Cold email works when it carries a number

Almost every cold email an agency sends says a version of the same thing: we noticed your website could be improved, we help businesses like yours, do you have fifteen minutes on Thursday. It is polite, it is generic, and it is indistinguishable from the other nine that arrived that week.

The alternative is not a better subject line. It is having something specific to say.

A number changes what the email is

Compare two openings. The first: "your site could load faster". The second: "your homepage takes 4.8 seconds to become usable on a mid-range Android phone. Google's threshold is 2.5, and the businesses ranking above you in Maps average 2.1".

The second is not a better-written version of the first. It is a different category of message. It proves the sender looked, it is checkable, and it hands the recipient a fact they can act on whether or not they ever reply. That last property is what makes it feel like help rather than an approach.

The measurement has to be true, and it has to be theirs

This only works if the number survives contact with the recipient's own developer. Which means: measured, not inferred; current, not cached from three months ago; and about their site specifically, not an industry statistic dressed up as a personal observation.

It is also why we let the code choose the angle rather than the model. The findings are ranked by severity and sellability before anything is written, and the drafting step only gets to phrase the case around a measurement that has already been selected. A model asked to be persuasive from a pile of raw audit data will reliably produce something persuasive and slightly false, and slightly false is fatal here.

Lead with the finding, not the offer

The pattern that works, in three sentences: here is the specific thing I measured, here is what it is costing you in terms you care about, here is the report if you want the detail. No pitch deck, no calendar link in the first message, no paragraph about our process.

The report link does the selling. It is a full diagnostic with their name on it, and it was free to produce. A meaningful share of the replies we see are not "yes let's talk" but "thanks, we had no idea about the certificate", and that is a conversation, which is all a first email should be trying to start.

Volume is the thing that kills it

The temptation, once drafting is automated, is to send more. Resist it: the moment the messages outrun the evidence, they collapse back into the generic email you started with, and you have taken your domain reputation along for the ride.

We built the approval gate for exactly this reason. Reading each draft before it goes is a small tax that keeps the whole approach honest. And if reading them is too slow, the answer is to send fewer, better ones, not to remove the reading.