There is an argument going around B2B marketing that attribution should be treated as a compass rather than a GPS. You cannot observe every touchpoint, so stop trying. Use directional signals, invest more in what consistently shows up near closed revenue, and accept that certainty is not available.
The argument is correct. It is also extremely convenient for the people making it loudest, because lowering the standard of proof from “this caused revenue” to “this correlates with revenue” makes upper-funnel advertising much easier to defend, and the companies advancing the compass framing sell upper-funnel advertising and the tools that measure it.
Both of those things can be true at once. This post takes the framing seriously and then asks the obvious follow-up question: what stops a compass from pointing wherever you were already walking?
Why the compass argument is right
The numbers behind it hold up, and they come from companies that publish their methodology.
Dreamdata’s 2026 LinkedIn Ads Benchmarks Report, drawn from over 66 million sessions and 3.5 million customer journeys, puts the average B2B buying journey at 272 days, up from 211 the previous year. Around 220 of those days, roughly seven months, happen before the buyer enters the sales pipeline at all. The average journey now runs to 88 touchpoints across 10 stakeholders, up from 76 touchpoints and 6.8 stakeholders.
Separately, Factors.ai reported in September 2026 that buyer engagement begins an average of 124 days before an opportunity is created in the CRM.
Those two figures measure different milestones, entry into the sales pipeline in one case and opportunity creation in the CRM in the other, so they are not interchangeable. They point the same way. A last-click model applied to a 272 day decision is measuring the final few weeks of a nine month process and calling the result the cause. That is not a small inaccuracy. It systematically credits whatever happens closest to the signature and starves whatever happened in month two, which is frequently the thing that got you onto the shortlist.
So yes, directional measurement is the honest description of what is actually available. Nobody is wrong about that.
And why it is convenient
Here is the part the vendor version of this argument tends to skip.
At 88 touchpoints across 10 stakeholders over 272 days, almost everything you did will appear somewhere in almost every deal you won. “Appears consistently in closed deals” is a very low bar when the journey is that long and that crowded. A model built on that standard can justify nearly any line of spend, which is precisely why it is comfortable to adopt and uncomfortable to test.
There is a detail in the Dreamdata report worth holding onto. Its headline finding is that LinkedIn delivers 121% return on ad spend against 67% for Google Search and 51% for Meta. The methodology note says that figure comes from a data-driven attribution model on closed-won deals over twelve months, and that impressions are excluded.
So the benchmark everyone quotes is click-based, while the advice published alongside this class of research is usually to give view-through engagement more weight. Those are not the same measurement, and it is worth knowing which one you are repeating in a board meeting.
This is not an accusation that anyone is lying. The research looks solid and the methodology is disclosed, which is more than most marketing statistics manage. It is a point about incentives: when an argument happens to favour the person making it, the argument needs checking harder, not less.
Guardrail one: write the prediction down before you move the money
Before you shift budget on a directional signal, write one sentence: what you expect to change, by roughly how much, and by when.
If you only form a view about what the data meant after the numbers arrive, you do not have a compass. You have a narrative with a chart attached. The discipline costs nothing and it is the difference between a measurement practice and a justification practice.
Guardrail two: hold something out
The only thing that separates correlation from causation is withholding the thing and seeing what happens.
Hold out a geography, a segment, or a period. Keep everything else constant. This is cheaper and more practical than most teams assume, and it is the single most informative exercise available in B2B measurement. One clean holdout tells you more than a year of dashboards.
A compass that has never once been checked against a landmark you already know is not an instrument. It is a decoration.
Guardrail three: fix the lookback window before you look at the result
The choice between a 180 day and a 365 day lookback window is a real one and it changes the answer materially. Choose it from your actual sales cycle, not from what makes the report look better. If your median cycle is nine months, a 90 day window is wrong. If your median cycle is six weeks, a 365 day window will credit things that had nothing to do with the sale.
Decide once, write it down, and do not quietly change it because the first result was disappointing. Changing the window after seeing the output is the most common way an honest team produces a dishonest number.
Guardrail four: ask who benefits from the model
When a platform, partner or agency recommends an attribution approach, ask one question: what does this model do to the numbers they report to you?
A model that lengthens lookback windows and credits view-through engagement will increase the measured contribution of display and paid social. That may be entirely correct. It is still worth saying out loud before you adopt it.
That question applies to us too. Brilad sells AI search work, and part of the argument below is that an unmeasurable slice of the journey is growing. That conclusion happens to make our services look necessary. Apply the same scepticism to it that you would apply to anyone else’s, and ask us for the holdout.
The part that genuinely is getting harder
One real change sits underneath all of this. A growing share of the pre-pipeline research now happens inside AI assistants, which return a complete answer and no referrer. That is not a tracking gap you can close with a longer lookback window, because no signal is created at all. We wrote about the measurement side of this in full-funnel attribution for B2B and about the people it hides in the buying group you cannot see.
The honest response is not to widen the model until everything is attributable. It is to measure that part directly, by asking the assistants the questions your buyers ask and recording what comes back, and to keep it separate from your attribution reporting rather than blended into it.
What to actually do
- Accept the directional framing. The arithmetic of a 272 day, 88 touchpoint journey leaves no alternative.
- Set your lookback window from your real sales cycle, document it, and leave it alone.
- Run one holdout per quarter. Pick the channel you are least certain about, not the one you most want validated.
- Write the prediction before the budget moves.
- Keep AI search visibility as its own measured number rather than folding it into attribution.
- When someone hands you a model, ask what it does to their reported performance.
If you want the simpler version of the underlying calculation, that is in how to calculate SEO and GEO ROI.
A compass is a good metaphor. It is worth remembering that a compass is useful because it points at something external and fixed. An instrument that confirms the direction you were already heading is not a compass, whatever it is called.
Sources: Dreamdata LinkedIn Ads Benchmarks Report 2026, published 10 March 2026, and Factors.ai, September 2026. Both are vendor-published research, which is the reason this post treats them the way it does.