B2B Marketing with Fexingo · 2026-06-30 · 11 min
Key moments - from our scoring
Substance score
62 / 100
Five dimensions, 20 points each
B2B marketers face a scaling problem: personalization works at the account level, but breaks down when targeting hundreds of accounts with small teams. Lucas walks through a concrete case study of SecurEdge, a cybersecurity vendor, that solved this using AI-powered personalization infrastructure. They fine-tuned an open-source LLM on their content library, integrated it with a CDP and intent data provider, and built an API layer that generated personalized landing pages in under 500 milliseconds - with the top 50 accounts pre-generated nightly. The personalization wasn't creepy or obvious; instead of referencing a prospect's research behavior directly, the system re-weighted value propositions and emphasized relevant features based on industry and inferred intent. Results were substantial: 35% more meetings booked, 45% longer page engagement, 60% more content downloads, and accounts consuming three or more personalized pages were twice as likely to accept sales meetings. The technology stack involved a CDP, microservices, LLM APIs, and required three months of engineering effort plus ongoing human curation. Cost was negligible at scale - roughly five cents per page - but the real investment was upfront infrastructure. The discussion covers governance too: anonymized intent data, no personal data sent to the LLM, 90-day data retention, and opt-out mechanisms. Interestingly, the biggest engagement lift came from previously underserved long-tail accounts, not top-tier ones.
Use a fine-tuned LLM integrated with a CDP and intent data provider to generate personalized landing pages on the fly, keyed to industry and research signals. SecurEdge reduced page generation to under 500 milliseconds and pre-cached top 50 accounts, achieving a 35% lift in meetings with only five marketers.
The marginal cost per page is roughly five cents; for 400 accounts generating five pages each over a quarter, that's about $100 in compute. The main investment is three months of engineering time upfront, but once built, the infrastructure scales to thousands of accounts with minimal additional cost.
Don't explicitly reference a prospect's research behavior. Instead, re-weight and emphasize value propositions relevant to their industry or inferred intent. SecurEdge highlighted zero-trust capabilities for accounts researching zero-trust, without saying 'we noticed you were reading about zero-trust.'
Ensure intent data is anonymized, prospects have consented to its use, and no personal data (names, emails) is sent to the LLM - only industry, job function, and research topics. Implement data retention policies (e.g., 90 days unless the prospect converts) and provide one-click opt-out mechanisms.
Long-tail and previously underserved accounts show the highest lift, because they jump from zero personalization to some, whereas top-tier accounts already had human-crafted content. This democratizes personalization across the full target list.
Our reviewer’s read on each dimension, with quotes from the episode.
The episode delivers concrete technical and tactical insights about AI-personalization implementation - fine-tuning models on proprietary content, intent data integration, real-time page generation with caching, compliance guardrails, and role shifts for marketers. However, it occasionally slips into general framing ('the question isn't whether to use AI, but how to integrate it responsibly') and includes a brief, somewhat forced sponsorship message that dilutes focus. Most claims are substantiated with the SecurEdge case study, but deeper exploration of failure modes or nuanced trade-offs is limited.
they fine-tuned an open-source model on their own content library. Blog posts, case studies, whitepapers, product documentation. Then they integrated that model with their CRM and a third-party intent data provider.
They optimized it to under 500 milliseconds by caching common templates and only regenerating the variable parts.
The framing of AI-driven personalization at scale for mid-market accounts is practical and somewhat fresh in its focus on the long-tail, and the specific technical approach (fine-tuning on proprietary content + intent signals + real-time generation) is less standard than generic 'AI will personalize everything' takes. However, the underlying concepts (CDPs, intent data, LLM guardrails, human-in-the-loop) are well-established in the industry. The episode lacks truly contrarian claims or first-principles rethinking; it mainly illustrates execution of known ideas.
It democratizes personalization. Instead of only your biggest accounts getting a tailored experience, now every account can feel like a priority.
The marketers shifted from writing every page to curating the model's output and refining the prompts. They became content curators rather than content creators.
The episode features Lucas, who appears to be the host, and Luna, a co-host or recurring guest, discussing a case study about 'SecurEdge' - a company neither host appears to have direct operational experience with. There is no named external guest with direct expertise. The hosts are relaying a second-hand account without clearly establishing their own hands-on background in building or deploying AI personalization systems. This significantly weakens guest caliber; the content feels like educated commentary rather than first-hand practitioner insight.
A cybersecurity company I spoke with - let's call them SecurEdge
I spoke with - let's call them SecurEdge - they were trying to break into the mid-market enterprise space.
The SecurEdge case study is rich in specifics: 400 target accounts, 5-person marketing team, 35% increase in meeting bookings, 500ms page load optimization, 45% increase in page engagement time, 60% increase in content downloads, accounts consuming 3+ pages were 2x more likely to accept meetings, 3-month build timeline, ~$5 cents per page generation, 50 pre-generated pages for top accounts, 100-day data retention policy. These named metrics and concrete numbers provide credible evidence. However, 'SecurEdge' is anonymized, and no other real companies are cited as additional proof points, which limits the breadth of validation.
Over the course of a quarter, they saw a 35 percent increase in meeting bookings from those targeted accounts
They optimized it to under 500 milliseconds by caching common templates and only regenerating the variable parts.
The dialogue between Lucas and Luna is conversational and well-paced, with Luna posing natural follow-ups ('What about the results?', 'What about the technology stack?', 'Did they have any issues?'). However, the questions tend to be responsive and clarifying rather than genuinely challenging or probing. There is little pushback on claims, no skepticism about the 35% lift or whether other factors could explain the uplift, and no pressure on edge cases or limitations. The conversation reads more as structured education than as critical examination. The mid-episode sponsorship message also breaks conversational flow.
Wait, so the page was generated on the fly? That sounds like it could take a few seconds.
But I imagine there's a risk of the personalization feeling creepy - like, 'How did you know I was researching zero-trust?'
Computed from the transcript - who did the talking, and the words that came up most.
In this episode, Lucas and Luna explore how B2B marketers are deploying AI-powered personalization engines to tailor content, email, and web experiences for hundreds of enterprise accounts without manual effort. They discuss a case study: a cybersecurity company that used a large language model to dynamically generate account-specific landing pages based on intent signals from third-party data. The result was a 35% increase in meeting bookings for targeted accounts. The hosts break down the tech stack - CRM integration, API orchestration, and model fine-tuning - and address common pitfalls like over-personalization creep and data privacy. They also touch on how this approach changes the role of the marketer from content creator to content curator. If you've ever wondered how to make 'at scale' feel genuinely one-to-one, this episode is for you. #AI #PersonalizationAtScale #B2BMarketing #EnterpriseMarketing #IntentData #ABM #DemandGen #MarketingTechnology #LLM #ContentPersonalization #CybersecurityCaseStudy #SalesEnablement #MarketingOps #DataPrivacy #ContentStrategy #FexingoBusiness #BusinessPodcast #Marketing Keep every episode free: buymeacoffee.com/fexingo
Transcribed and scored by The B2B Podcast Index.
Lucas: So we talk a lot on this show about personalization in B2B, but usually it's at the account level - tier one accounts get a custom landing page, tier two gets a slightly tailored email sequence, and so on. Luna: Right, and that works when you've got a few dozen key accounts. But what about when you're targeting, say, three hundred accounts in a quarter? Lucas: Exactly.
That's where ai powered personalization at scale comes in. And I've got a concrete example. A cybersecurity company I spoke with - let's call them SecurEdge - they were trying to break into the mid-market enterprise space. They had a list of about four hundred accounts, but their marketing team was only five people.
Luna: So they couldn't possibly build custom landing pages for each one. What did they do? Lucas: They used a large language model - specifically, they fine-tuned an open-source model on their own content library. Blog posts, case studies, whitepapers, product documentation.
Then they integrated that model with their CRM and a third-party intent data provider. Luna: Intent data - so they knew which accounts were researching specific topics, like ransomware or zero-trust architecture? Lucas: Exactly. And here's the clever part: they built an API layer that, when a target account visited their website, would check the intent signals for that account in real time.
Then the model would generate a personalized landing page - the headline, the hero image selection, the three key value propositions - all tailored to that account's industry and recent research topics. Luna: Wait, so the page was generated on the fly? That sounds like it could take a few seconds. Lucas: They optimized it to under 500 milliseconds by caching common templates and only regenerating the variable parts.
They also pre-generated pages for the top 50 accounts each night, so those were already served instantly. For the long tail, they used the real-time generation. Luna: And the results? Lucas: Over the course of a quarter, they saw a 35 percent increase in meeting bookings from those targeted accounts compared to their previous approach, which was a generic landing page with some basic company-name personalization.
Luna: That's a huge lift. But I imagine there's a risk of the personalization feeling creepy - like, 'How did you know I was researching zero-trust?' Lucas: That's a real concern. They mitigated it by not referencing specific intent signals directly.
So they wouldn't say, 'I see you've been reading about zero-trust.' Instead, the page would emphasize their zero-trust capabilities as a prominent feature. The personalization was in the emphasis, not in the data revelation. Luna: So it's about adjusting the hierarchy of information rather than calling out the prospect's behavior.
That feels more respectful. Lucas: Exactly. And they also made sure that any account could opt out of personalization entirely with a single click. That's important for compliance and trust.
Luna: What about the technology stack? I'm guessing they needed a decent data infrastructure to pull this off. Lucas: They did. They used a customer data platform, or CDP, to unify the intent data with their CRM.
Then they built a microservice that called the LLM API. The CDP also handled the opt-out logic. The whole thing took about three months to build and test. Luna: So it's not something you can spin up in a weekend.
But for a company with the right technical resources, it seems highly replicable. Lucas: Absolutely. And the model didn't just generate landing pages. They also used it to personalize email subject lines and body copy for the same accounts.
For example, if an account was in the healthcare vertical, the email would reference HIPAA compliance naturally. Luna: Did they have any issues with the model generating inaccurate or off-brand content? Lucas: They had a human review loop for the first few weeks. Every generated page was reviewed by a marketer before going live.
They also set up guardrails in the prompt - the model was instructed to avoid making specific claims about product features unless it was pulling directly from the documentation. And they had a fallback: if the model's output didn't meet a confidence threshold, it would serve a generic page. Luna: I like that. So it's not a black box.
You're using AI to augment, not replace, human judgment. Lucas: That's the key. The marketers shifted from writing every page to curating the model's output and refining the prompts. They became content curators rather than content creators.
Luna: That's a big shift in job role. How did the team feel about that? Lucas: Mixed at first. Some were excited about the scale, others worried about losing creative control.
But once they saw the results and realized they could focus on strategy and messaging frameworks, most came around. The team actually grew because the demand gen pipeline got bigger. Luna: So the technology didn't replace jobs - it changed them. Lucas: Exactly.
And that's a pattern we're seeing across B2B marketing. The question isn't whether to use AI, but how to integrate it responsibly. Luna: Speaking of which - you know, we've been sharing a lot of practical marketing frameworks on this show, and we deliberately keep it ad-free because we think that's the best way to deliver real value. Lucas: Yeah, and the response has been great.
If these conversations have sparked something you've actually used in your own work, and you want to support that kind of independent content, there's a simple way to do it. Luna: Exactly. It's buy me a coffee dot com slash fexingo. No subscription, no perks, just a way to keep the show free for everyone.
Lucas: And that's it. Back to the tech - one thing I found really interesting was how they measured success beyond just meetings booked. They also tracked page engagement time and content consumption per account. Luna: I was going to ask about measurement.
So they saw higher engagement? Lucas: Significantly. Average time on page increased by 45 percent for personalized pages compared to generic ones. And the number of content downloads - whitepapers, case studies - went up by 60 percent.
Luna: That makes sense. When the content feels relevant, people actually read it. Lucas: Right. And they also tracked which accounts went on to engage with sales.
They found that accounts that consumed three or more personalized pages were twice as likely to accept a meeting invitation. Luna: So the personalization wasn't just a gimmick. It was actually qualifying leads better. Lucas: Exactly.
It helped prioritize which accounts were genuinely interested. Sales loved that. Luna: What about the cost? Running an LLM at scale isn't cheap.
Lucas: They estimated it cost about five cents per page generation, plus the fixed cost of the model fine-tuning and infrastructure. For four hundred accounts, even if each generated five pages over a quarter, that's maybe a hundred dollars in compute. The bigger cost was the three months of engineering time. Luna: So the marginal cost is tiny.
The investment is upfront. Lucas: Exactly. And once it's built, you can scale to thousands of accounts with essentially the same infrastructure. That's the beauty of it.
Luna: I'm curious - did they run into any privacy or compliance issues beyond the opt-out? GDPR, CCPA, that kind of thing? Lucas: They did have to work with legal to ensure that the intent data they were using was properly anonymized and that they had consent for its use. They also made sure that no personal data - like names or email addresses - was ever sent to the LLM.
The model only saw industry, job function, and research topics. Luna: That's a smart precaution. It also reduces the risk of the model accidentally memorizing sensitive data. Lucas: Yeah.
And they had a data retention policy: any personalized page data was deleted after 90 days unless the prospect converted. Luna: So they were thinking about this from multiple angles - not just marketing effectiveness, but also governance. Lucas: Absolutely. And I think that's the blueprint for any B2B marketer looking to implement AI personalization at scale.
Start with a clear use case, invest in data hygiene, build in human oversight, and measure everything. Luna: One thing I'm wondering: how do you decide which accounts get the AI treatment? Is it all accounts, or do you tier them? Lucas: In SecurEdge's case, they started with their top 100 accounts - tier one - and then expanded to the full list of 400.
They found that the long-tail accounts actually had a higher lift in engagement because they went from zero personalization to some, whereas the top accounts already had some human-crafted content. Luna: That's interesting. So the biggest gains come from accounts that were previously underserved. Lucas: Exactly.
It democratizes personalization. Instead of only your biggest accounts getting a tailored experience, now every account can feel like a priority. Luna: That's a powerful shift. I think a lot of marketers will want to try this.
What would you say is the first step? Lucas: Audit your current content library. If you don't have a solid base of blog posts, case studies, and product documentation, the model won't have good material to learn from. Then get your data infrastructure in order - CRM, CDP, intent data source.
And then start small. Pick ten accounts, build a prototype, and measure the difference. Luna: Start small, prove the concept, then scale. That's a classic B2B playbook, but with a modern twist.
Lucas: Exactly. And the tools are getting easier to use every quarter. What used to require a dedicated data science team can now be done with a couple of engineers and an API key. Luna: So the barrier to entry is lowering.
That's good news for smaller marketing teams. Lucas: It is. And I think we'll see this become table stakes in the next couple of years. Just like dynamic content in email used to be a differentiator, now it's expected.
Luna: Right. So the question becomes: how do you do it better than your competitors? Lucas: And that's where the human touch still matters. The AI can generate, but the human curates.
The best campaigns will combine speed and scale with strategic insight. Luna: Well, that's a great place to leave it. Next episode, we'll look at another angle of AI in B2B marketing - this time, how it's changing content syndication. Lucas: Looking forward to it.
Until then, keep experimenting.
Other episodes covering the same guests and topics, from across The B2B Podcast Index.
Radiology Can't Keep Up. Here's Where AI Actually Helps | Dr. Nina KottlerRethink Imaging · on Large language models96 / 100
Episode 7: AI & the Power of a "Thin Core"Architecting the AI Enterprise · on Prompt engineering82 / 100
The Hidden Risk in AI-Generated Tests and Requirements - Olivier DenooSoftware Testing Unleashed · on Prompt engineering79 / 100
Guest data and direct bookings with the top 3 CRM solutions on Mews MarketplaceMatt Talks Hospitality: Real conversations for innovative hoteliers · on Customer Data Platform (CDP)79 / 100
John Suarez on How Businesses Should Actually Use AI Without Losing Human JudgementMarketing for SMEs · on Prompt engineering78 / 100
Our Cybersecurity Mission: AI In DevelopmentOur Cybersecurity Mission · on Prompt engineering77 / 100
How B2B Marketers Use White Space Analysis to Map Enterprise Account Growth57 / 100How Enterprise Sales Teams Use Social Selling to Close DealsHow B2B Marketers Use Teardown Videos to Win Enterprise DealsHow B2B Marketers Use Intent Data to Prioritize AccountsHow B2B Marketers Use Expansion Revenue to Grow Enterprise Accounts