Attribution Readout — 2026-10-05

The question

Four tools report 35 to 60 conversions for the same month. Which number do we trust, which model do we use, and where should the budget go? This is a fictional B2B SaaS with a 90-day sales cycle.

Source of truth

The CRM, at 35 closed deals, defines how many conversions happened. It's the only system where revenue is real. The other sources only explain where those deals came from.

There is a timing problem to fix first. With a 90-day cycle, the 35 deals closed last month were mostly sourced from touches 2–4 months ago. The ad platforms' 50 and 40 are probably mostly leads, demos, or form fills in the last 30 days. So the numbers are likely not measuring the same event in the same period. Before comparing anything, confirm what each tool counts as a "conversion":

What each source says

Google and Meta together claim 90 conversions against 35 real deals. Never sum them. Overlap is certain, and some of what they claim are leads that never closed. GA4's 60 above the CRM's 35 is a sign it's counting something other than closed deals. Don't try to make these tie out. Explain the gap instead.

Model comparison

Use first-touch and position-based (U-shaped) side by side, computed from CRM or warehouse data. Do not use the GA4 or Google Ads last-click or data-driven views as the decision model.

• Last-touch will over-credit branded search and direct. Those are the final steps of a 90-day journey, not its cause.
• Data-driven attribution needs roughly 200 conversions and 2,000 interactions a month. At 35 deals, it would likely collapse toward last-click.
• First-touch shows what creates demand. Position-based (40/40/20) shows what created the opportunity and what closed it, which fits the B2B "created" and "closed" moments.
• The gap between first-touch and last-touch is itself the insight. Channels that rank much higher in first-touch are your demand creators.

Confidence and gaps

• Direct at 45% of sessions is a measurement problem, not a channel. It's probably a mix of stripped referrers, dark social, dropped UTMs, app-to-web handoffs, and people who found you elsewhere and typed your name. It most likely hides real top-of-funnel work. High confidence on this diagnosis, low confidence on the split.
• Branded search probably harvests demand that other channels, such as Meta, content, and word of mouth, created. Segment it from non-branded before judging any search spend.
• Offline touches, like sales conversations, events, and referrals, are invisible to MTA. This is where B2B tracking is weakest.
• AI search is a likely contributor to the direct and branded-search inflation. See the ai-seo skill for that.
• Sample size: 35 deals is small. One month of data can swing a lot, so look at trends over several months rather than this month's split.

Recommendation

What to trust

1. The CRM for the deal count (35).
2. Platform numbers only as directional trends, not as counts. If every source shows a channel rising or falling, believe the direction.
3. Self-reported attribution as the tiebreaker where platforms fight over the same deals.

Which model
First-touch and position-based, side by side, on CRM data. Report the first-touch vs. last-touch gap every month.

Budget (moderate confidence, no big reallocations yet)

1. Don't cut anything based on this month's last-touch view. Cutting Meta or other upper-funnel spend because it "converts less" is the classic mistake. Its effect shows up months later, as direct or branded search.
2. Don't scale based on platform-reported conversions. Google and Meta claim 2.5× the real deals between them.
3. Hold the total budget flat for 1–2 quarters while you fix measurement. Shift only small amounts, around 10%, toward channels that look good in both first-touch and position-based views.
4. Judge channels on pipeline and revenue per dollar from the CRM, not on platform conversions. Use cohorts that match the 90-day lag, for example deals closed this month against spend from 90 days ago.
5. Split branded from non-branded search. Cap branded spend if it's mostly capturing demand that already existed.

Fixes to make in the next 30 days

1. Add "How did you first hear about us?" to the demo-request form. Use a pick-list plus free text, and write the answer to the CRM. This is the most direct way to see inside the 45% direct. Discount the shares you get, since recall favors memorable touches.
2. Audit what each platform counts as a conversion. Align everything to a CRM-validated event such as a qualified opportunity or closed-won, and import offline conversions back to Google and Meta.
3. Check UTM discipline and the referrer exclusions (OAuth, checkout, self-referrals) so direct shrinks to what it truly is.
4. Write the first-touch source and a full touch path into the CRM, rolled up to the account, since one B2B deal involves several people.
5. Fix the identify gap so that anonymous browsing is joined to the demo request. The analytics skill and first-party-tracking.md cover this.

Tiebreaker test worth running: a geo holdout or an on/off test on Meta or non-branded paid search. Because the cycle is 90 days, measure lift on pipeline created (qualified opportunities) rather than closed deals. Closed deals would take too long to read. This is what settles whether Meta and Google are both truly driving deals or just claiming them.

Bottom line: 35 is the real number. The other three are different measurements of different things, and the inflated platform counts are normal. Keep the budget stable, run first-touch and position-based together, add a self-reported question, and test incrementality before moving real money.