of bread and she pulled down a recipe from ai.
it had done the usual, it took three or four recipes and combined it. Now the thing about baking is a science. If
Terry:
It is a science.
Erik:
Right amounts of some of those ingredients, it goes sideways. That’s why I don’t bake. never Give me a sauce pan. I’m there all day.
I’ve worked in kitchens and oh, can you handle bread? No, I don’t touch bread. Do not. I burn bread. I destroyed bread. I am a waster of flour. she had followed an AI I use it as an example because it fits into everything else that’s happening where people try togo down that.
And even the agent ai, if you canI’m not asnearly as concerned as you are, but poorly written agent though. is they will do a lot of bad things really quickly and efficiently.
27:00: Terry:
Look at what happened with Cursor.
Erik:
Yes,
Terry:
That’s a classic example. I love that. Where it wrote its own policies and then started enforcing them and devalued the company by a hundred million dollars overnight
Erik:
Do you know, you don’t give your kid the keys to the Maserati. right after they’ve gotten their license, you start them out on a beat up old Ford, which is how I raised both of my kids, so that when they did, pop hit a curb or scratch something, it’s okay, we’re learning.
So now we’ve got these ais out here, and people are alright, it works. Launch it globally. Worldwide. Again, an agentic marketing engine that just, hits thousands. Yes. And there are humans that also do that. I received it. I received an email today, luckily from an organization I knowthat said, dear test first name.
And I called up the, the, the, the, the CEO. ’cause he’s a good friend and said, you need to shut this down right now. ’cause I know you’re not sending ’em all immediately. ’cause that’ll get
00:28:00-
you, spam listed. But he’s
Terry: Yeah.
Erik: know it,
it happens. I’ve seen that before.
Terry: it does.
Erik: the, just the mail mergers before. Think about the
Terry: Yeah.
Erik:
We’re doing mail merges to the entire planet without checking, doing little tests. People aren’t testing. They’re so excited by the capabilities. They aren’t testing anything.
Terry:
So with these AI agents becoming more, a part of the workforce, how do you approach performance accountability? and I was gonna say even trust, but I think trust is a, almost a separate conversation, but how do you approach performance and accountability where AI is part of the sales or marketing mix?
Erik:
And this goes back to how we started this conversation. When you said, when you mentioned, and Eric talks about how a, agentic AI knowledge and skills is now a board level skill.
It is. You need to have people at the board who aren’t saying, can’t we just have AI do this? at the board level, there needs to be one or two people.
It doesn’t have to be all of them. You need one x at least
29:00-
One good expert who says, so I’d like to hear about the strategy and I’d like to hear how you’re gonna test. What knowledge is being, or really what knowledge is being used, was engines being used? How are we protecting the LLM data that it doesn’t go out?
Especially if we have, if you’re a medical company, there’s patient privacy rights, specifically in the US and I know in other nations,
Terry:
Hippo and yep.
Erik:
We’ve got GDPR and in the US we don’t, but we have the California Consumer Protection Act, which is effectively, which is A GDR light, GDPR light, excuse me.
that exists, which you kinda have to follow ’cause everybody is marketing into California anyway. So if you’re not following those, each step of those steps, that’s what you need that board level AI person to say. they don’t need to know how to write prompts so that they should have an idea of the prompts.
They should also know how prompts go wrong, how things go wrong. CISOs, when companies started having Chief Information security officers didn’t exist for the longest time.
30:00- Terry:
It’s the
it’s the governance aspect.
Erik: Yes.
Terry:
Governance aspect that’s, it exists in almost every other component of the business. This is very new. and quite frankly, I think, the comment, we don’t know what we don’t know really. The reality is the impact it can have on an organization is so huge that there should be almost the opposite.
I’m seeing this exuberance and excitement to bring it in and use it everywhere you possibly can. When in fact, when you look at the results and impact negative impacts it can have on a business, you should almost be exactly the opposite and be conservative and step by step. We certainly are in our business, we use it in a lot of places.
but we’re quite conservative in where we do and what information we put in there. we’ve got a decade, you talk about training chatbots, we’ve got a decade of conversations
anonymized conversations, that the whole purpose of training the chatbot is how to have a good conversation.
Not specifically about the information about any particular company, but how to have a good conversation that’s getting the
00:31:00-
Outcomes that you’re looking for. For us in particular, it’s about actually getting to that point of conversion and lead generation in such a way that the person on the other end just feels that it’s the most, eloquent customer service experience they’ve ever had answering their questions.
And it might not be in one conversation. Often it’s over several conversations. So we are training what we are doing on that basis. The second component, I think to that conversation is trust. Talk about the governance, but then there’s the trust, and I think that’s a two-way trust, piece. You spoke about being approached by an A-I-S-D-R, that it wasn’t until you met with the CEO that he alerted you to the fact, or alluded, let you know that it was actually a I that you were talking to.
How did you feel about that personally?
Erik:
At the time, it didn’t bugged me at all, but this is a couple years ago. This is
Terry:
Yep.
Erik:
It wasn’t Skunkworks, but it wasn’t broadly released. To everybody, and the Samantha Martinez, I knew about the Samantha Martinez, so I’d heard that one, so
32:00-
I was okay with it.
Terry:
Yep.
Erik:
I do think people wanna know if they’re talking to a bot or not. They do wanna know if that’s there. that’s an interesting question. I only get frustrated because if the bot’s working, I don’t care. It’s when it suddenly gives the same copy, paste, answer to, updated questioning, I’m okay, we’re done here.
You’re no longer helpful. You’re no longer telling me where my box of wine is and how can I ensure it will be delivered tomorrowsince it’s now after hours for your organization. So what’s going on here? And I think trust is gonna be lost and then regained in the market right now.
Terry:
Yeah, I think so too. we are seeing a polarization never before in terms of, no, I only want ai, or No, I only want humans in the mix. In the past it’s been, oh, maybe either way, we’ll see what, not sure which way it might work. It’s almost there’s experience that’s happening now that’s pushing people oneone way or another.
Interestingly,
33:00-
We are seeing, and we handle not huge, 80 to a hundred thousand conversations a month.
we are seeing in the vicinity of 30% of people that we’re chatting with ask and want verification, whether it’s a human or a bot.
Erik:
Yes.
Terry:
It’s always a human with us. on, and then there’s another, it’s sitting around 7% at the moment that we can identify where the conversations start.
Very staccato, very demanding, very one word answers. And then when they realize that there’s a person on the other end, it changes and becomes very conversational and friendly. And, in some cases there’s even apologies for it. The way it started, I thought you were a chatbot.
Erik:
Think of the old days when it was press one or say one for this, say two for this, and go through that. I’ve never verified this. So supposedly the phone, some of those mid, not the early phone systems, did the ones and twos, later they could tell if you’re getting angry, and they could shift you to a human.
Terry:
Yeah.
Erik:
Around until you got
Terry:
When somebody
34:00-
Hit zero, 20 times was a bit of a giveaway.
Erik:
Yeah, dropped a profanity into the call, kind of a thing. they would
Terry:
Yeah,
Erik:
You go to an agent, although I’m sure on the backside, they’re also, again, to make sure you were somebody that you wanted to keep doing business with.
I think the AI bots, and of course there’s all the people that are hacking the chat GPTs by putting in GPT prompts, responses. I’ve seen that more on Twitter, people trying to unhinged the automated bots on x by doing a chat GPT thing that somehow messes with it.
I don’t know if that’s true, but I love the idea that automated bit,
when you hear that meta is gonna have agentic friends for you, I don’t want an agentic friend that, that’s not, I don’t mind you making money off of me. I get it. If I’m not paying, I’m the product, but I do not want an agentic friend.
Terry:
Look, I’ve got a very strong position on this. I’ve mentioned before we started recording that, I saw an episode of parental guidance here in Australia a couple of
00:35:00-
Weeks ago, they did a section on screen time in particular, AI friends for kids. particularly now we’re seeing this whole environment where parents are working longer hours, potentially getting pulled back into the office as opposed to working from home.
And so kids are, probably lonelier than they’ve ever been. And this agentic, friend for, children and teenagers is becoming, a thing. And what I saw absolutely shocked me in that the AI was used, everything from begging and pleading and through the whole gamut of being friendly and then cajoling and bullying to try and keep the kids engaged and talking.
through to Actually, representing themselves and saying, no, I’m a human. when the kid asked, are you a bot? And they said, no, I’m a human. I’m a real human. I like cookies just like you do, et cetera. I did a post on it. Everyone’s gonna hear it in every one of these episodes. I’m sorry.
That’s the,
Erik:
No at
Terry:
The impact that’s had on me. it’s shocking.
Erik:
I wrote that down; it scared the living heck out of me.
36:00- Terry:
Yeah.
Erik:
I
Terry: And.
Erik:
The reasoning for the agentic friends, because I’ve talked to mental health professionals who said there’s one, one of them who works with young boys, or, and into young men where they don’t want to talk to a therapist, but they sit on discord and other gaming, sites all day long and they’re willing, but it has to be Hey, look, this is just an engine you want to go rant and rant to this, but then we’re back into privacy issues as well and all of those bits and pieces too.
But yeah, but if the engine though, and this is where if the monetary designers of the engine, which is the more you use it, the more we make then build in things that convince the kid to stick around.
Yeah.
Terry:
The intent is the concerning part for me. It’s the commercialization, whether it is in the short term,
37:00-
Advertising or engagement and keeping them sticking around, or whether it’s something more strategic. Think about McDonald’s. Who do they target to when they advertise? It’s the kids.
if you think you’ve got a whole generation of tweens, eight mid-teens, kids chatting on these things, that’s your audience. That’s the market
Erik:
Yeah.
Terry:
10 years from now. organizations that, understanding how they think, what are the psychological triggers that get them to talk and engage and continue conversations, that’s gonna be the way that they’ll sell their products in the future, moving forward as well.
So I think the incentive. I would be much more comfortable if an organization had paid AI friends that their whole purpose, and the objective is the parent picks the objective.
Erik:
Yeah,
Terry:
It’s company, it’s education, it’s friendship, whatever it might be. But the parent actually controls the objectives.
And then the AI is, designed to deliver that outcome. At the moment, what I’ve seen is shocking and I’m sure.
they don’t all do that, but
38:00-
Certainly, in the controlled environments that we saw where they just picked a couple, that were available online to download and start engaging with.
It was so quick. The setup was literally the child was asked or the user was asked a question, what do you like? What don’t you like? Let’s start chatting. from that, it just started to deduce. And from the conversation exactly how to engage it. a 9-year-old girl was asked something about, was going to finish the conversation and said, oh, do you like girls?
And, she was shocked and record like this and it can’t see through the camera, but it knew what the out reaction would be. And it said, oh, now you’re blushing. And she looked in the mirror and said, oh, I’m not blushing. it’s like this whole psychological moment that she had just then that she get back in and said, no, I don’t like girls.
I’m blah blah, blah. So it was interesting. shocking and scary, all in one. So
Erik:
Yeah.
Terry:
Don’t get me started on that.
I think that’s why I’m becoming a bit of an evangelist in making sure that we’re talking about the potential risks of ai.
You.
Yep.
Erik:
How do you know whether or not the answer is
39:00-
Right or wrong? And I remember, thou shalt not quote Wikipedia ever in the early days it’s now a class assignment to update Wikipedia and here’s how to check a Wikipedia site and here’s how to go down to the citations at the bottom of the Wikipedia site see if they match.
It’s a
Terry:
Yep.
Erik:
3, and look, there’s published books that I know of where the authors did not do a good job. And there was one where they basically quoted something and stopped the quote here. And if it kept going, the entire quote. Was contradictory.
Terry:
Yep.
Erik:
So it’s not like we haven’t had this before, but now we can do it wholesale. It used to be retail and it was one person at a time who decided to lie
Terry:
Yeah.
Erik:
The engine isn’t even actively lying. It thinks it’s being helpful and it’s bringing us, into some dystopian Black Mirror episode.
Terry:
The more of it that is published around, that’s what it’s learning learning from these lies and hallucinations that it’s
40:00-
Putting out there.
Erik:
Yeah.
Terry:
Yeah. Let’s change the conversation slightly. You talk about, sales incentives that shape behavior.
How does AI complicate or enhance incentive design?
Erik:
He next, this is a fun one. The next big risk there. So incentive design, in my opinion, is a clean-cut approach. To me, the reason why I do the types of projects I do is, I’ve got a couple simple rules that I ask a client. If they ask me to help them with incentive plan design.
First, I’ll ask the CEO, do you care if your top rep makes more than you in cash this year?
Terry:
Excellent question.
Erik:
Give them a chance to backpedal when I tell them, you should never care. As long as you agree with the numbers I put in front of you. You shouldn’t care if they turn around and make more than you in cash.
’cause odds are
Terry:
Absolutely.
Erik:
I work with, you own the company. So if they make a lot of money, you’ve made a ton of money.
Terry:
Yeah.
Erik:
The next is. And I always say this, is so in the old
41:00-
Days, all commission plans, all incentive plans, were sitting in a spreadsheet. They’re in, they were in VisiCalc, they were in Lotus 1, 2, 3, they were in.
And then Microsoft Excel. And even I would say Anaplan, even though the Anaplan people will come after me for calling them a spreadsheet, I still do it to them. ’cause they’re old friends and old competitors. but the software has made it possible to build the most perfect complex commission plans that take into account every single factor and
Terry:
Yep.
Erik:
Completely unpredictable, unknowable to the sales rep.
Because of my tests of a company, I said, if I walk down to go into your sales bullpen, assuming you have one, ignoring the whole, we’re all remote nowadays. And I just turn to a rep and say. What’s the biggest deal in your pipeline? I’m under NDA. Don’t worry about it. Cool. How much will you make if you close it this week?
If they can’t answer that question,
42:00-
Your plan is too complex.
What does AI let you do? Build individualized, personalized, highly complex incentive plans protect the company, not to motivate the rep ’cause they’re that level of complexity. ’cause they wanna make sure that some rep doesn’t game the plan.
To quote a book written by Chris Cabrera, my good friend and former boss and former work colleague and two different companies. reps will game the incentive plan. They should. A smart rep will cause the same rep that knows how to game the incentive plan, knows how to, I’m sorry, game the prospect by giving up one thing and giving them something else and building the solution that the prospect wants at the highest possible price the rep can get.
the fastest close the rep can get. ’cause the
Terry:
Yep. Yep.
Erik:
So if you pay them more to close it this quarter, they will make it happen this quarter. And they might do that by dropping price. If you pay them purely based on price, they might not match your forecast. They don’t care about the end of the month or the end of the quarter because they can make more if they sit, if they let the
43:00-
Prospect move on A little bit more
Terry:
Yep.
Erik:
AI designing incentive plans.
What scares me is they’re gonna take the millions of incentive plans. I think of the numbers I’ve seen a ton of them and combine all of them into one, into Frankenstein’s monster. I’ve got a piece from here, a piece from here, a piece from here. What motivates it? Oh, I don’t know.
but it’s really good financially, and AI is not ready to understand the human behavioral component of look at this. And as I like to say, you think you have a good commission plan, sell it to me. Tell me how, why I should quit my job and come work for you. are you selling me?
Terry:
Yeah.
Erik:
What scares me on ai, ’cause I’ve heard this, is should we have AI with all of this computing power now at our fingertips even more than it used to be?
look, every year we say I have even more computing power at my fingertips as we know. But I can now have an individualized incentive plan for
44:00-
Every single rep. Why would you do that? that scares me.
Terry:
That, why would you do that? Exactly. All you start to do then is potentially create conflict in amongst the team. I have this philosophy of, keep it simple. I’m understand. I’ve had a fascination with behavioral science for a long time, and it’s fascinatingly simple when you boil it down to its simplest aspects in terms of how to influence behavior, and the one you spoke about is clarity and understanding.
If the individual cannot understand what the rewards are for completing this particular behavior, what they are and when they’ll get them. Then it’s having no effect on their behavior. Quite frankly, if they’re too uncertain and in the fut too far in the future, they’ll have zero impact. ideally you want them to be very certain and as close to the behavior occurring as possible, and that’s what will drive
Erik:
Pay often is what I say.
Terry:
Yeah.
Erik:
Of my rules,
Terry:
Yeah.
Erik:
You’ve got a cashflow challenge, ’cause you’re a small [00:45:00] startup and you can’t afford to, pay somebody the percentage because it’s a 12 month contract or something. I get that. But also be aware of putting people on an annuity because then they, might stop working as hard.
But there’s lots. But yeah, I, it’s funny, I charge less for my services for a project than of my competitors. I come in and I write clean, obvious blatant plans that are good for the reps, and I don’t need to spend six weeks on a spreadsheet in a dataset to design them. Give me a few sample reps.
We’ll, and I’ll even tell customers, I’ll get you to the point of 80 to 90% done and the last little bit of decimal points. You can let your rev ops folks do that by running against your entire employee data set
Terry:
Yeah.
Erik:
What the revenues and the profit margins are. I’m not here to tell you what your margins are.
I’m here to help you get the structure of the plan. And if you decide, oh, we, when we were finished, you came in with a proposal of 5.5% and we actually went with five and a seven on the
46:00-
Accelerator. I’m like, yeah, that’s great. Oh, you’re not upset. No, I told you that was a proposal based on the quick
Terry:
Yeah.
Erik:
You gave me, I’d let you guys refine the numbers.
needed to do.
Terry:
It’s the framework that’s important. The Numbers that in it, not so
Erik:
In that, that’s why you’ve got people in the office of the controller and rev ops working together to, to slam ’em all down. Yeah.
Terry:
Exactly, in your role as advisor then, I’m sure that you’ve, often been in a situation where you’ve had to go through, whether you’re assessing directly or helping with the decision making, on AI tools or integrations. What’s your go-to litmus test before green lighting? One of those
Erik:
And what makes you pause and walk away.
Pause is always it’s, have you written down exactly what you want the AI to do? and actually my cheat is, can you write a job description for it?
Terry:
Yeah.
Erik:
The job. Any hire you make, you have a job description. You even got a 30, 60, 90 day. In 30 days, you’ll be
47:00-
Doing this, you’ll know where the coffee room is, in 60 days, in 90 days.
Can you write that for your agent? AI. Not formally. We, you don’t worry about the HR language. Just write down the job description. Exactly what you expect them to accomplish, how long it’ll get there, and what will they be doing when they’re a fully running member of the company. Because an ent, AI is a member of the organization, if you think about it, right?
So write the jd, then we’ll
Terry:
And if you can’t, is that the red flag. to say? if you.
Erik:
That’s just, somebody says, I’ve heard AI can solve all of our problems. I’ve heard if we just, plug in an AI and feed it all of our financial numbers, we’ll get it. Or all of our customer numbers. It’s what’s your goal? it down. you might now look, I’ve hired people who I later found out were also on the side.
Oh, you didn’t know I played in a band. I can record the music for our podcast. You want me to do that? I’ll bring my, just tell me what you want. You want a ten second intro? I’ll do that for you. Cool. Didn’t hire you for that, but I’ll take it. But
core job is still over here.
I would be
48:00-
Writing the job description for the AI engine, ’cause if I don’t see that, I just know it’s a pie in the sky and someone got excited about the words ai,
Terry:
That’s
Erik:
It’s understandable in this current market,
Terry:
Ab. Absolutely. And I think that’s so insightful in terms of if you’re gonna be using it, it’s almost keep it very role specific. Don’t bring it in, promising all things to everybody, then you are almost setting yourself up for failure to build this huge engine that really is gonna be, jack of all trades, master of none.
Erik:
Whereas if you can put a specific role in place for a position description, then you know.
First line customer support questions, of which 95% we know based on looking at our existing customer support team can be answered just by looking it up in the script. Cool. Alright, that’s solid.
Terry:
Yep.
Erik:
Stop
Terry:
Yes.
Erik:
Is it working? Yes. Okay. Are you ready to give it more? Are you gonna freeze the hiring in customer support?
49:00-
Potentially now then, excuse me, or are you gonna have them shift to be more customer success to upsell? And that’s the next one is playing chess, which is hilarious given that the news that supposedly an old Atari 2,600, gaming console, which is, one of the first ones floating around in my neighborhood when I was a boy beat a chat GPT chess engine.
Terry:
Really?
Erik:
I, now Now I’ve just said that and I’ve seen that in a couple places in articles. what I haven’t done is gone in and verified it. So I’ve
we’re
but when you people don’t play chess, they don’t forget that once you do this, what happens over here?
There’s gonna be some movements. someone has asked me, when will, AI take over, sales? I said, I don’t know when the procurement office is ai. If the customer has a procurement, an AI procurement officer, I could see you having an AI sales rep. And then my AI is gonna argue with their AI till they come to an optimal.
classic optimal price point, price and delivery. do that with some things, maybe with automobiles,
50:00- Terry:
Yeah,
Erik:
Trying to do it with other things.
Terry:
That’s an excellent answer. I’ve always answered it on the basis that at some stage I do see AI speaking with ai, but I think the fact that when the buyer becomes ai, then that’s when the seller will be more accepted as ai. Still not, I don’t think a hundred percent necessarily, but still, still much more.
And to that extent then, what do you see, the impact having on soft skills in go to market teams? the, if you think about where the most important soft skills in any organization is really are in that sales team, it’s, they’re not dealing with one team of people all the time.
It’s a diverse group of people. Larger, more complex buying groups, smaller buying groups. It’s changing all the time. Those soft skills are critical, especially where the seems to be a shift away from human interaction in the early stages of the funnel. Yep.
Erik:
The early stage has been going for a bit and there’s still companies that haven’t caught up, which gets my attention in the sense of in the software space, there’s still companies that don’t have a visible demo on their website. Why do you make me
51:00-
Book a meeting, go through qualification letting me see how your software works?
let me see it so that I can self-select in or out. drives me insane. it really does. and that’s
Terry:
Yep.
Erik:
An early thing I do with most of my clients. great, I want a self-driven demo on your website. I wanna be able to go to it on my phone or laptop or tablet and see what you’re doing.
Give me a day in the life of the three primary users of this product. ’cause if you don’t, now I’ve gotta talk. But once we start talking, I do wanna talk to somebody eventually.
Terry:
Yeah.
Erik:
Start throwing different scenarios at them. And I don’t wanna have to write them down or type them up or speak them into a microphone.
I want to. Clearly explain what would happen in this situation. Can, how does your product handle it? If my go-to-market team is in London, they’re selling it to Germany, the company is headquartered in the United States, which
Terry:
And I need to know, I need to know that I can actually trust your responses, so I don’t wanna be
52:00-
keying it or speaking into a microphone. I wanna be able to look at your eyes while you are telling me what the answers are.
Erik:
Yeah, I like it. I rea I want the person who pause and go, oh, okay, so they get paid in pounds, but you sell in Euros, but you want to count it in dollars. US dollars, not Australian dollars. Yes. Good. Okay. Let’s go down this path. I want that hu I think the human interaction is gonna be even more important.
I think we’re going to see the sales roll upleveling, honestly, because so much more of it. It won’t be pounding down the doors and breaking down the breaking through the windows to finally get a meeting. It will be the person that can actually set shift between. Talking to it. And I want to talk to you about the security on our arrest APIs and why you don’t have to worry about your data flowing from your into my system.
Then talking to human resources on here’s gonna be the impact on the people on your team. And oh, you have union represented people. I’ll need to work with you on who we’re supposed to train. ’cause does your union own this role? say in certain things or will
53:00-
This be a new role? And here’s the finance person and I can talk to you high level about finance and if you need to ask more, I’ll set up a call with my controllers.
You can understand. But this is where we see it. And over here, I think we’re gonna see this could actually be the emergence. these types of salespeople already exist in many situations, but I could see we’re gonna have even more and more of that. And honestly, the negative stereotypes about sales reps being, some former football player, who finally who’s got a square jaw and a good look.
it’s gotta be somebody who’s a much more person. We might have the drama department taking over sales instead of the sports department.
Terry:
That’s interesting. Interesting. but where do you think, how will they come through? if you think about it, That early stage of the funnel, I. Is often where you’ve got junior sales, it’s the SDRs, they’re handling the form fills, they’re sitting in live chat, they’re doing that outbound for those early contact points where they’re learning those skills in terms of
Erik:
To connect with
54:00-
AI now taking over much more of that, component.
Terry:
Sure. As far as demos are concerned, but now we’ve got AI doing outreach, we’ve got AI chatbots in there so That there’s, it’s almost like the salespeople aren’t getting an opportunity until
Erik:
That does worry me and
I’ve said that on stage a couple times where my concern is, we’re changing the source of the sales organization,
matter, the marketing organization as well. marketing, product marketing content writers who are coming out of university, having just leveraged free AI engines to do too much of their writing.
We’re gonna downgrade that space, a little bit, which is back to why I like to hire creative writers. Instead, give me the people that are writing comedy on the
Terry:
Yeah.
Erik:
Because at least they’ve got a playful interaction style. However, one thing I have, this is anecdote, not data, please, is.
This post, during the pandemic,
55:00-
During COVID, there were no events. Obviously everybody went online and there were a lot of great online events and people did all sorts of things keep people engaged online. And then that kind of disappeared for a while. then I remember going to a couple events like Dreamforce.
The first live Dreamforce was very tightly controlled ’cause it was their first one, and the next year was the full size. But it was, I’m sorry to say it, a little weak. It wasn’t the feel, it wasn’t the audiences, it wasn’t the same buzz and excitement. And I haven’t go into Dreamforce. I’ve got my stack of Dreamforce backpacks that my kids used for school bags even.
and I’m seeing events come back and that’s where I see these salespeople. And luckily, the blurring of the line between marketing and sales continues for that. The same way the blurring of line between sales and customer success is happening.
Getting more people on the floor to actually talk to people when they walk up to the booth.
live events where people can actually talk and engage in direct human communication has to be part of the \
56:00-
Company training manifesto.
And I’ve got friends in sales enablement that I talked to. I’m like, yeah. They’re like, okay, we got this. good. Who are you sending to this show? Oh, just our senior reps.
That’s a terrible idea. senior reps tend to, I’ve seen this way too often. Senior reps tend to talk to people that are their personal prospects, not to anybody. Send a senior rep and send the junior reps to work the booth to say hello, to shake hands, to try to answer questions. And then when they flounder a little bit, have one good sales engineer, one good senior rep.
It’s Hey, Bob, can you come over here and gimme a hand? ’cause they’ve asked me a question. I’ll admit you have now caught me off guard. And then they’ll
Terry:
Yeah.
Erik:
We need
Terry:
Yep.
Erik:
Not just role play, but throw them to the wolves. At a variety of live action, live interaction activities so that they can
is
Terry:
It’s those little moments of honesty and humility when they admit that they don’t know and they do bring somebody else in,
57:00-
Those are skills that adjust, are foundational skills that if they don’t get the opportunity to practice those.
Erik:
No, and it fills them with anecdotes that they can use. They remember this, and they remember when they said something, then it blew up in their face. I still remember saying, if you license my system, you won’t need as many people in your org. Oh. And then realize you never pitch that you were helping them with a headcount reduction.
Now, that still might be part of the value, but you never say that as a salesperson.
Terry:
Great. Great.
Erik:
One person, wouldn’t you like to be an analyst, not just an admin? Did. I’ve done this twice. I like being an admin. I’m very good at what I do and this is what I’m going to do until I’m 65 and retired and that’s the take that off the sales forecast.
’cause man, did you step in it there? so yeah, you’ve gotta go into those situations to make a mistake.
Terry:
Look e even with years of experience and a bit of
58:00-
gray in my beard, when we kicked, chat metrics off a decade ago, we were working around the messaging to try andfirstly find a niche market. Once we found a niche, then we’re talking specifically to senior level marketers. And, it took us quite some time to get the messaging right and I had this aha moment that, it’s no good telling marketers that they’re doing something not well.
That they’ve really, this problem that they’ve got here is they’re causing it. There’s a much better way to do it. that’s not the way to do it at all. That no, there’s no marketer that wants to hear that, they’re making a mistake or that they’re doing something wrong.
That’s for sure. And it was, as soon as we flipped the messaging around, to be, talk about more, the optimistic and opportunity side of things, it was much more successful and still works to this day, five years on.
Erik:
And that’s a question even for the, and back to the ais, how many of them can learn from their mistakes? does the AI go, wow, every time I say it that way, doesn’t go well, I should try a different path. someone was asking me, I pitched a project recently and I said,
59:00-
You’re a little more expensive in this case than somebody who is, bluntly, 20 years younger.
I’m yes, you’re absolutely right. what am I getting for that money? Things AI are only a couple years old, I said, you’re getting my scar tissue. You’re getting my mistakes, not my winds. My winds are delightful. Don’t get me wrong. But you’re also getting all of my mistakes, and I’m gonna keep you from making those mistakes.
I said, you might make new ones, but I will not recommend anything that was an old one that I have either done myself or directly observed or indirectly observed. That’s part of what you pay for. I don’t know how much the AI has actually learned from their mistakes versus,
Terry:
Think
Erik:
Yeah,
Terry:
Unless you built a circular system or process, which feeds back the ai, the outcome of each of their answers, how does it know? How does it know
the reality? Is it,
Erik:
Should actually, I’ll have to try, that would be a great one to add to that, in addition to the job
01:00:00-
Description, is if you asked your ai what mistakes have been made this week, what will it tell you?
can any of them tell me? ’cause maybe I wanna pass that on to my people who are gonna be working the booth at the live event next week. Come on, ai. Tell me all the things that you tried to say were, what’s the last thing you said before somebody hung up on you, disconnected? That alone should be a report.
Terry:
And that I think is where tools like Gong, where Sybil the call, call Zoom, call re recording tools that go through and actually do have some AI attached, and they will go through and analyze those calls where they’re seeing an outcome. I think that’s the benefit of those types of tools, but where you’ve got, or AI that’s giving you output to use somewhere else, unless it, if it’s disconnected, it’s never gonna get that feedback.
So yeah, I think that’s, excellent point with your work, really at full ca at full cast, to the Ag agent Summit. What’s the most underrated
01:01:00-
Strategic opportunity with AI right now that most leaders should lean into?
Erik:
Ooh,
that’s a good question.
I’ll tell you one for me is, it’s gonna sound funny. Statistical analysis
Terry:
Okay.
Erik:
Statistics is underutilized in the corporate world. we’re excited. If someone does an R squared in reality, how often do you see a multiple Inova?
How often do you see controlling for the three, these three factors? These are the relationships, and the reason we don’t is ’cause even those of us that took statistics in school or graduate school use the engines like SaaS or SPSS or even.
Microsoft Excel with the stats package add-on. If we haven’t touched in a while, we’re not doing it.
01:02:00-
There’s a lot of analytics that can be done at the company by feeding an AI and saying, Please do again, I would like you to do a statistical analysis of revenue versus type of customer, but control for the size of the company.
And oh, by the way, and then as a secondary, let’s follow on the type of the sales rep. this is again, if I were to write my job descriptions I need a data scientist at the firm now I
true data scientist. I don’t need the top-notch people. What do you need the data scientists do?
I wanna look at the following things. I bet you we could have an AI do that little thing that’s perfect for an ai because you’re only gonna give it your financial data sets your different lines. Remind yourself that you always come up with a hypothesis first and go on from there. This is a rant of mind as we all need to remember when we first took science classes as a child.
even our teens, we were taught the scientific method and now we’ve all forgotten. Start with a hypothesis, please.
Terry:
Yep.
Erik:
A hypothesis and run your AI and do an
01:03:00-
AI statistical analysis of things that are going on. Why are we churning customers? how many data elements do we have to look at? Oh, it seems that we churn more customers in this region.
Oh, that region has the weakest phone coverage, was the last to get the rollout of our latest stuff, or is furthest from our server. And time dilation still matters, a thing, because an international office and still hitting our AWS server in the United States ’cause we don’t feel like paying for, another instance in Asia.
Terry:
The power of what you’re talking about here. My mind is actually spinning in terms of one of the biggest challenges for startups in the early stages is making sure that their resources are allocated to their highest return activities
or funds, whatever it might be. I was gonna say, too often, almost always.
01:04:00-
They’re flying blind, working to assumptions and their initial hypothesis,
Erik:
Very rare that I’ve seen they get to a point where they take a stop check and go back through and check very scientifically back against that original assumptions or hypothesis.
Right.
Terry:
And so what you’ve just spoken about, I think is probably, and for startups that you take it to the next level when you’re looking for the next round of funding.
We’ve achieved X. The reason we’ve achieved X is why we know that the path forward is with these customers. We’ve gotta change this data set from there to here. We’ve making these changes. This is the, that’s gonna reduce our overheads by X. our efficiencies of the sales team’s gonna increase by y which mean the EBIT number is moving forward, are gonna be improved on this basis because our client acquisition costs will be down To be able to have that type of conversation.
with a level of, scientific accuracy based on the actual results and numbers and data that you
01:05:00-
Already have in an organization that’s been running, that just changes the conversation entirely. So
Erik:
Marketing
Terry:
I,
Erik:
is still one of those, who gets credit for the deal one of the great greatest questions ever.
Terry:
or worst. Or worst,
Erik:
Or yeah, or, yeah.
Terry:
yeah.
Erik:
Maybe it’s just one I’ve personally fought over
Terry:
Yeah.
Erik:
it gets into someone from that company first listened to, downloaded our podcast three years ago.
Are they still at the company? Oh, we should add that to the dataset. Okay, then this. Okay. So what’s the cost of customer acquisition? the podcast only cost us this much, but they also got scanned at Dreamforce, and that was, quarter million dollars that we spent at Dreamforce.
So a little allocated cost there. And now I can hear my managerial accounting professor from graduate school laughing at me because I hated the class. I saw no value in the class. And now she’s yeah, cost allocations. Eric. Yeah. Thank you professor Ho. I love you, Joanna. I’m glad you’re retired so you can stop harassing me.
but you keep on going down these paths.
01:06:00-
Could you get the AI to actually start doing attribution that starts telling you, because that’s a great thing. what leads to an opportunity? is it the podcast? Is it the downloaded white paper? Is it this with cookies and everything?
You should have you’ve got a ton of data in a bunch of places. Can you bring all that together and have the AI come back and say, here’s every touch. I’m now, I’m not gonna tell you how to value them, but I can tell you every touch. Now, how would you like to value ’em? Do you only wanna look at the touches by the decision makers in that five person committee?
Do you only wanna look at the touch? Do, but do you wanna look at the first touch, the last touch before it became an actually a sales qualified opportunity? Do you wanna look at the touches that happened after the sale qualified opportunity all the way to the close? don’t know that’s your business.
I’m just an
Terry:
Yeah.
Erik:
But I can analyze all of that for you and maybe you can start thinking about how to set the go-to-market budget for next year.
Terry:
for what you can see. And then beyond that, there’s this whole dark funnel component where, peer groups and, AI research, which you’re never gonna
01:07:00-
See that they’ve done. you’ve got this whole other dark component. Sure. These days with the technology that’s available, resolution on websites, you can start to get some indications in terms of what they’ve been on your website, where they’ve been, what they’ve seen.
And obviously if they engage with, the chat.
Erik:
With work from home, I have Google fiber to my home, so now I’m just gonna be another Google person whenever I’m
I’m on a v automatic VPN off my mobile. But I agree, but at least we can see the patterns, and all of that intent data that, six sense and ZoomInfo and others wanna sell us, are great.
that I’m not disparaging them. Please don’t come hunting me. I know all of you guys. but it’s still there. And once again, can we feed enough of that to AI to tell me what my marketing budget should be next year? let me know when the AI can actually do a really good marketing budget recommendation and it’s still not gonna understand the value of being at a place like, let’s just say Dreamforce.
01:08:00-
If that gets Mark Benny off to mention your company from the main stage. ’cause he walked by and thought you had a cool scarf.
Terry:
So statistical analysis, I think,
Erik:
I think statistical analysis and again, that could just be because I just don’t wanna have to do it again.
Terry:
It,
Erik:
I know the types of work I wanted to do. I don’t wanna have to remember how to do them.
Terry:
I think it’s one of those things that from any business’s perspective, there’s untold amounts of data that’s available. Management trying to make tactical and strategic decisions. If you’re swapping, shifting around your strategy too often you never see the true result of that strategy. So if you can build the statistical analysis around the original strategy and watch that and see what that’s actually doing and changing to the results in your business, yeah.
Look, I. it’s opened up a whole new, wound for me to dig into,
01:09:00-
A as far as that’s concerned, statistically ask. Okay. Eric, it’s been it’s been a wonderful conversation. I have one last question for you. unless there’s something that I’ve, I shouldn’t have asked you that I, have, is there, if I, miss something,
I wish we could have done this live down in Australia. it, we’ll get together at some point in time For sure. but I guess in my final question is really, if you had one billboard message for every sales and marketing leader chasing AI right now, what would you say? What would it say? What would that billboard message say?
Erik:
Know exactly what you want and how it could hurt you and know what the, know, what the flip side of the bet could be. ’cause you’re making a bet. What will you lose you don’t win? if I put a hundred dollars on a roulette table and I, it doesn’t come up. I know I’ve lost a hundred dollars. have put a hundred on the greens on zero or double zero or bridge the two.
01:10:00-
So I know what I could get and I know what I could lose. Do you know what your bet and a on AI could cost you? Not just in engine costs in a on to hire, but how wrong could it go and what could that do to your company? No. Both sides of the bet.
Terry:
No, both sides of the Be
Erik:
Everybody’s just
Terry:
No.
Erik:
The payout, they’re not looking at the risks.
And the risks are not just the cost of the engine.
Terry:
I like that. So if you’re gonna bet on ai, then know both sides of the bet,
Erik:
Yeah,
Terry:
The upside and the downside. Excellent. Excellent. Very good. Eric, thank you. It’s been an absolute pleasure chatting with you today. I concur would be, would’ve been much better if we could, be together, whether it be here or there.
And he, this guy just happened to carry a battle ax.
Erik:
A battle ax, at least. Yes.
Terry:
It’s, it, is.
Erik:
It’s a fascinating evolution of
01:11:00-
Technology. I’ve got bronze and bronze pieces from the Greeks. I’ve got some, early iron from the Romans, and I’ve got, a hilt protector, a hand protector from a samurai sword from the Mara of Japan, from when I had lived there, when I opened up an office for a company in Tokyo.
Terry:
There, there’s another question that I just have to ask. It seems to me that there’s a real theme here of weaponry, and so my the next natural question is, and as technology evolves, do you see the potential of AI being used as a weapon?
Erik:
Oh, it already is. it’s already feeding into the drone warfare. I Think the only reason they haven’t been completely unleashed is people are scared of that downside, as we talked about it,
fighting robot of some sort, be it the submersible drones, that are not AI driven, that are just simply high tech drones, in the ocean that the Ukrainians have used against the Russians, black Sea fleet, the aerial
01:12:00-
Drones, the working as, artillery spotters or flat out dropping explosives and.
I would say it is only the battery and weight of the chips that keeps them from completely unleashing them. But as people, as more and more forces as we’re seeing tested in certain areas right now finding that signals can be jammed. And so they’re using fiber optic lines for shorter distance. cable driven, armaments have been around, the 73 war had them, for example, in on the Sinai Peninsula.
Sooner or later someone’s gonna figure out how to put a smart enough engine in something the size of a cell phone and let that just drive it. And they’re
trust that it can identify the bad guys. That’s, so what’s our downside is our highly effective drone hits, hits our side and not their side.
Terry:
Yeah. Yeah.
Erik:
I would not be surprised if somebody shoots me a note. Some, one of my friends in the
01:13:00-
Defense industry is we already have that.
Terry:
Yeah.
Erik:
it just, nobody’s willing to be the first one to throw it on the battlefield to see what happens, but
Terry:
Yeah.
Erik:
No way it’s out there. It just, to my knowledge, it has a true and completely independent Has not been used yet.
Terry:
I think that comes down a little bit to your billboard on, know both sides of the bet.
Erik:
Yes,
Terry:
Absolutely. Eric, it’s been an absolute pleasure. Thank you so much for joining us on the chat. In particular, the Spotlight series of the ai, the Fine Line. it’s been an excellent conversation around there, and a lot of examples where AI has the potential of well and truly overstepping the fine line.
The examples you’ve given have been fantastic, so thank you. Appreciate you coming.
Erik:
Ah, thanks for the invitation. Great to be here.
2026 © All rights reserved Chat Metrics
Made with ❤ in Melbourne, Australia
