The proof: an 85 percent dark-social funnel your software scores at zero

If the argument so far sounds like a hunch, the evidence is not. The clearest study on this ran for twelve months and produced numbers stark enough to end the debate.

Refine Labs (now Passetto / Vault GTM) ran a mandatory free-text "How did you hear about us?" field on their highest-intent form in parallel with their tracked analytics, across 620 declared-intent conversions. The two pictures barely overlapped. Software attribution sourced 78 percent of those conversions to web search; buyers themselves named web search only 12 percent of the time. Buyers instead attributed 85 percent of conversions to dark social, social feeds, podcasts, word-of-mouth and community, the channels software either undercounts or cannot see at all. Podcast and community produced zero in the software view, because no event ever fires for them.

The gap widens, not narrows, when you weight by money. Looking at closed-won revenue, software sourced roughly 79 percent to web search while customers credited it only 3 percent, and 98 percent of closed-won revenue traced back to dark-social origins in the buyers' own words. The single finding to print on a wall: the channel buyers named behind the overwhelming majority of revenue was, to the dashboard, completely invisible. Run that company off the report and you would conclude podcasts and community did nothing and cut them. The buyers would have told you they were the whole game.

Now the honest counterweight, because a one-sided case is a weak one. Self-reported data on its own is messy. Ruler Analytics found that around 72 percent of inbound demos give vague or no detail in the self-report field, the "Google", the "online", the "a friend told me". On its own, that is thin gruel. But the same research found that combining self-reported attribution with a tracked attribution tool reached 81 percent accuracy on lead source, against roughly 16 percent of attribution data being completely lost when you run no self-report at all.

Read it as overlap, not as a winner. The software sees the lower funnel precisely and the upper funnel not at all. The form sees the upper funnel directionally and the lower funnel poorly. The truth lives in the overlap, and you only get the overlap if you run both. The measurement gap is not an argument for throwing out your analytics. It is an argument for stapling a question to it.