Holly Enneking, VP of Marketing at Markup AI and co-author of Startup CXO, has spent the past year watching attribution data quietly stop meaning anything, and she’s got a better way to measure what’s actually working.
Here’s the thing about marketing attribution. It was built for a world where a buyer clicked a link, then another link, then landed on a demo page. That world is gone. Buyers now read an AI summary, watch a LinkedIn video a friend sent them, lurk in a Slack community, scroll Reddit, and then show up to book a demo already most of the way to a decision. None of that leaves a UTM parameter behind. Holly Enneking has spent the last year in the middle of that breakdown, and her take is blunt: the dashboards still show numbers, everyone just assumes those numbers are still true.
Why is marketing attribution breaking down right now?
Enneking’s read is that attribution wasn’t so much broken by AI as exposed by it. The tracking infrastructure (UTMs, click paths, last-touch models) was already shaky. AI-driven search and “dark social” just made the gap between what’s tracked and what’s actually happening impossible to ignore.
“Attribution models were really built for a world where buyers were clicking on links in a sequence. Now, buyers are reading some AI summaries, they are watching a LinkedIn video that someone sent them, they’re finding content in Slack communities… What’s really hard is that the dashboards are still showing something, and so then there’s an assumption that what is being tracked is accurate, but so much of it is being missed.”
— Holly Enneking, VP of Marketing, Markup AI
That’s a dangerous gap for any team reporting up to leadership or a board. The dashboard hasn’t gone blank. It just quietly stopped being the full picture, which is arguably worse than no data at all.
If you can’t trust the dashboard, what should you actually measure?
Enneking’s answer is to stop trying to patch attribution and start tying activity directly back to revenue, the one language every leadership team and board already speaks. Concretely, that means going back to basics: having sales ask prospects how they actually found you.
“I think really the best thing that marketers can be doing is figuring out how to find ways to connect what you’re doing back to revenue… We’ve been doing this with our team and have gotten some really fantastic insights around prompts that they were using in ChatGPT, for example, in order to learn about us… There would be no attribution even with Google Analytics’ new rollout of additional source information. That wouldn’t be data that I would have if it hadn’t been asked.”
— Holly Enneking, VP of Marketing, Markup AI
Beyond direct asking, she points teams toward pipeline velocity and churn trends, broader signals that aren’t tied to any single channel or tactic but that reveal where things are actually moving. Pair that with a real understanding of what’s in-market and who it’s targeting, and you can start making informed calls about where to lean in and where to pull back, even without a clean attribution trail.
There’s a human wrinkle worth calling out here too, one David flagged in the conversation: when you ask a buyer how they heard about you, they might mention only one touch out of ten, but it’s the one that stuck. Learning which touch actually lands is only possible if you ask.
What does personalization actually mean beyond a first-name merge field?
Enneking breaks personalization into two buckets. The first is the basics: accurate names, companies, locations. Get that wrong and any personalization effort collapses instantly, because bad personalization destroys trust faster than generic messaging ever would.
“Bad personalization is just immediate trust loss, lack of credibility… it is hard to come back from something like that, especially when there’s so much content, so much potential for engagement. Your audience, it just takes one wrong move to lose somebody.”
— Holly Enneking, VP of Marketing, Markup AI
The second bucket is experiential: meeting someone based on the context they’re actually in. That might mean tailoring a follow-up based on what happened at an event, or adjusting language and use cases for different personas rather than blasting every ICP with identical copy. David connected this directly to what Two Segment (BDEX’s newest product) is built to do, using data enrichment and AI-driven personality analysis to segment audiences not just by demographics or purchase history, but by how different people actually think and respond, since a message that moves a pragmatic buyer won’t move a creative one the same way.
On the flip side, Enneking is wary of over-segmenting to the point of burnout. Building 50 variations of a campaign because AI makes it technically easy doesn’t mean it’s useful.
“Is 50 different iterations actually serving you and helping you more effectively communicate?… I think this is absolutely a quality over quantity game of what is the most important thing for you to learn from these different segments.”
— Holly Enneking, VP of Marketing, Markup AI
She also flagged MQLs as a vanity metric worth retiring in their current form. Hitting a lead volume number means nothing if those leads never turn into a discovery call, a booked meeting, or an opportunity.
Can AI replace human judgment in marketing analytics, or just support it?
This is where Enneking got the most direct. AI is good at surfacing patterns; it’s not equipped to explain why those patterns exist or what to do about them. A churn spike could signal a product gap instead of a customer success failure. An underperforming campaign could be a creative miss or simply bad timing. Telling those apart takes a person who knows the context. Pattern-matching alone can’t do it.
“I worry for the companies that I see who are like, ‘I laid off my entire marketing team and I run everything through Claude or ChatGPT.’ What you have done is you have really stripped away any of the judgment, the knowledge, the expertise, the lived day-to-day work of a marketer… AI is really great at helping us see what is happening. We need a human still who can tell us why.”
— Holly Enneking, VP of Marketing, Markup AI
She also raised a subtler failure mode: feed an LLM lopsided data (lots of marketing detail, thin sales detail) and it will hand back an analysis skewed toward marketing simply because that’s what it had to work with, not because marketing is actually the bigger problem. Garbage in isn’t always obvious. Sometimes it’s just uneven in.
“LLMs are really built on the past. They are not forward-looking… Being able to apply what you know and what you’re experiencing in real-time, and where you think things are moving in the future, that is a real lens that an LLM simply cannot apply.”
— Holly Enneking, VP of Marketing, Markup AI
The bigger picture
Every thread in this conversation ties back to the same idea: the old proxies for success (click-through rate, MQL volume, last-touch attribution) are measuring a world that doesn’t exist anymore, and clinging to them just because they’re familiar is its own risk. Enneking’s advice isn’t to throw out the data; it’s to hold it more loosely, contextualize it constantly for your team, and keep asking the qualitative question, “where did you actually hear about us?”, that no dashboard will ever surface on its own.
The uncomfortable part is that this isn’t a problem AI is going to solve for you. If anything, AI-driven search and content consumption is what broke the old model in the first place. What replaces it is judgment, communication, and a willingness to say “here’s what we’re seeing, here’s what it might mean, and here’s the giant grain of salt to take it with,” repeated as many times as it takes for the rest of the org to actually absorb it.
Connect with Holly Enneking on LinkedIn or visit markup.ai to learn more about building AI-native content workflows.
Watch the full episode on YouTube.
This article was adapted from an episode of Deconstructing Data, BDEX’s weekly podcast on data-driven marketing. Tune in live every Thursday at 4:15 PM Eastern on LinkedIn.
About BDEX: For companies that need clean identity data to power their products, BDEX offers unmatched quality and execution. Visit bdex.com and click “Talk to an Expert” to get started.
