Short observations from live campaigns.

Three subject-line tests this fortnight, all in compliance. Same pattern across them: lines that named a specific regulatory event (“On the new ICO consultation”) beat generic ones (“Quick question”) by between 2.6 and 3.4 times on cold reply rate. Sample sizes between 280 and 410 sends per arm.

We've stopped running the generic ones. The cost of testing them is real: every cold send to a compliance director is a small claim on their attention, and “quick question” burns the claim cheaply. The new rule on these lists is, if a subject line wouldn't run as the first line of a feature, we don't test it.

Length matters too. Anything over eight words gets cut by the preview pane on most devices compliance buyers read on first. Short isn't necessarily better, but mid-length anchored to a real event is.

Two campaigns turned down this fortnight, both for the same reason: the buyer wasn't actually who the founder thought.

The first was a fintech consultancy who'd defined their ICP as CFOs at mid-market UK firms. Three calls in we found the actual buyer was the FD's external accountant, and the accountant had no budget authority. We wrote that into the brief and didn't run the campaign. There's nothing to send to the right person at the right firm yet.

The second was a data consultancy whose stated ICP was heads of analytics at 100-to-500-person businesses. Their last six closed deals had all been to heads of operations, two layers below where they thought they were selling. The brief recommended fixing the position before running outbound. Sometimes the work the brief does is point at a different problem.

Three new sending domains this week. The thing nobody tells you about email warm-ups: the best signal isn't volume, it's the spread of conversation patterns. The slowest of the three, talking to thirty existing inboxes a day for ten days, cleared deliverability at 96% on first send-day. The fastest, the one we tried to push to a hundred a day from the off, ended at 78%.

Patience is the actual variable. Skip the warmup and the campaign flames out in week two regardless of how good the messaging is.

Wrote a brief this fortnight that ended with “don't run this”. Founder of a four-person consultancy, average deal £18k, nine-month sales cycle. The deals were too small to clear our economics; the cycle was too long to feed signal back into the messaging fast enough.

Sent the brief, said the honest read aloud, didn't take the work. He said it was the most useful £0 conversation he'd had this year. The version of the model where this happens is the version of the model that works.

Pattern from the last three campaigns in cyber security, data and AI, and compliance.

Cyber: highest cold-reply rate (7-9%), but the “remove me” rate is also highest. Compliance: lowest cold-reply rate (3-4%), but conversation-to-meeting rate is best by a margin. Data and AI: middle on both, highest variance.

Variance is the thing. Across campaigns, results aren't a smooth curve, they're a sigmoid. Don't trust week-one numbers. Don't trust week-two either, if you're honest.

List build update. We refused 47 names this fortnight that the criteria let through. Reasons: recently changed role, on an extended holiday, in a different department in name only, their company just announced an acquisition.

The criteria are necessary; they aren't sufficient. A human still has to look at each one and decide whether the moment is right. The hardest thing to automate in outbound isn't the writing. It's the discrimination of one prospect from another.