Last-touch attribution lies to you about developer buyers, and it lies in a specific direction: it credits the channel that was standing closest to the signup and erases every channel that did the actual convincing. A developer reads a Hacker News thread on Tuesday, sees your tool mentioned in a Slack community on Thursday, opens your docs at midnight the following week, and types your name into Google because they already decided. Your analytics record that last step and stamp the signup “branded search” or “Direct.” The thread, the community, the docs, the coworker who first said your name out loud all show zero. So you defund the things that worked and pour money into the thing that got the credit, which was just the finish line. This post is how I stop that from happening.
What is the dark funnel, and why does it hit developer tools so hard?
The dark funnel is all the evaluation that happens in places you cannot instrument, before a buyer ever identifies themselves to you. For developer tools it is not an edge case. It is most of the funnel.
Think about where a developer actually forms an opinion about your product. A Hacker News comment section. A Discord where someone asks “what are you all using for this.” A conference talk they half-watched from the third row. Your open-source repo they starred on a weekend. A Reddit thread from eight months ago that still ranks. Your docs, read at 1am, with no account and no cookie consent that matters because they are just skimming. None of that fires a tracked event tied to a person. By the time they sign up, the persuasion is done and it happened entirely offstage.
This is worse for devtools than for most B2B categories because the audience is allergic to the parts of the funnel you can see. Developers block trackers, use privacy-focused browsers, ignore gated content, and refuse to fill out a form to “download the whitepaper.” The exact behaviors that make a great technical buyer also make them invisible to your analytics. If you sell to AI engineers specifically, it is more extreme still, because that buyer runs their own trial against their own data before they will talk to anyone.
The scale is not a hunch. Gartner found that B2B buyers spend only 17 percent of the total purchase journey meeting with potential suppliers (Gartner). The other 83 percent is research you are not in the room for. For a self-serve developer product, even the 17 percent overstates what you get to witness, because a lot of buyers never take a sales call at all.
Why does last-touch attribution credit the wrong channel?
Because it is built for a world where buyers move down a visible path, and developers do not move that way.
The classic model assumes a click, a form, a nurture sequence, a conversion, each step logged and stitched to one identity. That path is legible. Last-touch just reads off the final step and assigns full credit. When the real journey is a click on an ad, followed by a form fill, followed by an email that gets opened, the model is at least pointed at something real. It is still crude, but it is not lying.
Developer journeys break every assumption underneath it. There is no form, because they signed up self-serve. There is no ad click, because they came from a link a coworker pasted into Slack. The “first touch” your tools can see is a branded Google search, which is not a first touch at all. It is the shadow of a podcast episode, a launch, or a friend’s recommendation. Branded search is what demand looks like after something else created it. Crediting it is crediting the last domino and congratulating it for the whole chain.
There is a timing problem stacked on top. At any given moment most of your potential buyers are not in the market, which is the premise behind the widely cited 95-5 rule from the LinkedIn B2B Institute: roughly 95 percent of business buyers are out of market at any time. The developer who signs up this week may have first heard of you seven months ago when they had zero need. No attribution window on earth reaches back seven months to the talk that planted the seed. So it credits whatever happened in the last few days, which was the easy part.
So should I just give up on measuring where signups come from?
No. Give up on precision, not on knowing. The goal shifts from “assign one source to every signup” to “understand which investments move the whole business.” Three things get you there, and none of them is a fancier attribution model.
Ask people directly. Put a “how did you hear about us” field on signup. Make it a free text box, not a dropdown, because the dropdown you write will not include “my coworker Priya told me” or “that thread on Hacker News,” which are the answers you most need to see. Self-reported attribution is messy and it is also the cleanest signal you get when your tracking is blind, because it captures the dark-funnel touch the buyer actually remembers. Read the raw answers every week. Do not just tally them into buckets. The specific phrasings are where you learn which podcast, which community, which talk.
Ask again on sales calls. Any time a human talks to a prospect, whether that is a founder-led sales call or an onboarding chat, ask how they first came across you and what made them look closer. You will get a richer story than any form field: the sequence of touches, which one tipped them, who else on their team was already using it. Log it somewhere structured. Ten of these are worth more than a month of UTM data for understanding your real funnel.
Watch whether the curve moves. This is the one people skip and it is the most honest. When you do a strong launch, land a well-placed conference talk, or get a big mention in a popular newsletter, does total qualified signup volume rise in the days and weeks after? You will not be able to draw a clean line from that talk to a named signup. You do not need to. Track your baseline, make the investment, and watch the aggregate. If dark-funnel spend consistently correlates with the whole curve lifting, that is stronger evidence than any single-touch model pretending to know.
How do I report this to a founder without looking like I am dodging the numbers?
Lead with the honesty and then show the movement. Tell the founder plainly that developer purchases mostly happen where you cannot track them, so precise per-channel attribution is a fiction, and any dashboard claiming otherwise is confidently wrong. Founders who build technical products understand instrumentation limits better than most executives. They will respect “here is what is real and here is what is noise” far more than a pie chart with false decimals.
Then report what you can defend. Show total qualified signups over time as the trend line that matters, annotated with your big dark-funnel bets, so the launch, the talk, and the newsletter placement sit on the timeline where the founder can see what lined up with a lift. Show the “how did you hear about us” answers as a ranked list with real quotes, not just percentages. Show the self-reported sourcing from sales calls next to it. The picture that emerges is directionally true, which is worth infinitely more than a precise number that points at the wrong thing.
A line I repeat: report the direction with confidence and the magnitude with humility. You can say Hacker News, your docs, and word of mouth are driving evaluation, because your surveys and calls say so every week. You cannot say 34 percent of pipeline came from Hacker News, and you should refuse to, because that number is invented. The same discipline applies at the top of the funnel, which I get into in the post on measuring AI search visibility: measure what is real, watch the trend, and never manufacture precision the data cannot support.
What does this change about where you spend?
It stops you from starving the channels that do the real work. Once you accept that branded search and Direct are mostly downstream effects rather than sources, you quit optimizing for the finish line and start funding the things that create demand in the dark: the launches, the community presence, the docs worth reading at midnight, the content a developer forwards to a teammate. Those are the touches your surveys keep surfacing and your last-touch dashboard keeps zeroing out.
None of this means abandoning measurement. It means measuring the funnel you actually have instead of the one your analytics tool wishes you had. For the mechanics of instrumenting the part of the journey that does happen inside your product, I wrote that up in how to measure a product-led funnel. This post is the other half: making peace with the part you will never see, and getting good at reading its shadow.
