Your B2B pipeline has three definitions of a lead, so no channel can be judged

Open the CRM of almost any B2B company and count how many different things are called a lead.

1. Marketing's lead — somebody filled in a form. It is a row with a timestamp and a campaign name.
2. Sales' lead — somebody agreed to a meeting. It is a calendar entry.
3. The CRM's lead — a row whose source field says Web, or is empty, because the form never carried the source and a human retyped it three weeks later.

Three definitions, one report, one argument. And the argument is not really about channels: it is about the
fact that the event which ends the test and the event sales acts on are two different events.

What this does to a channel decision

Take a channel that looks mediocre: 4,000 sends, 120 form fills, eleven meetings booked, two closed. And a
channel that looks great: 900 sends, 61 form fills, one meeting, nothing closed.

The second one wins on every metric marketing can see. The first one wins on every metric sales cares about.
Both teams are reading real numbers from the same database, and neither can switch a channel off, because
"off" needs a number everyone already agreed is the one that decides.

The four definitions that have to be written down

The event. The test ends on the event the sales team already acts on: a meeting held, with the source of
the record attached to it. Opens, clicks and form fills are traffic. Traffic is an input to a test, not the
verdict of one.

The source. Written by the page, not by a person. If a human fills it in later it is a guess, and a guess
in a required field is worse than an empty one, because it looks like data. The labelled url and the matching
event call are written together, at the same time, by the same builder — that is the whole trick.

The reply. A human typing back. Out-of-office messages, "please remove me", bounces and auto-responders
are not replies, and if they count as replies in one arm and not the other, the test compares two different
things.

The stop rule. Written before the first send, with the date on which the number will be read. A channel
that is never allowed to be switched off is not a test, it is a subscription with a monthly invoice.

The arithmetic that usually ends the argument

A strict two-variant test at a 3% base rate and a +20% relative lift needs 13,914 contacts per arm —
27,828 in total — at 80% power and alpha 0.05. That is the standard two-proportion computation, not an
opinion, and you can run it yourself in a browser or from a terminal:

npx marketing-mindset sample --base 0.03 --lift 0.2

At 500 sends a month, filling one arm takes about two years and four months. That number is the actual
answer to most "which channel is better" debates: at that volume, neither channel can be judged, so the
honest options are a cheaper readable unit (a reply instead of a meeting), a higher base rate (a narrower
audience whose problem you obviously solve), or a different channel — not a longer campaign.

Two more numbers that belong before the send, not after:

• Paid: stop or double at 1–3× the target cost per order with no readable return in 48–72 hours.
• Email: 1,500–2,000 sends per variant when the readable unit is the reply, and a sending ramp that starts at 20 a day and grows by half daily to 200 so the domain survives long enough to produce a number.

Why this is engineering work and not a campaign

Every item above is an artifact: an event in an analytics container, a required field in a CRM, a computed
floor, a ramp schedule, a written rule with a date. None of them is a creative decision, and none of them
belongs to the person whose bonus depends on the campaign working.

That is the shape of the job I do: build the measurement first, so a channel can be switched off without a
meeting, then run the queue of experiments on top of it. The artifacts from running exactly this on my own
domain are public — including the free calculators, which do not require handing over an email address:

• Email test planner — the contact floor for your base rate and the lift you hope to see.
• Kill-rule calculator — the stop rule and the read date, in writing.
• Warmup planner — the sending ramp and the four stop conditions that pause it.
• Outbound deliverability checklist — 12 weighted checks, 100 points, the three most valuable fixes shown first.

Where I do this work

B2B, with the measurement built before the spend:

• Sprint $900 — one focused build in five working days: the offer page with schema, five hypotheses with kill rules, one channel live, tracking that makes the result readable.
• Engine $1,900/month — a running queue of pages, tools and distribution with a monthly read-out.
• Full Build $2,900 — site architecture, the full page set, the AI-readable layer, tooling, distribution and handover.

No invented case studies and no benchmark numbers I did not compute. Scope and prices:
marketing engineer for B2B.

The one-line version

Write down which event ends the test, make the page fill in the source, compute the floor before the first
send, and give the channel a date on which it is allowed to be switched off. Everything else in B2B
marketing is a discussion about definitions dressed up as a discussion about performance.