The Disappearing Funnel: Marketing Analytics in the Age of LLMs

Marketing Analytics in 2025: Navigating the LLM Black Box

If you want to understand the state of marketing analytics in 2025, picture this: you’re at a wedding, and the DJ has just handed over the playlist to an AI. The crowd’s requests are now filtered through a black box, and all you can do is watch people’s feet to guess if the music’s any good. Welcome to the new era of funnel tracking — where the customer journey has migrated from the open dance floor of the web into the velvet-roped VIP lounge of Large Language Models (LLMs), and marketers are left peering through the keyhole, trying to figure out if anyone’s actually dancing.

The Old Funnel: From Messy Middle to Measurable Metrics

Let’s break it down. For years, we obsessed over the messy middle — that glorious, chaotic space between awareness and conversion where customers bounced from Google to Reddit to your landing page, leaving a breadcrumb trail of clicks, scrolls, and abandoned carts. We built dashboards, attribution models, and enough funnel diagrams to wallpaper the Salesforce Tower. The premise was simple: if you could see it, you could optimize it.

The Incognito Customer Journey

But now, the customer journey has gone incognito. Instead of clicking through your beautifully optimized site, buyers are asking ChatGPT, “What’s the best HR software for startups?” or letting Perplexity or Google’s AI Mode do the heavy lifting. The decision-making happens in closed AI environments, and your analytics stack — once the all-seeing eye of Sauron — is suddenly wearing a blindfold. The old funnel? It’s not just leaky; it’s been replaced by a maze with invisible walls.

LLM Tracking Analytics: Reconstructing the Invisible Funnel

So, what’s a marketer to do when the funnel disappears into the AI ether? Enter LLM tracking analytics — the latest, greatest, and, let’s be honest, weirdest attempt to restore visibility to the customer journey. Here’s how it works (and why you should care):

• Synthetic data is the stuff you manufacture in the lab — think of it as the marketing equivalent of crash-test dummies. You feed carefully crafted prompts into an LLM (“Who’s the best CMO in San Francisco?” — spoiler: it’s not always me), and see how the AI responds. This tells you how your brand could show up in AI-generated answers, under ideal, controlled conditions.
• Observational data — the real-world clickstream. This is the digital exhaust from actual users: what they see, what they click, where they bail. It’s messy, unpredictable, and, crucially, it’s the only thing that tells you what’s actually happening out there in the wild.

Reconciling Synthetic and Observational Data

Here’s the catch: synthetic data is like practicing your TED Talk in front of the mirror. You look great, you sound great, but the mirror never interrupts you with a question about ROI. Observational data is the real audience — sometimes distracted, sometimes skeptical, occasionally asleep. The magic (and the migraine) comes from trying to reconcile the two.

Vendor Approaches: Creativity and Limitations

Vendors are getting creative. Some are brute-forcing the system, injecting thousands of synthetic personas into LLMs to see how the AI handles every possible scenario. Others are mapping the entire citation ecosystem, analyzing millions of AI responses to figure out where your brand gets mentioned, ignored, or — worst of all — misrepresented. It’s like sending an army of interns to every wedding in town to report back on whether anyone played your song.

The Uncomfortable Truth About Lab Data

But here’s the uncomfortable truth: lab data is a fantasy league. It’s useful for spotting bugs, bias, or missed opportunities, but it doesn’t predict what real customers will do when faced with a dozen competing answers and a toddler screaming in the background. Synthetic personas might help you stress-test your messaging, but they won’t tell you if your next campaign will actually move the needle — or just move your CEO to ask, “Why are we spending six figures on AI dashboards?”

Why This Matters: The Death of the Old Playbook

So why does this matter? Because the old playbook is dead. The funnel is no longer a neat sequence of stages; it’s a feedback loop, a choose-your-own-adventure where the customer might skip straight from “Who are you?” to “Take my money,” or wander off into a rabbit hole of AI-generated reviews. Attribution models that credit the last click are about as useful as giving the wedding DJ all the credit for a happy marriage.

Embracing Ambiguity: The New Marketing Mindset

The real risk? Marketers who cling to synthetic dashboards and vanity metrics are optimizing for a world that doesn’t exist. You can’t manage what you can’t measure, but you also can’t measure what you can’t see. The winners will be the brands that get comfortable with ambiguity, that use synthetic data to map possibilities but anchor their strategy in the messy, unpredictable reality of clickstream behavior.

LLM Tracking Analytics: A Flashlight, Not a GPS

Here’s my take: LLM tracking analytics are a necessary evil — a flashlight in the fog, not a GPS. Use them to spot opportunities, diagnose problems, and stress-test your assumptions. But don’t let the dashboards seduce you into thinking you’ve regained control. The real work is in comparing your ideal scenarios to what’s actually happening, and having the humility to admit when the two don’t match.

Instinct Over Illusion: The Poker Analogy

And if you’re still looking for a silver bullet, remember: marketing isn’t chess, it’s poker. You play the odds, read the table, and sometimes win with a bluff. The brands that survive aren’t the ones with perfect data — they’re the ones that know when to bet on instinct, when to fold, and when to change the game entirely.

Conclusion: Optimize for the Dance Floor, Not the Mirror

So next time you’re tempted to trust the lab results, ask yourself: are you optimizing for the mirror, or for the dance floor? Because in the age of AI, the only funnel that matters is the one your customers are actually moving through — even if you can’t see every step.

And if all else fails, just remember: in marketing, as in weddings, sometimes the best moments happen when nobody’s watching the DJ.