Campaign review · Jun to Aug 2026
Campaign review · Jun to Aug 2026
Thirty-two sends to 3,333 nonprofits running Salesforce, over eight weeks. What the final numbers show, what they cannot show, and what the next campaign should be.
Nonprofits already running Salesforce, dealt round-robin by organization into eight groups so no organization sat entirely on one side of the test. Odd groups got a plain email from a named person. Even groups got the branded Civis version. Same subject line, same message, same day.
Three reasons to look past the CRM, then storage cost, then analytics and AI, then putting AI in the hands of the whole team. Bodies ran 213 to 300 words.
Plain against branded. Subject, copy, send time and audience construction were held constant. Subject lines were never tested.
Read the overview, request a demo, or reply. The demo request was the conversion event we instrumented.
Counted as people, because people are what a campaign is meant to produce. Every figure is as of August 20.1
Scroll the chart sideways to read it.
Zero. No verified recipient clicked the case study, a reason card, or the demo button on the landing page either.
97.6% delivered and zero spam complaints across 13,512 messages. Reaching the inbox was never the problem.
Strip it out and 217 people out of 4,002 clicked anything on purpose. The headline click number is roughly double the human one.
With the one group whose numbers are a security-scanner artifact4 excluded, the plain lane leads by 1.6 points of open rate and by a tenth of a point on clicks. Across nine clean same-day pairs, plain won the open seven times and the click four times.
| Persona5 | Plain, open | Branded, open | Plain, click | Branded, click |
|---|---|---|---|---|
| Leadership | 34.8% | 37.8% | 4.3% | 5.2% |
| Technical | 32.5% | 27.8% | 5.3% | 5.3% |
| Fundraising and marketing | 32.2% | 32.9% | 3.8% | 3.8% |
The two lanes changed the sender name as well as the design, a named person against the company. Open rate is decided by the sender and the subject, before any design is visible, so the 1.6 points may be about who the mail came from. This campaign cannot separate the two.
On the fourth email, branded won both metrics, 19.8% against 19.2% on opens and 3.05% against 2.27% on clicks. That message went out on August 19 and its numbers are still filling in.
By touch order. After the first email the line is flat, inside two points, for three more messages and six more weeks. Clicks per open recovered on the last one.
Orange marks the highest bar in each chart.
The same first message, same design, unchanged, returned 29.3% open in late June and 16.8% in mid-July. The day we sent moved the number more than the message did or the design did, so some of the fall above is the calendar.
Three extra emails bought no lift in opens, and the last message earned the most clicks per open of any in the campaign. It was also the newest argument, which points to fewer, newer and shorter messages.
None of the three is the copy.
Book a demo, to a cold audience that could not name the company, 13,844 times, zero conversions. Two thirds of B2B buyers now say they would rather buy without talking to anyone at all.7
Four messages of 213 to 300 words across up to six weeks. Three of them bought no incremental engagement, and the last one, on the shortest runway, performed best per open.
One lane asked people to reply and we cannot count the replies. The one substantive reply we know about surfaced by accident, weeks later, while looking for something else.
No reply count exists for either arm. One arm's replies land in an inbox nobody labels, the other in an inbox our reporting cannot read. If people were interested, this is where the evidence went.
The first two sends went out before per-person tracking existed. More than half the visits to the landing page across the whole campaign cannot be attributed to anybody, and that cannot be fixed after the fact.
Apple accounts for 62% of all email opens and its privacy feature affects 55 to 60% of them, so a large share of any open rate is a machine fetching a pixel.6 Ours also carries security scanners: 56 accounts hold 57% of the clicks.
The landing page opened its own form on arrival, which manufactured 30 form opens that nobody chose. None of them became a request.
Published cold-email benchmarks are vendor marketing with no disclosed methodology, so nothing from that category is on this page. These four have a named sample.
67% of B2B buyers prefer a buying experience with no sales rep in it, up from 61% a year earlier.7 A demo request is a late-funnel action and we asked for it first.
Across 40 million messages, replies peak between 50 and 125 words and decline from there. Ours ran 213 to 300.8
Apple is 62% of opens and its privacy feature affects 55 to 60% of them.6 Any test decided on open rate alone is decided partly by machines.
Across nearly 1,300 nonprofits, technology staffing is the strongest predictor of digital readiness, and the average organization has about one data person.9 That is the argument our last message made, and the one it should have opened with.
Working title Short List. Run a short-form campaign to the people this one already warmed up, then to fresh cold audience in the same format.
Slim banner, one accent, body under three sentences, one link. Keep the named human sender, which is the variable most likely to have earned the open. Drop the unbranded wrapper, which earned nothing after it.
Open with the 127 people worth a call and the 772 who engaged with two or more messages, then extend to fresh cold audience in the same format. Every standing exclusion and every mailability flag is checked when the list is built.
Design is settled and subjects were never tried. Short form makes subject testing cheap, because only the envelope changes between the arms. Read the result on clicks and replies, not on opens.
Who it targets, the emails rendered as they would arrive, the three-week cadence, and the funnel we would instrument. Two minutes of scrolling.
The walkthrough is a mock. Nothing in it is staged to send, and it carries no projected numbers.