Picture slide seven of the quarterly board deck. The waterfall chart. Inquiries on the left, closed won on the right, five tidy bars descending in between, each stage annotated with a conversion rate carried to two decimal places. The room nods. The chart is clean, defensible, and describing a buyer who no longer exists.
Linear. Sequential. Comfortable. That is the world the waterfall assumes: a prospect enters at the top, descends stage by stage, and surfaces politely in your CRM at every checkpoint. Buying has not worked that way for years, and the distance between the chart and the behavior it claims to summarize is now wide enough to hide an entire pipeline problem inside.
Here is what actually happens. A buying committee forms quietly, months before a budget line exists. Six people research in parallel, not in sequence. They ask ChatGPT to explain the category and Claude to compare the vendors in it. They read reviews, lurk in peer communities, and paste AI-generated summaries into Slack threads you will never see. None of this touches your website. None of it fires a pixel. By the time someone fills out your form, the shortlist is written and the favorite is often already chosen. Your waterfall starts measuring at the exact moment the decision stops being interesting.
The MQL Mirage
The MQL was always a proxy. It never measured intent; it measured willingness to trade an email address for a PDF. That was tolerable when gated content was where research happened. It is not tolerable now, when a growing share of buyer research happens inside AI answers that your form cannot intercept and your analytics cannot see.
The mirage works like this: MQL volume holds steady, so demand feels fine. But the MQL is no longer a top-of-funnel signal. It is a bottom-of-funnel formality, the administrative step a nearly decided buyer performs on the way to pricing. Celebrating it as demand creation is measuring the exit and calling it the entrance. The form fill is not the start of the journey. It is the receipt.
The Attribution Apocalypse
Multi-touch attribution rested on one load-bearing assumption: the touches were visible. Weight them however you like, first, last, U-shaped, algorithmic. The model can only distribute credit across events it can observe.
Now watch the assumption fail. A director asks an answer engine which platforms handle her compliance requirements. The engine synthesizes your documentation, your reviews, and an analyst's comparison, then delivers a verdict in seconds. She visits nothing. Weeks later a colleague types your URL directly, and your attribution software solemnly records: Direct.
The software is not lying. It is faithfully reporting the shrinking fraction of the journey it can still see, and your budget reallocates toward that fraction every quarter. Money flows to what is measurable, not to what is working. That is the apocalypse. Not that attribution broke loudly, but that it kept producing confident numbers after it broke.
The Efficiency Trap
When pipeline softens, the reflex is to optimize. Cut cost per lead. Squeeze the paid mix. Retire the channels with the worst reported ROI, which, given the apocalypse above, are frequently the channels doing the invisible work.
This is the trap: you become steadily more efficient at harvesting the visible slice of demand while the invisible majority of the journey goes unattended. Efficiency inside a broken map does not get you to the destination faster. It gets you to the wrong destination with excellent unit economics.
The cost of inertia
We have watched this movie before. Across 26 years of building demand, since the first AdWords auctions, we have worked through four migrations: search, then SEO as a service, then social, then mobile. Each time, buyers moved first and measurement moved last, and the teams that repriced their assumptions early took share from the teams that defended their dashboards. AI answers are the fifth migration. The pattern is holding.
Inertia compounds quietly. Every quarter the waterfall gets presented, the model beneath it drifts a little further from reality, and the plan built on it allocates a little more budget to the wrong layer. Nothing dramatic happens. That is the problem. Erosion never books a meeting to announce itself.
The evidence-first alternative
The fix is not a prettier dashboard. It is treating go-to-market as a research program: informed priors instead of opinions, hypotheses with success criteria and kill lines written before launch, parallel experiments across the places buyers actually decide, and honest readouts that are allowed to say this is not working. One client, CloudResearch, described the posture in their public review: "They do not guess at where you stand, they show you the data and then tell you what to do about it."
Keep the waterfall if the board likes it. Just stop believing it is the territory. The question worth sitting with is simple: how much of your buyer's real journey does your reporting actually witness, and how long are you willing to keep funding the part it invents?
If you want to see what the engines say about you before your buyers do, ask us about the AI Search Diagnostic, or start the conversation with Danton.