Benchmarks as Priors — Funnel Reference Data for Small-N Decisions¶

For factory operators. A factory running 2–3 experiments/week rarely has enough data for a stable estimate. Benchmarks are how you avoid starting from zero — but only if you use them as priors, not targets. This document gives the verified funnel reference data, the method for turning a benchmark into a prior that shrinks thin estimates, and the stage-based budget allocation that priors inform. It is the data layer beneath the experiment-loop.

The thesis: a benchmark is a distribution, not a goal line. "Average visitor-to-lead is 1.4%" doesn't mean hit 1.4%; it means before you have data, 1.4% is your best guess, and your noisy 3% from 100 visitors should be pulled partway back toward it.

1. The Problem Benchmarks Solve¶

At small N, raw rates are wildly unstable: 3 leads from 100 visitors reads as "3%," but the 95% interval spans roughly 1–8%. Acting on the point estimate over-reacts to noise. A benchmark gives an informed starting belief so each thin experiment borrows strength from the industry base rate — the same empirical-Bayes shrinkage the bayesian-decision-rule uses, with the benchmark supplying the prior.

2. Benchmarks as Bayesian Priors (the method)¶

Encode a benchmark rate p₀ as a Beta prior with a chosen strength k₀ (in pseudo-observations): α₀ = p₀·k₀, β₀ = (1−p₀)·k₀. Observe s successes in n trials → posterior Beta(α₀+s, β₀+n−s). The posterior mean is a weighted blend of benchmark and data; when n ≪ k₀ the benchmark dominates, and as n grows the data takes over.

Worked example (computed for this KB; benchmark p₀=1.4% visitor→lead, strength k₀=200):

Observed	Naive rate	Posterior mean (shrunk)
3 / 100	3.0%	1.93%
30 / 1,000	3.0%	2.73%

Same 3% signal, but at 100 visitors it shrinks hard toward the 1.4% prior; at 1,000 it's allowed to move most of the way to 3%. k₀ is the operator's dial — how much you trust the benchmark vs. this specific page. Set it from how well the benchmark's segment matches yours.

3. When a Benchmark Is a Valid Prior — vs. a Trap¶

A benchmark is only a valid prior if it matches your situation. Traps:
- Segment mismatch — a $10M–$100M-ARR mid-market benchmark is a bad prior for a pre-PMF startup or PLG self-serve motion.
- Definition drift — everyone defines "MQL" differently; a Lead→MQL benchmark is only meaningful against a comparable definition.
- Point-target fallacy — treating the average as a goal invites gaming the metric instead of the outcome (cf. marketing-attribution's caution on optimizing measurable proxies).
- Survivorship — published benchmarks skew toward companies successful enough to report.

The discipline: match segment/stage/channel, treat the benchmark as a distribution, and weaken k₀ when the match is loose.

4. Verified B2B SaaS Funnel Benchmarks¶

Average six-step funnel (First Page Sage, 50+ B2B SaaS clients over a decade, $10M–$100M revenue focus — fetched & verified):

Stage	Average
Visitor → Lead	1.4%
Lead → MQL	41%
MQL → SQL	39%
SQL → Opportunity	40%
Opportunity → Close	36%

By channel (visitor→lead / MQL→SQL): SEO 2.1% / 51% · LinkedIn 2.2% / 30% · Email 1.3% / 46% · Webinar 0.9% / 39% · PPC 0.7% / 26%. SEO and email convert deeper in the funnel; PPC fills top-of-funnel but qualifies worse.

(Trial-to-paid is not in this source; other aggregators report ~8–25% depending on product complexity — treat as secondary/unverified until fetched.)

5. Budget Allocation by Stage¶

Benchmarks and iROAS together set the spend mix. Allocation is stage-dependent (see gtm-archetype and the growth-stage maps): early-stage concentrates on a few founder/content/PLG channels; growth-stage diversifies into paid + ABM + events; scale-stage runs the full mix with NRR as the primary metric. The cross-cutting rule is causal, not benchmark-bound: set the mix from stage norms, then move money toward the highest marginal incremental-roas until it equalizes or hits the ROAS floor. Benchmarks tell you where to start; incrementality tells you where to move.

6. Factory Integration¶

Benchmarks are the prior-supplier for the experiment-loop: they seed the bayesian-decision-rule's priors so thin tests don't over-react, and they inform the minimum-detectable-effect gate (knowing the base rate sets the effect size worth testing). The flow is: benchmark sets the prior → the live experiment updates it → incrementality-testing sets the causal iROAS → the loop decides. All agent-ownable: maintain a benchmark table by segment/stage/channel, attach a strength k₀, and apply shrinkage automatically before any small-N call.

Provenance Note¶

Primary-sourced, fetched & verified: the six-step funnel averages and the per-channel conversion table (First Page Sage, B2B SaaS Funnel Conversion Benchmarks, fetched).
Computed for this KB (reproducible): the Beta-prior shrinkage table in §2 (Beta(p₀·k₀, (1−p₀)·k₀) updated by observed s/n).
Secondary — not fetched: trial-to-paid ranges (other aggregators); flagged as unverified.

Sources¶

First Page Sage — B2B SaaS Funnel Conversion Benchmarks
OpenView / KeyBanc SaaS benchmark reports (CAC payback, NRR, Rule of 40) — secondary references for stage benchmarks

Concepts¶

Extracted from this source: marketing-benchmark-prior · marketing-budget-allocation