Methodology

The rules in one paragraph

Every benchmark on ABMBenchmarks.com links to a public page, names its publisher and year, states whether it was measured or surveyed, and carries a description of the sample it came from. We never adjust a published figure. When we derive a number (for example, a cost per opportunity from a cost per lead and a conversion rate), the arithmetic is printed next to the result. The site is sponsored by Metadata.io, whose 2026 B2B Benchmark Report is one of the reports we index, and that report is held to the same rules as everyone else's. This page spells out each rule so you can check our work.

Sourcing rules

A figure is eligible for this site only if a reader can load the source page without paying. That is the first and least negotiable rule. Lead-gated reports are tolerated because the reader can still get the document; paywalled analyst subscriptions are not, because we cannot link to them and you cannot check them.

Beyond that, four things must be true before a number goes in a table:

1. Named publisher and year. "Industry data suggests" is not a source. Metadata 2026, Demandbase 2024, WordStream 2023 are sources.
2. Defined metric. The report has to say what it means by "lead," "opportunity" or "conversion." If it does not, the row is marked undefined and the figure is excluded from any derived calculation.
3. Described sample. The report must say something about who is in the data: dollars of spend analyzed, number of respondents, number of accounts, or number of customers. "Thousands of campaigns" without a count does not pass.
4. Primary, not secondhand. We cite the report that produced the figure, not a blog post that quoted it. If we can only find the secondhand version, we say so in the row note and downgrade the row's confidence label.

We do not use vendor sales decks, webinar slides, screenshots posted to social media, or numbers relayed to us in conversation. If a vendor wants a figure included, the path is to publish it on a public page and send us the link.

Survey vs measured: how every figure is labeled

Every benchmark table on this site carries a "basis" column with one of three labels, because a cost per lead reported from memory in a survey and a cost per lead pulled from ad-platform billing are different kinds of evidence.

The label is not a quality score. Surveys are the right instrument for "what share of B2B teams run ABM," and measured data cannot answer that. But for anything with a dollar sign, measured beats surveyed, and when both exist for the same metric the measured figure is listed first and the survey figure is listed second with a note about the gap.

How figures are normalized across reports

We do not change any published number. Normalization on this site means making the presentation consistent, so that figures with different definitions can sit in the same table without one of them quietly misleading you. Here is what that involves.

Units are stated on every row

A LinkedIn Lead Gen Form submission, a gated-content download, a demo request and a CRM-qualified MQL are all called "leads" somewhere. Each row names the unit the publisher used. When two rows in one table use different units, the caption says so, and the rows are grouped by unit rather than sorted by value.

Ranges, medians and means stay as published

If a report gives a range, we print the range. If it gives a median and a mean, we print both, labeled. We never take a midpoint of a range and present it as a point estimate. Where one report publishes a median and another a mean for the same metric, the table notes that the two are not directly comparable, because paid-media cost distributions have long right tails and the mean is usually higher.

Time periods are attached

Each figure carries the period it describes (calendar year, trailing twelve months, a single quarter). Reports that do not state a period are labeled with the publication year and flagged "period not stated."

Currency and geography

All figures are printed in the currency the publisher used, which is US dollars in every report currently indexed. We do not convert. Where a report is global, its geography column says "global"; where it is US-only, it says so. Mixing global CPC data with US-only CPC data is a common way benchmark tables mislead, so the geography column is never blank.

Disagreements are shown side by side

When two reports disagree, we print both and add a one-line reason for the likely gap. The most common reasons, in order, are: measured vs surveyed; a different lead definition; a different sample (venture-backed software companies vs a general B2B population); and a different time period. We do not pick a winner. The guide explains how to decide which sample is closer to your own company.

Derived figures: the arithmetic is always shown

Some pages on this site present numbers no report published directly, built from figures that were published. Every derived figure is marked "derived" in its row, and the calculation appears in the row note or immediately beneath the table. The rules for derivation are strict.

• Inputs must come from the same report. We never divide one publisher's CPL by another publisher's conversion rate; the samples are different and the result would be meaningless.
• Inputs must share a unit and period.
• Only one arithmetic step is allowed. A cost per opportunity from CPL and lead-to-opportunity rate is acceptable; a cost per closed-won deal from CPL, lead-to-opportunity and opportunity-to-close rates is two steps, compounds two rounded inputs, and is not published.
• Inputs are shown at the precision the publisher used. If a report says "roughly 10%," the derived figure is labeled approximate.

Worked example, using placeholder values so the arithmetic is clear: if a report states a measured cost per lead of $200 and a lead-to-opportunity rate of 10% for the same sample and period, the derived cost per opportunity is $200 ÷ 0.10 = $2,000. The row would read: Cost per opportunity, derived: $200 CPL ÷ 10% lead-to-opportunity = $2,000 (same report, same period). If either input is missing, the cell says "not derivable" rather than borrowing an input from elsewhere.

Derived figures never appear in page titles, hero text or the statistics roundup. They live only in tables where the inputs are visible.

What we count and what we exclude

The sponsor's report under these rules

Metadata.io's 2026 report passes the sourcing tests because it is public, attributed, defines its units (closed-won revenue rather than leads) and describes its sample ($57.6 million of B2B ad spend across its customer base) (1, 2). It is labeled measured. It also carries a sample-bias note on every page where its numbers appear: the customer base skews toward venture-backed B2B software companies running paid social and paid search through one platform, which is not the same thing as "B2B." Those two facts, the pass and the caveat, are presented together every time. The full report summary uses the same template as the Demandbase and 6sense summaries on the reports index.

Update cadence

The date at the top of every page is the date of the last review, not the date the page was first published. Reviews happen on a schedule that depends on how fast the underlying data moves.

• Report summaries: updated within 30 days of a new edition. Metadata's report is annual; Demandbase's has been roughly annual; Demand Gen Report publishes several surveys per year.
• Channel pages such as LinkedIn Ads benchmarks: reviewed quarterly, because platform costs move faster than reports.
• Industry and roundup pages such as CPL by industry and create vs capture: reviewed quarterly, re-dated only when a figure changes.
• Retired figures: when a publisher releases a new edition, the old figure is replaced, and the previous value is kept in a "prior edition" note for one cycle so year-over-year movement is visible.

A page not reviewed within twelve months is either refreshed or removed. Stale benchmarks are worse than no benchmarks, because people plan budgets on them.

Corrections policy

Anyone can request a correction: a publisher, a vendor that competes with our sponsor, an analyst, a reader. Send the page URL, the figure in question and a link to the public source you believe is correct, using the contact details on the about page. We check the public source. If it supports the change, we make it, usually within a week, and add a dated correction note at the bottom of the page stating what changed and why. We do not silently edit numbers. If a publisher revises its own report after we summarized it, we update to the revised figure and note the revision date. If a correction request is not supported by a public source, we reply explaining what would be needed, and the page stays as it is.