Attribution sounds like an enterprise problem. It is not. The wrong attribution model quietly misleads small teams into cutting the channels that actually work, and that mistake is more expensive than the data team you think you cannot afford.
Here is the trap. A small B2B company runs last-touch attribution because it is the default in their tool. Content marketing rarely gets the final-click credit, so it looks worthless. They cut it. Six months later pipeline dries up, because content was what brought in and educated the buyers who later converted through another channel. The model got the wrong channel cut.
What attribution actually is
Attribution is how you assign credit for a sale to the marketing that led to it. A B2B buyer might find you through a search, read three posts, get a referral, attend a webinar, then convert. Attribution decides which of those gets the credit.
The model you pick changes which channels look valuable. That is the whole reason this matters. It is not an academic exercise. It is the lens you make budget decisions through, and a distorted lens distorts the decisions.
Single-touch: simple and wrong in opposite directions
First-touch gives all the credit to the channel that first brought the buyer in. Last-touch gives all the credit to the final interaction before the sale.
First-touch over-credits awareness channels and under-credits the channels that close. Last-touch does the reverse. Both are easy to run. Both are wrong, just in opposite directions. The mistake is treating either one as the truth instead of as one viewpoint.
Used together, though, they are useful. First-touch tells you what brings people in. Last-touch tells you what closes them. The gap between the two views tells you where the assisting work is happening. Two simple wrong models, read together, point at something closer to right.
Multi-touch and data-driven: more accurate, more demanding
Multi-touch attribution spreads credit across several touchpoints. Linear gives equal credit to every touch. Time-decay gives more to touches near the sale. Position-based weights the first and last touch heavily.
Data-driven attribution goes further and uses an algorithm to assign credit based on what actually correlates with conversions in your data. [Google Analytics 4 uses data-driven attribution as its default now](https://support.google.com/analytics/answer/10596866). It is the most sophisticated option.
The catch for small companies: these models need clean data and enough conversion volume to be reliable. A B2B SMB closing a handful of deals a month does not have the volume to make a data-driven model trustworthy. The algorithm needs patterns, and a small number of conversions does not produce stable patterns. The sophisticated model can be less reliable than the simple one when the data is thin.
The dark funnel breaks every model
Here is the thing no attribution model handles: much of the B2B buying journey happens where tracking cannot see it. Private conversations. Podcasts. Slack groups. A recommendation over coffee. None of it shows up in any model, because the touchpoints are invisible.
This is not a small gap. For B2B it is often most of the journey. The buyer who converts through a Google search of your brand name did not start there. Something sent them to search your name, and that something is in the dark funnel where no tool can follow. The dark funnel post covers how large this blind spot really is.
What a small B2B team should actually do
Stop looking for the one correct model. There is not one. Run a blend.
Use last-touch and first-touch as two reference points. Watch assisted conversions in GA4 to see the channels doing the middle work. And add a "how did you hear about us" field to your lead and demo-request forms.
That last one is the highest-value attribution practice a small B2B team has, and it is free. The self-reported answer captures the dark-funnel touchpoints no model can see. A buyer who writes "heard you on a podcast" just told you something GA4 will never know. For B2B, the self-reported answer often correlates with reality better than the technical attribution does.
You do not need a tool to start
GA4 gives you data-driven and assisted-conversion views for free. The form field costs nothing. A CRM ties the lead to the eventual deal. That is enough to run honest attribution for a small B2B company.
Dedicated attribution tools earn their cost at higher volume and complexity. Below that, they are a line item that produces precision you cannot act on. Start with GA4, the CRM, and the self-reported question. Add tools when the volume justifies them, not before.
The takeaway
Attribution is not about finding the one true model. It is about not letting a single distorted view talk you into cutting a channel that works.
Run a blend. Read first-touch and last-touch together. Watch assisted conversions. And ask every lead how they heard about you, because for B2B the honest self-reported answer beats the technical model that cannot see the dark funnel.
When to ignore your attribution model (and what to do instead)
There are moments when the attribution data is actively misleading and following it would be a mistake. Knowing when to set the model aside is as important as knowing how to read it.
Ignore your model during channel launches. A new channel — a podcast, a LinkedIn organic push, a referral program — has almost no data yet. Any attribution model will show it performing poorly, because it has not run long enough to accumulate conversions. Cutting a three-week-old channel because the attribution looks weak is not analytical discipline; it is impatience dressed up as data.
Ignore your model when deal volume is below 20 conversions a month. Data-driven attribution needs enough events to find real patterns. Below that threshold, the algorithm is producing confident-looking numbers from noise. You will make worse decisions following a data-driven model on thin data than you would following your judgment plus the self-reported question.
Ignore your model when a channel shift coincides with an external event — a competitor going dark, a platform algorithm change, a referral burst from one partner. The model will attribute the change to whatever channel happened to be active. It cannot see the real cause.
What to do instead in each case: go back to first principles. Talk to the customers who closed. Ask your sales team what conversations they are having. Run the self-reported form question and read the answers manually. Direct observation beats a confused model every time.
The attribution stack for a lean B2B team: what you actually need to run
Most attribution guides describe enterprise stacks. Here is what a team of two to ten actually needs to run, using tools you likely already have.
Layer one: GA4 with a custom channel grouping. Set up a channel group that separates your traffic accurately — branded search, unbranded search, direct, LinkedIn, referral, email. The default GA4 groupings lump things together in ways that misrepresent your specific mix. This takes one afternoon to configure and pays back indefinitely. Use the assisted-conversions report, not just the last-click report.
Layer two: the CRM source field. Every new contact in HubSpot or your CRM of choice should have a "lead source" field that the sales team fills in based on what they know, not just what the tracking recorded. Sales often knows things the pixel never captured.
Layer three: the self-reported form question. "How did you hear about us?" on every lead form and demo request. Review these manually once a month. Look for patterns that contradict what GA4 is telling you. When they conflict, the self-report is usually closer to right for B2B.
Layer four: a monthly 15-minute attribution review. Not a dashboard you build once and never open. An actual monthly ritual where someone looks at the three layers together, flags any channel that the data is consistently undercounting versus what the sales team reports, and adjusts the budget conversation accordingly. That review, done consistently, is worth more than any attribution tool you could buy.