Best of LinkedIn: AI in B2B Marketing CW 03/ 04

00:00:00: And, uh...

00:00:24: Sharpening strategies is definitely the theme today.

00:00:27: Yeah.

00:00:27: I mean, if you just step back and look at the posts from weeks three and four, the whole conversation feels like it's fundamentally shifted.

00:00:33: It really does.

00:00:34: It's like we've finally, finally moved past the, hey, look what this chatbot can write for me phase.

00:00:39: That feels like ancient history now.

00:00:41: Totally.

00:00:41: The novelties gone and, you know, the operators moved in.

00:00:44: We're not just chatting with bots anymore.

00:00:46: The real focus is on deploying autonomous systems.

00:00:49: It's much less about generation and way more about.

00:00:52: Execution.

00:00:53: Execution, that's the word.

00:00:54: And looking at the source material for this deep dive, we're really seeing three big shifts.

00:00:59: There's this rise of agentic workflows, which we need to unpack that term.

00:01:03: Then there's this kind of terrifying concept called the gluttonous pipeline.

00:01:09: And finally, the thing no one wants to talk about, data infrastructure.

00:01:13: It really is the move from AI as a toy to AI as a coworker.

00:01:19: And honestly, looking at the data, for some teams, it's a very expensive co-worker that doesn't really know what it's doing yet, but for others.

00:01:27: I mean, is rewriting the entire go-to-market playbook?

00:01:29: Okay,

00:01:30: so let's start with that first one, agentic workflows.

00:01:33: I think for a lot of people listening, the word agent just sounds like another fancy buzzword for a chatbot.

00:01:38: What's the real difference?

00:01:39: It's a crucial difference.

00:01:40: A chatbot, you know, just sits there, it waits for you to ask it something, then it gives an answer.

00:01:44: It's totally reactive.

00:01:45: Okay.

00:01:46: An agent is proactive.

00:01:48: You give it a goal and it figures out the steps to get there.

00:01:51: To really illustrate this, we've got to look at a post from Jonathan MK.

00:01:55: He was talking with Danielle Sacker, who's the VP of GTM at Mind Studio.

00:01:59: Mind Studio.

00:01:59: They're a big name in the custom AI space, right?

00:02:02: Exactly.

00:02:03: And Sacker isn't just talking theory.

00:02:04: She's running actual workflows that are automating what she calls the drudge work, the top five percent of research and prep.

00:02:11: that just burns out sales reps.

00:02:13: actually look like though because every sales tool out there claims it automates prep but you know it usually just spits out the company's about s-page.

00:02:20: This is different.

00:02:22: she's talking about deep deep pitch optimization.

00:02:26: her agents are pulling live research on a prospect in real time analyzing it against their value prop and then outputting a tailored strategy.

00:02:34: she's collapsing what used to be three hours of meeting prep.

00:02:38: Yeah, you know googling the prospect reading their ten K checking LinkedIn all of that.

00:02:42: She's collapsing it into two minutes.

00:02:44: Wow three hours to two minutes is what that's a massive efficiency game.

00:02:47: You're literally giving a rep half their day back.

00:02:50: you

00:02:50: are and she's claiming it helps them close three times more deals.

00:02:54: But it goes even deeper.

00:02:56: They're even automating things like contract reviews.

00:02:58: The agent scans the document and flags high-risk terms before a human lawyer even sees it.

00:03:05: So it's a filter.

00:03:06: It's doing the readings so the human can just do the thinking.

00:03:08: Precisely.

00:03:09: And this fits perfectly with something Isabella Bedoya wrote about Genspark.

00:03:13: That's a Silicon Valley AI agent unicorn.

00:03:17: Bedoya had this great line that really stuck with me.

00:03:19: What was it?

00:03:20: Most people use AI like Google.

00:03:21: Smart people use it like employees.

00:03:24: Use it like employees?

00:03:25: That's a really strong mental model.

00:03:27: Break that down a bit for us.

00:03:28: Well,

00:03:29: think about Google.

00:03:30: You ask for info, you get a list of links, and then you have to do the work.

00:03:33: If you use AI like an employee, you're assigning a job.

00:03:37: Bedoya talks about this all-in-one workspace where the AI isn't just answering a prompt, it's building a website or running market research or designing visuals.

00:03:46: It's executing.

00:03:47: Okay, but this is where I start to get a little skeptical.

00:03:49: We hear autonomous agents, and I just picture a manager pushing a button and then walking away to play golf, but that usually ends in disaster, doesn't it?

00:03:58: It absolutely does.

00:03:59: And that's where Carolyn Haley comes in

00:04:01: with a

00:04:02: big necessary reality check.

00:04:04: She points out that most marketing teams are just throwing agents at, say, content creation without any kind of map.

00:04:11: Without a map.

00:04:11: Yeah, without a process map.

00:04:12: They just say, agent, write me a blog post.

00:04:15: Yeah.

00:04:15: I mean, that's way too vague.

00:04:16: It's like telling an intern, go do marketing.

00:04:17: Yeah.

00:04:18: Haley says you have to map the workflow first.

00:04:21: A content workflow isn't one step.

00:04:23: Right.

00:04:23: It's topic ID, then research, then a first draft.

00:04:27: visuals, distribution.

00:04:29: So for every single one of those stages, you have to define the inputs, the outputs.

00:04:33: Exactly.

00:04:34: And, crucially, the human checkpoints.

00:04:37: The agent needs to know what done looks like for each step.

00:04:40: If you don't have that process map, you don't have an agent.

00:04:42: You just have a hallucination machine running at light speed.

00:04:44: You're just automating chaos.

00:04:46: It's interesting you bring up structure because looking at the notes, it seems like the European market is taking a slightly different angle on this.

00:04:53: They are.

00:04:54: Michelle Lieben pointed this out.

00:04:55: He said, if you look at European tools like Adio.

00:04:58: apify or lovable, they're not as obsessed with the generative stuff like writing poems or emails.

00:05:05: They're focusing really heavily on data extraction and CRM population.

00:05:10: So less creative, more...

00:05:14: more operational, it's less.

00:05:16: write me a LinkedIn post and more update these five hundred CRM records based on this call transcript.

00:05:22: Which let's be honest in B to P. that's often where the real money is.

00:05:25: It's all about data hygiene and accuracy.

00:05:27: It sets the stage for everything else.

00:05:29: That makes a ton of sense.

00:05:30: Reliability over just pure creativity.

00:05:32: Now speaking of reliability or maybe the lack of it, we have to move to the second big theme.

00:05:36: We've got these agents and of course people are pointing them at sales.

00:05:39: We have to talk about the explosion of the AISDR.

00:05:42: Oh yeah.

00:05:43: This was easily the most contentious topic in the source material this week.

00:05:47: There's a real divide forming here.

00:05:49: I can see why.

00:05:50: I mean, on one hand, you've got these incredible success stories.

00:05:52: Ryan Teas will share a case study about event lead follow-up.

00:05:54: that just sounds like a dream.

00:05:56: And that's a perfect use case for this, right?

00:05:58: You go to a trade show, you scan a thousand badges.

00:06:01: Normally, a human rep cherry picks maybe the top fifty hot ones, and the rest just die in a spreadsheet.

00:06:07: Right.

00:06:07: It's physically impossible for a human to follow up with everyone instantly.

00:06:11: Exactly.

00:06:11: So, Teasol... used AI to work one hundred percent of those leads immediately with personalization.

00:06:18: And the result was a three hundred fifty-two percent increase in pipeline.

00:06:21: Which is an incredible number.

00:06:23: So the argument there is pretty clear.

00:06:25: Speed and volume win.

00:06:26: If you can touch every single lead instantly, why wouldn't you?

00:06:29: In that very specific context, an event where people expect a follow-up, yes, but there's a really dark side to this kind of scale.

00:06:37: when you applied to cold outreach.

00:06:39: And this brings us to that term you mentioned, the gluttonous pipeline.

00:06:42: Exactly.

00:06:43: This comes from Maximus Greenwald.

00:06:44: He calls it the third deadly sin of AI marketing.

00:06:48: I love the biblical framing.

00:06:49: So what exactly is a gluttonous pipeline?

00:06:52: It's the illusion of success through just raw volume.

00:06:55: You're booking thirty meetings a week, so your dashboard looks amazing.

00:06:58: But you only close one deal.

00:07:00: You're just stuffing the funnel with more, but more isn't better.

00:07:03: You're just burning through your addressable market way faster.

00:07:06: You're annoying twenty-nine people just to get one customer.

00:07:09: Precisely.

00:07:10: And you're wasting your sales team's time with twenty-nine bad meetings.

00:07:14: Greenwald says the whole shift needs to be from volume to scoring.

00:07:17: Instead of using AI to blast everyone, use agents to rigorously rank leads by fit so you only talk to the ones that actually matter.

00:07:25: And the stats that back this up are... They're pretty sobering.

00:07:28: Mahesh Ayer shared a figure that really stood out to me.

00:07:31: He called it the point three percent problem.

00:07:33: That number should terrify any VP of sales.

00:07:36: Ayer mentioned a company that had a zero point three percent conversion rate from AISDR outreach all the way to revenue.

00:07:43: That's

00:07:44: statistically nothing.

00:07:45: You might as well just be guessing.

00:07:46: You basically are.

00:07:47: And Ayer explains why it happens.

00:07:49: The AI targeting is, you know, technically precise.

00:07:51: It matches keywords, but it has zero judgment.

00:07:55: It lacks the human nuance to understand context.

00:07:58: He had a pretty funny personal example of this, didn't he?

00:08:01: He did, and it's both hilarious and painful.

00:08:04: An algorithm pitched him Meti PCC sales training.

00:08:08: Now for anyone listening who doesn't know, Meti PCC is a very specific high-level sales methodology.

00:08:14: And Mahesh Ayer is an expert in it.

00:08:16: He

00:08:16: teaches it.

00:08:17: But the AI just saw that he engaged with Meta PCC content and said, aha, laid.

00:08:23: It couldn't tell a seller from a buyer.

00:08:26: Another one pitched him US only health plans when his profile clearly says he's in India.

00:08:31: That's

00:08:31: just sloppy.

00:08:32: It's worse than sloppy.

00:08:33: It's brand damaging.

00:08:34: And when he called them on it, the response was, sorry, this is an AI.

00:08:38: Ayer says that's not an excuse.

00:08:40: That's a confession.

00:08:41: A confession of laziness, really?

00:08:43: It's saying we didn't care enough to check.

00:08:44: So if the tools are this flawed, what's the fix?

00:08:46: Do we just not use them?

00:08:48: Coonstam has a good take on this.

00:08:49: The answer isn't don't use them.

00:08:51: It's don't trust them blindly.

00:08:52: He says, don't deploy tools, build systems.

00:08:55: OK, what's the difference in this context?

00:08:57: Well, a system has boundaries.

00:08:59: It has feedback loops.

00:09:00: It means you define your ideal customer profile and your personas so rigorously that the AI can't really mess it up.

00:09:07: And most importantly, you install daily feedback loops.

00:09:11: Meaning a human actually looks at what's going out the door.

00:09:14: Yes,

00:09:15: you check the output daily.

00:09:16: You catch those errors like pitching the teacher before they multiply by ten thousand.

00:09:21: can't just set it and forget it.

00:09:22: You need a human in the loop.

00:09:23: It sounds like we're moving from AI as a magic wand to AI as high maintenance intern.

00:09:30: That's a very good analogy.

00:09:31: An intern with infinite energy, but zero common sense.

00:09:36: And if an intern has no common sense, it usually means they lack context.

00:09:39: Which brings us right to our third theme.

00:09:41: Data integrity, the really unsexy plumbing of all this.

00:09:44: The foundation of the entire house.

00:09:47: And Sheriff Shahan from Common Room really hit the nail on the head here.

00:09:50: He says most teams are rushing to build all this AI on top of completely broken CRMs.

00:09:55: We've all heard garbage in, garbage out, but does AI change the stakes?

00:09:59: Shaheen argues it amplifies it.

00:10:01: It doesn't just process bad data, it weaponizes it.

00:10:05: So if you have duplicate contacts or old job titles, your AI will just confidently route leaves to the wrong person and personalize messages with false info.

00:10:14: at a speedy human could never match.

00:10:15: So instead of sending one bad email a day, you're sending a thousand an hour.

00:10:19: Exactly.

00:10:21: And Andrew Riesenfeld, looking back on his time at DocuSign, brings up this really critical technical concept he calls Context Graphs.

00:10:28: Context Graph?

00:10:29: That sounds pretty technical.

00:10:31: Can you break that down?

00:10:32: Sure.

00:10:32: Think about a huge company like Cisco.

00:10:34: In your CRM, you might have records for Cisco Systems, Cisco Meraki, Cisco California.

00:10:40: Riesenfeld calls it the Eighteen Cisco's Problem.

00:10:43: A human knows that's all the same company.

00:10:45: But a database doesn't.

00:10:46: A

00:10:46: basic database sees eighteen different companies.

00:10:49: And without what's called entity resolution, the ability to connect those dots, your AI treats them like total strangers.

00:10:56: It might pitch Cisco Meraki like they've never heard of you, even if you have a massive deal with Cisco systems.

00:11:00: That's just embarrassing.

00:11:01: It

00:11:02: leads to what reasonfell calls expensive autocomplete.

00:11:06: You're generating text, sure, but you're not generating value because the AI doesn't understand the relationships in your data.

00:11:12: Expensive autocomplete, that is a spinging line, but it feels so accurate for a lot of what's out there.

00:11:20: It should be.

00:11:21: And Sandeep Gulati breaks this down even more with his seven layers of the LLM stack.

00:11:26: We probably won't list all seven, but what's the core idea there?

00:11:29: The

00:11:29: core idea is that most leaders are only looking at layer three, which is the model.

00:11:33: They're asking, are we using GBT-IV or CLOD?

00:11:36: They think the intelligence is in the model itself.

00:11:38: But it's not.

00:11:39: But Gulati

00:11:39: argues that trust is built way down in layer one, your data sources, and layer two, pre-processing.

00:11:45: So the model's the brain, but layer one is the memory it's working with.

00:11:48: Exactly.

00:11:49: A genius brain with false memories is just going to hallucinate.

00:11:52: So the advice for leaders is stop asking what can AI do and start asking which layer of our data stack is the weakest.

00:11:59: Because if your layer one is a mess, your application is just going to be a very efficient way to annoy your customers.

00:12:03: That's a powerful way to frame it.

00:12:05: So we've got the agents, the SDRs, the data plumbing.

00:12:08: Where is all this heading?

00:12:10: Let's look at twenty twenty six and beyond.

00:12:12: The landscape is definitely changing and we're seeing new roles pop up.

00:12:15: Brendan Short highlights this fascinating trend, the rise of the GTM engineer.

00:12:21: GTM engineer, go to market engineer.

00:12:24: Is that a marketer who codes or a dev who sits in sales?

00:12:27: It's increasingly a developer sitting inside the revenue org.

00:12:31: Some of you can actually build these systems we're talking about, connect the API, set up the agents, and short links this to the idea of micro campaigns.

00:12:38: How is that different from what we do now?

00:12:41: Well, right now we build these massive lists, right?

00:12:43: Ten thousand people, blast them over a month.

00:12:46: A micro campaign is a super targeted list of say, fifty to two hundred fifty contacts.

00:12:51: that's only valid for maybe a week.

00:12:53: It's based on a specific buying signal.

00:12:55: Like they just hired a new VP of sales or installed a competitor's tech.

00:12:59: Exactly.

00:13:00: High speed, high precision and the cost efficiency is wild.

00:13:03: George Vico did a cost breakdown that's going to make some CFOs really happy.

00:13:07: A traditional Model Five SDR's legacy tech that costs about forty eight thousand dollars a month.

00:13:12: Okay, sounds about right for a team that size.

00:13:14: Versus the twenty twenty six model.

00:13:16: Yeah.

00:13:17: One GTM engineer, plus the AI infrastructure.

00:13:20: The cost.

00:13:21: Around twenty three thousand dollars a month.

00:13:23: Wow.

00:13:24: Half the cost.

00:13:25: Half the cost for potentially much better results because of that precision.

00:13:30: That creates a huge strategic gap between the companies that figure this out and the ones that don't.

00:13:35: It does.

00:13:35: But are companies actually figuring it out?

00:13:37: I mean broadly speaking.

00:13:38: That's

00:13:38: the big problem.

00:13:40: Adam Schoenfeld, analyze data from Lenny's podcast.

00:13:43: which is a huge bellwether in the product world, and found something really telling.

00:13:47: What do you find?

00:13:48: He just counted mentions of AI in product and engineering topics.

00:13:52: Five hundred eighty-eight mentions in GTM and sales topics, only fifty-eight.

00:13:56: That's a ten to one difference.

00:13:58: It is.

00:13:58: The product teams are surfing this wave and the sales teams are just sort of paddling.

00:14:02: Schoenfeld notes there's no cursor for sales yet.

00:14:05: You know cursor, the AI code editor.

00:14:06: Right,

00:14:06: developers love it.

00:14:07: It helps them rate code way faster.

00:14:09: There's

00:14:10: no real equivalent for GTM yet.

00:14:12: The mindset in go-to-market is still lagging.

00:14:15: We're still just trying to bolt AI onto old workflows instead of completely reimagining the workflow itself.

00:14:20: So it sounds like that GTM engineer role might be the key to bridging that gap.

00:14:25: I think so.

00:14:26: They bring that engineering mindset to the sales problem.

00:14:29: And this leads to one final really crucial thought on defensibility from Vladislav Muzhulovsky.

00:14:36: Defensibility is a huge topic.

00:14:38: Everyone's worried.

00:14:39: AI just makes everything easy to copy.

00:14:41: And Moziliski says they're right.

00:14:43: AI is killing what he calls feature moats.

00:14:46: Meaning just having a cool feature isn't enough to protect your business anymore?

00:14:50: Not at all.

00:14:51: Because code is cheap now.

00:14:53: If you build a cool feature, your competitor can use AI to replicate the code in a weekend.

00:14:57: So your defensibility doesn't come from features.

00:15:00: It comes from workflow ownership and trust.

00:15:02: So if you're deeply embedded in how a company operates like we talked about with those European tools managing the CRM, that becomes your moat.

00:15:09: That's your moat.

00:15:11: It's about building a system that's harder to rip out than it is to copy.

00:15:15: If your agent knows my business better than I do, I'm never leaving you.

00:15:18: That brings it all full circle, doesn't it?

00:15:20: It's not about who has the flashiest AI.

00:15:23: It's about who owns the workflow and who has the trusted data.

00:15:26: And

00:15:27: who has the discipline to map it all out before they hit go.

00:15:29: So if we recap this whole deep dive, it feels like we've charted a path from chaos to structure.

00:15:36: I think so.

00:15:36: First, we're moving from chatting with AI to deploying agents, but you have to map the workflow.

00:15:40: first.

00:15:41: Like Caroline Healy said, don't automate chaos.

00:15:44: Right.

00:15:44: Second, AISDRs can blow up your pipeline, but you have to watch out for the gluttonous pipeline, quality over quantity, and you need a human in the loop to avoid that .

00:15:53: three percent problem.

00:15:55: Third, your AI is only as good as your data.

00:15:58: If you haven't solved the eighteen Cisco's problem in your CRM, you are not ready for advanced agents.

00:16:04: Fix layer one before you even worry about layer seven.

00:16:07: And finally, the future seems to belong to that GTM engineer, the person who can actually bridge the gap between the tech and the strategy.

00:16:15: It's a brave new world out there.

00:16:16: Here's a thought to leave you with.

00:16:18: We talked a lot about efficiency and cost cutting today, but as AI makes average content and outreach basically free, the value of genuine human connection is going to skyrocket.

00:16:31: So, while you're building these agents, you should ask yourself, are you using them to replace human connection, or are you using them to clear the deck so you can actually have

00:16:39: more of it?

00:16:39: That is the ultimate question, isn't

00:16:41: it?

00:16:41: If you enjoyed this episode, new episodes drop every two weeks.

00:16:44: Also, check out our other editions on account-based marketing, field marketing, channel marketing, Mar-Tech, go-to-market, and social selling.

00:16:51: Thanks for listening, and don't forget to subscribe.