Open your CRM source report. Organic Search, Direct, a slice of Paid, a long tail of Other. Now think about your last five deals and how those buyers actually found you. A post someone forwarded. A referral. A talk. A thread in a community you are not even in. Chances are the two lists do not match.
This is for founders and CROs at early stage to Series C companies deciding where next quarter's marketing and event money goes. The usual advice is to buy a better attribution tool. For a founder led company the better fix is smaller: one free text question on your contact form, the answer logged word for word, and read beside the software field. That is self reported attribution: you ask the buyer, in their own words, how they found you, and you write the answer down. Keep both fields. Where they disagree is the finding.
The source field records the last thing the software saw
A CRM source field is usually filled from tracking data: the referring page, a campaign tag, the page the form sat on. That data is honest about what it saw. It cannot see a conversation in a group chat or a post that was forwarded.
In a typical setup, a buyer who hears about you in one of those places and then searches your company name is recorded as Organic Search. A buyer who types your address straight in is recorded as Direct. That is an inference about how most setups behave, so check how yours is configured. But the logic is hard to escape: the field credits whoever stood closest to the form.
The result is a report that favors capture over creation. Search and direct collect demand. They rarely start it. Fund channels by that report alone and you will pay most for the part of the journey that was already decided, and starve the part that decided it.
One company's numbers, and how far they travel
Refine Labs, a B2B demand generation consultancy, tested this on its own pipeline. In an article by its chief executive, Megan Bowen, dated October 15, 2024, the firm reported a 12 month test across 620 conversions, collecting software attribution and a self reported answer for each one.
Software credited web search, direct and organic, with 78% of conversions. Self reported answers put what Refine Labs calls dark social, meaning social media, podcasts, word of mouth and communities, at 85%, while customers named web search as their source only 12% of the time. Read those figures carefully. The firm's own note says the self reported shares are counted as mentions, and buyers often name more than one source. So the two methods sit on different bases, and the numbers should not be subtracted from each other.
Read it for what it is. One consultancy, reporting on its own conversions, from people who had already raised a hand, in an article that also sells its services. It is not a benchmark for B2B, for founders or for you. What it shows is narrower and still useful: on the same people, two methods can give very different answers. Your gap might be smaller. It might be as wide. You will not know until you ask.
The fix is one question and a second field
Four steps, none of them expensive.
- Add the question to every contact and demo form. "How did you hear about us?" Mandatory. Free text. No dropdown, no suggested answers. Refine Labs describes using exactly that on its own form, so the buyer answers unaided. A dropdown hands them your categories instead of their memory.
- Ask it out loud on the first call. Then follow with a second question: "What was the first thing you saw?" The follow up is ours, not Refine Labs's, and its job is to get past the last thing they remember to the first.
- Store the answer in its own field, word for word. No mapping to a category at entry. The buyer's sentence is the data.
- Leave the software source field alone. Replacing it throws away the only thing that tells you where demand was captured. Refine Labs calls the combination hybrid attribution: software shows where demand is captured, the buyer shows where it was created. We agree with the split.
Read a few dozen answers, not a model
Most attribution advice assumes volume. With thousands of conversions a month you need classification rules and modeling. A founder led company with a few dozen inbound conversations a quarter does not. That is our view of scale, not a statistic. At that size you can read every answer in twenty minutes.
Once a month, put the two fields side by side for every new opportunity. Apply one rule: the first source the buyer names wins, and if they name several, log the first and note the rest. Then mark each row as agreeing or disagreeing with the software.
The disagreements are the output. Picture a row where the CRM says Organic Search and the buyer wrote that a founder in a group chat sent them your post. No dashboard would have told you that the group chat is a channel, or that you were not counting it.
Three spend decisions the answers should gate
Before you cut a channel. If tracking gives it almost no credit, check whether buyers name it. A newsletter, a LinkedIn presence or a community can show nothing in the software and plenty in the answers. RivoAxis has written about founders collecting likes while the pipeline stays empty. The answers are how you tell which of the two you have.
Before you scale a paid channel. If the software gives paid the credit but buyers never mention it, you may be paying to collect demand that would have arrived anyway. That is a hypothesis to test, not a verdict, but you should hold it before you double the budget.
Before you buy a booth or a sponsorship. Events are where tracking breaks hardest, because the meeting happens in a hall and the form gets filled in weeks later. We covered how event credit goes missing in the post event pipeline leak. The verbatim field is the cheapest way to see whether an event is being named at all.
Where self reported answers break
The objection is fair. Blend, a B2B marketing and HubSpot agency that promotes its own services, lists recency bias and vague answers such as "Google Ads" among the pitfalls. Buyers remember the last thing they saw. They write "LinkedIn" when they mean one post by one person. They name three sources, or forget. Drop that into a spreadsheet as gospel and you have replaced one bad instrument with another.
All true, and it is why this approach keeps both fields and adds the follow up question. Treat the buyer as a second witness. Two witnesses who disagree tell you where to look. Two who agree tell you the software was probably right for that deal, which is also worth knowing.
There are conditions where we would not lean on it. With fewer than ten inbound conversations a quarter, which is our rule of thumb and not a research finding, you do not have a pattern. Talk to each buyer instead. And never allocate budget to a decimal point from this data. Use it for direction, and for the questions it makes you ask.
Give the field an owner
Source is one of the six definitions we told founders to write before a first RevOps hire. The same discipline applies here. One person owns the verbatim field. One rule classifies the answers. Nobody edits a buyer's words afterwards. If a rep can overwrite the answer with the label they prefer, you have rebuilt the software field with extra steps.
It is the same trap we described in why "lost to price" is not a reason: a CRM field that looks like a fact and works like a guess. Loss reason sits at the back of the funnel and source sits at the front. Both need the buyer's own words before anyone trusts them.
Somewhere between your pipeline report and your last ten buyers there is a gap, and next quarter's budget will follow whichever one you trust. A Revenue Diagnosis conversation with RivoAxis settles which source of truth sets that budget and who owns the field.
Book a Revenue Diagnosis