Section 1
Key takeaways
• Attribution software measures the last trackable touch, not the moment a buyer decided, and on the same pipeline, those can be wildly different sources. Refine Labs found podcasts drove 53% of self-reported influence and 0% of software-attributed revenue . • The decision usually forms before you can see it: 6sense found the winning vendor was already on the buyer's Day One shortlist 95% of the time , and 94% of buying groups ranked vendors before first contact . • When you simply ask buyers how they found you, word of mouth ranks #3 (18%) while podcasts barely register (0.14%), proof that "hyped" and "effective" are different axes . • You don't measure the dark funnel. You feed it and read its shadow: the direct/branded demand it produces downstream. • Channels that look free in a click model and channels that look expensive can both be lying. Only a holdout test tells you which spend actually creates buyers.
Section 2
Why your attribution dashboard is lying to you (politely)
Start with a concrete case. A B2B services firm, say a 12-person fractional-CFO practice billing high-five-figure retainers, runs paid search, publishes a weekly newsletter, and the two partners go on industry podcasts. Their dashboard says 78% of closed deals came from "direct" or "branded search," about 15% from paid, and the rest scattered. The obvious read: branded search is the engine, cut the podcasts, double the Google budget. That read is backwards, and you can prove it without leaving their own data. "Branded search" means someone typed the firm's name into Google. People don't type a name they've never heard. Something made them search, a partner's podcast appearance, a referral from a former client, a comment in a CFO peer group. The branded search isn't the cause of demand. It's the receipt for demand created upstream, in the dark. This is what Refine Labs named the Attribution Mirage: the tracker reliably credits the channels closest to the click (search, retargeting, "direct") and reliably blanks the channels that actually create the want. In their 12-month study of 620 declared-intent conversions tied to $21.5M in closed-won ARR, podcasts drove 53% of self-reported influence, $11.4M of revenue, and registered exactly 0% in software-based attribution . Same pipeline. Same buyers. A roughly 90% gap between what moved them and what got credited. If you'd run that firm by the dashboard, you'd have killed the channel responsible for over half the revenue and poured money into the channel that was just collecting names the podcast generated. The mirage doesn't make you slightly wrong. It makes you confidently, systematically wrong in the exact direction that hurts most.
Section 3
What "direct traffic" actually is
It helps to be precise about the jargon, because the labels do the deceiving. Attribution is the practice of assigning credit for a conversion to one or more marketing touches. Last-touch attribution gives 100% of the credit to the final touch before conversion. Multi-touch attribution spreads credit across the touches it can see. The shared, fatal assumption is that the touches it can see are the touches that mattered. "Direct traffic" is the tell. In analytics, "direct" is a residual bucket, it's where the platform files any visit it can't trace to a referring source. Someone hears your name on a podcast during a commute, searches it that night on their phone, then types the URL on their laptop at work two days later: direct. Someone gets your case study forwarded in a private Slack and pastes the link into their browser: direct. "Direct" rarely means "they navigated straight to you out of brand love." It usually means "the dark funnel happened here and we lost the thread." So when a founder says "most of our business is word of mouth, we don't really know how it works," they've actually described their growth engine accurately. The problem is they've also described it as unmanageable, and that's the part that's wrong. You can't tag the dark funnel, but you can read it. We'll get to how. First, the harder truth about when the decision happens, because it reframes the whole job. If positioning is upstream of all of this, so is the fact that most of your market isn't ready to buy today, the demand you're reading was planted long before the click.
Section 4
The decision is made before you can see it
The dark funnel would be a manageable measurement nuisance if buyers made up their minds late, during your sales process, where you have visibility. They don't. 6sense's 2025 Buyer Experience Report, built on roughly 4,000 vetted buyers (each required to have spent at least $25,000 in the prior two years), found the winning vendor was already on the buyer's Day One shortlist 95% of the time, up from 85% the year before . Read that slowly. The vendor who wins is, almost always, a vendor the buyer had already mentally selected before the formal evaluation began. The shortlist isn't built during the deal. It's built in the dark, and the deal mostly confirms it. It gets sharper. In the same body of research, 94% of buying groups had ranked their preferred vendors before first contact, and they bought that preliminary favorite 77% of the time . As 6sense's research team put it: "Buyers are choosing a preliminary winner much earlier than they have in the past." The implication for how you spend is uncomfortable: by the time a lead is trackable enough to attribute, the preference that determines the outcome has usually already set. And the window keeps shrinking. The same report shows roughly 60% of the buying journey now happens before sellers are engaged, a shift from a 70/30 to a 60/40 split, while the point of first contact slid from about 69% of the journey to 61%, and the average cycle compressed from 11.3 months to 10.1 . The trackable portion of the funnel, the part where your CRM and your ad pixels can actually watch a human behave, is getting smaller every year. You are trying to steer using a window that's closing. This is why "just get better attribution software" is the wrong instinct. Better software measures the visible part with more precision. The visible part is shrinking. You're buying a sharper lens for a smaller and smaller slice of the picture. The work isn't to see the dark funnel directly. It's to manage the thing you can't see by the evidence it leaves behind, the same way you infer wind from a moving flag.
Section 5
Does the hyped channel even work? Ask the buyers
Here's the counterintuitive twist that keeps founders honest: the dark funnel does not reward the channels you'd expect. It's tempting to read "podcasts drove 53% of influence" and conclude the move is to chase whatever's loud and trendy. The data says no. HockeyStack's 2024 Self-Reported Attribution Report compiled 8,528 responses from high-intent buyers who were asked, at the point of booking a demo, how they actually heard about the vendor. Word of mouth ranked #3 at 18%, 1,544 responses, ahead of every paid and content channel except search and social. Podcasts, the channel that dominates marketing-Twitter discourse? Named by 12 people out of 8,528. That's 0.14% . Hold both findings in your head at once, because they're not contradictory, they're the whole point. In Refine Labs' specific portfolio, podcasts were a dark-funnel powerhouse worth $11.4M . Across HockeyStack's broad cross-section, podcasts were a rounding error and word of mouth was a top-three engine . The lesson isn't "podcasts win" or "word of mouth wins." It's that the dark funnel is specific to your buyers, and the only way to know which dark channel works for you is to ask them, not to copy whatever a louder company swears by. This is the trap behind most "what's working in marketing now" content. It reports the average dark channel and dresses it up as a universal one. For a fractional-CFO firm, the dark engine might be a single closed CFO community and a handful of repeat referrers. For a vertical SaaS shop, it might be one industry podcast and a Slack group. The channels are invisible to your tracker either way. The difference between guessing and knowing is whether you've built a habit of asking. Once you know which dark channels move your buyers, the next question is what pushes them to start looking at all, which is the work of mapping the triggers that move buyers into the market.
Section 6
The cost of optimizing toward the visible
Before the framework, sit with the failure mode, because it's the expensive one and almost everyone has lived it. When you optimize toward what the tracker can see, you don't just under-credit the dark funnel, you actively defund it. The budget review comes. Search and retargeting show clean, attributable ROI because they're standing closest to the conversion, collecting receipts for demand other channels created. The podcast, the community sponsorship, the founder's writing, the conference talk, they show "no pipeline" because their influence got filed under "direct." The rational-looking move is to cut the unattributable spend and feed the attributable spend. So you do. And for a quarter or two, efficiency metrics look great, because you're still harvesting demand the dark channels generated before you cut them. Then the pipeline thins, not dramatically, just a slow softening of inbound, a quieter top of funnel, deals that suddenly need more nurturing because buyers arrive less convinced. The dashboard can't explain it, because the dashboard never saw the thing you cut. You optimized the harvest and starved the planting, and the lag between the two is exactly long enough to hide the cause. The same lag shows up in how fast a buying signal goes cold, which is why the way intent decays when you wait belongs in the same operating picture. That lag is the dark funnel's signature, and it's also the key to managing it. If the damage from cutting a channel shows up months later in untracked demand, then the right instrument isn't an attribution report. It's a system built around the lag.
Section 7
The BGA framework: the Shadow Demand Ledger
You don't measure the dark funnel. You feed it, then read its shadow. The Shadow Demand Ledger is three instruments that together let you manage spend you can't attribute, replacing the false precision of click-attribution with three honest, lagging echoes of real demand. Run all three; any one alone misleads. 1. The Ask, capture self-reported attribution at the moment of intent Put one open or semi-open field on every demo request, inquiry form, and discovery call: "How did you really hear about us?" Phrase it as a human question, not a dropdown of your marketing channels, dropdowns force buyers to pick the option closest to "direct" and bury the real answer. This is the method that surfaced word of mouth as the #3 source at 18% in HockeyStack's data, a signal completely invisible to GA4 . • Concrete action: Free-text field, made prominent but optional, asked at the highest-intent moment (demo booking, not newsletter signup, where intent is too low to bother answering honestly). • Metric / rule of thumb: Aim for a response rate above ~70% of high-intent inquiries. Tag answers into a handful of buckets monthly. Treat any channel that self-reports materially higher than it attributes as a confirmed dark-funnel engine, protect its budget, don't cut it. • Caveat to stay honest: Self-reported data is fuzzy. HockeyStack flagged roughly 20% of responses as invalid . You're not after decimal precision; you're after the order of magnitude the tracker is hiding. "Roughly a fifth of our serious buyers name our podcast and the tracker says zero" is all the precision this decision needs. 2. The Echo, watch branded and direct demand as the trailing signature Dark-funnel activity you can't tag still leaves a visible wake: branded search volume, direct traffic, inbound DMs, and unprompted referrals. These are lagging indicators of upstream influence. When a partner does a strong podcast run or a piece of writing circulates in a community, you won't see it as "podcast → pipeline." You'll see branded searches tick up two to six weeks later, and "where did you hear about us" answers start naming things you can't track. • Concrete action: Track branded/direct volume as a weekly time series. Annotate it with your dark-funnel activity, every podcast, talk, big post, community sponsorship. You're not proving causation per event; you're watching whether the baseline of "people who already want us" is rising or falling. • Metric / rule of thumb: Branded-search and direct demand should grow at least in line with your dark-funnel investment. If you increase founder visibility and branded demand stays flat for a quarter, the activity isn't landing, adjust the content, not the measurement. If you cut a dark channel and branded demand softens 4–8 weeks later, you just found your cause-and-effect the hard way; reverse it. • Why it works: It respects the lag. The echo arrives after the cause, which is exactly why click-attribution misses it and why a time series catches it. 3. The Lift, prove channel impact with holdout tests, not click trails For the channels you can control geographically or by segment, stop asking "what did this click lead to?" and start asking "did demand rise where we ran this versus where we didn't?" This is incrementality testing, you hold out a comparable audience or region from a channel and measure the difference in demand. It's the only method that survives the dark funnel, because it measures the outcome (more buyers) instead of the path (which is untrackable). • Concrete action: Pick a channel and a controllable boundary, two similar regions, two matched account lists, two time periods. Run the channel in one, hold the other dark, and compare total qualified demand, not attributed clicks. • Metric / rule of thumb: A channel earns continued or increased budget only if the exposed group produces measurably more pipeline than the holdout. Run it as a recurring discipline on your largest line items, not a one-off. If a "high-ROI" attributed channel shows little lift when you hold it out, it was harvesting, not creating, reprice it accordingly. • Reality check: Holdouts cost you something, deliberately not marketing to a group feels wrong and forfeits some near-term pipeline. That cost is the price of truth, and it's far cheaper than a year of defunding the wrong channel. Reserve it for decisions big enough to justify the spend. Together: The Ask tells you which dark channels buyers credit. The Echo tells you whether dark activity is growing demand over time. The Lift tells you which spend actually creates buyers versus merely collecting them. Attribution tells you who to thank for the last click. The Shadow Demand Ledger tells you what's making buyers. Once you can see which channels create demand, nurturing the buyers who aren't ready yet keeps the harvest from leaking. If you want to go deeper on the demand analytics can't see, the Growth Reader develops the thinking behind reading demand you can't attribute.
Section 8
You're running the Shadow Demand Ledger right when…
You're running it right when a budget meeting can't end with "cut the podcast, it shows no pipeline", because someone pulls up the self-reported data and the branded-demand time series and the room can see the channel's shadow even though the tracker can't. You're running it right when "direct traffic" stops being an answer and starts being a question you investigate. When you've made peace with the fact that your best channel may never be fully attributable, and you've stopped trying to force it to be, you feed it, you watch its echo, and you holdout-test the spend big enough to matter. You're running it right when you'd rather be approximately correct about where demand comes from than precisely correct about where the last click came from. And you're running it wrong the moment a clean attribution report talks you into defunding a channel your buyers keep naming with their own mouths.