Methodology
Three numbers appear across multiple posts on this site: top 2 of 18 referring domains accounted for ~64% of trial revenue last quarter; Domain Rating explained ~12% of referral CVR variance; topical-relevance scoring tracked closer to ~58%. These power the Backlink RPV Scorecard and the deeper revenue-per-referring-domain argument. This page explains the sample, the join, the regression, and where it does not apply.
Every backlink-evaluation tool on the market scores referring domains on the same primary axis: Domain Rating (Ahrefs), Domain Authority (Moz), or Authority Score (Semrush). These are link-graph proxies for how much PageRank-flavoured authority a domain has accumulated. They tell you whether a link is theoretically valuable for rankings.
They do not tell you whether the link sends a buyer.
We wanted to answer a simple operator question:
Among the referring domains pointing at our own marketing site, which ones actually delivered visitors who turned into paying Stripe customers, and how strongly does Domain Rating predict that outcome compared to alternative scoring inputs?
The end goal was a scoring rubric that an operator can apply to any referring domain to decide whether it's worth pursuing, monitoring, or ignoring — not in terms of authority, but in terms of expected revenue per visitor (RPV).
We ran the measurement on the attrifast.com marketing site and one sister bootstrapped SaaS we operate, across the same 16-week window from January 6, 2026 through April 26, 2026 used for the return-delay-penalty methodology. Eligibility for inclusion in the sample:
Referer header or a UTM utm_source pair (filters out direct/(none))18 referring domains met all three criteria in the window. They split roughly as:
Combined, those 18 RDs sent ~1,950 sessions and produced 57 Stripe checkout completions across the two sites in the window. The sample is small. The numbers below are directional, not statistically tight. A site with 100x our referral traffic should run its own measurement; we publish the methodology so others can.
On every landing, the server-side handler captured the Referer header (or any present UTM source-medium pair), normalized the referring domain (stripped subdomain on most, kept it for known multi-property hosts like news.ycombinator.com vs ycombinator.com), and wrote both the raw referrer and the normalized RD into a session row keyed by a first-party session ID.
On Stripe's checkout.session.completed webhook, the server joined the Stripe customer back to the session row via metadata stamped at checkout-link creation time, recording the originating RD on the customer row.
The pseudo-SQL of the join, simplified:
SELECT
s.referring_domain AS rd,
COUNT(DISTINCT s.session_id) AS sessions,
COUNT(DISTINCT c.customer_id) AS converters,
SUM(c.amount_total) / 100.0 AS gross_revenue_usd,
COUNT(DISTINCT c.customer_id) * 1.0 / COUNT(DISTINCT s.session_id)
AS conversion_rate,
SUM(c.amount_total) / 100.0 / COUNT(DISTINCT s.session_id)
AS revenue_per_visitor
FROM sessions s
LEFT JOIN stripe_checkout_completed c
ON c.session_id = s.session_id
AND c.status = 'complete'
WHERE s.referring_domain IS NOT NULL
AND s.created_at BETWEEN '2026-01-06' AND '2026-04-26'
GROUP BY rd
HAVING sessions >= 10 AND converters >= 1
ORDER BY revenue_per_visitor DESC;revenue_per_visitor (RPV) is the headline metric. It folds traffic volume and conversion probability into a single comparable number across RDs of very different sizes.
We report three derived numbers in posts that cite this page. Each is a summary of what the 18-RD table showed.
Sorting the 18 RDs by gross revenue contribution:
| Rank | RD category | Share of gross revenue | Cumulative share |
|---|---|---|---|
| 1 | Niche SaaS community | 38% | 38% |
| 2 | Industry-blog category | 26% | 64% |
| 3 | Newsletter show-note | 9% | 73% |
| 4–5 | Mixed | 11% | 84% |
| 6–18 | Mixed (long tail) | 16% | 100% |
The top 2 RDs combined drove ~64% of gross revenuefrom this cohort. That's the "directional read of last quarter suggests the top 2 accounted for on the order of two-thirds of trial revenue" phrasing used in the linked blog posts.
We fit a simple univariate linear regression of conversion rate (CVR) on Domain Rating across the 18 RDs:
That's the ~12% of variance explainedfigure. The relationship is positive but weak. Most of the high-DR aggregators in the sample (DR 72-89) had middle-of-the- road or below-median CVRs because their audiences were broad and off-topic. The R-squared is reported with caveats: n=18 and the regression is not significant.
We manually scored each RD on a 0-3 topical-relevance scale against attrifast's ICP (Stripe-based SaaS founders, revenue attribution, growth analytics):
Fitting CVR on this topical-relevance score:
That's the ~58% of variance explainedfigure. Topical relevance was a much stronger CVR predictor than DR in this sample, by roughly 5x in R-squared terms.This is what licenses the "score links on relevance, not authority" argument in the blog posts.
For completeness, the correlation between each scoring input and CVR:
| Input | r | R-squared | p-value (n=18) | Significant? |
|---|---|---|---|---|
| Domain Rating (DR) | 0.35 | 0.122 | 0.16 | No |
| Topical relevance (0-3) | 0.76 | 0.578 | 0.0002 | Yes |
| Audience overlap with ICP (0-2) | 0.71 | 0.504 | 0.0009 | Yes |
| Page-level traffic (log10) | 0.22 | 0.048 | 0.38 | No |
| Anchor naturalness (0-1) | 0.41 | 0.168 | 0.09 | Marginal |
Both significant inputs (topical relevance and ICP audience overlap) are content-quality signals, not link-graph signals. Both link-graph signals (DR, page traffic) failed to reach significance.
The Backlink RPV Scorecard in the "backlinks with seo" post is a 5-axis 0-10 rubric: Topical Relevance (0-3), Audience Overlap (0-2), Page-Level Traffic (0-1), Editorial Integrity (0-1), Revenue Intent (0-1), plus Domain Authority as a tiebreaker (0-2).
The point weights are not arbitrary. They roughly track the R-squared values above: topical relevance gets the most weight because it predicts the most variance; DR gets the least because its predictive power is weak. The cutoff thresholds (pursue at 7+, monitor 4-6, ignore under 4) were calibrated by back-fitting against the 18-RD sample — domains that hit 7+ in the rubric are the same ones that sit in the top quartile by RPV.
If you have a similar tracking setup (server-side referrer capture + Stripe webhook join), the SQL above adapts in roughly 30 lines. The two manual scoring inputs (topical relevance, audience overlap) take ~30 seconds per RD to score; budget 10-15 minutes for an 18-RD audit.
Two pieces of advice from running it ourselves:
If you don't have server-side referrer capture wired to Stripe yet, that's what Attrifast's revenue attribution view does out of the box.
If you want to run this on your own data and you already have server-side referrer capture plus Stripe webhooks, the SQL above is the bulk of the work. Email vincent@attrifast.com with "rpv methodology" in the subject and I'll share our actual SQL plus the BigQuery schema and the topical-relevance rubric template. Faster path: connect Stripe via Attrifast's revenue attribution and let the same join run automatically.
2026-05-11: Initial publication. Next planned re-measurement is December 2026 with a larger sample (target 50+ RDs) once cumulative referral traffic supports it. A v2 of the topical-relevance score using embedding similarity is planned for Q3 2026.