Quality over volume: what the tool can and cannot do
Once your mail reliably lands in the inbox, the next question is what earns a reply, and here the data demolishes the volume mindset that most tools are tuned to encourage. The lever is relevance, not reach, and understanding why changes how you deploy the AI's leverage entirely.
What the numbers actually say
Across billions of interactions in 2025, the average cold email reply rate was 3.43 per cent. The top quartile of senders cleared 5.5 per cent, and the top decile exceeded 10.7 per cent. That spread, from an average of around three per cent to a top-decile of over ten, is enormous, and it is not explained by who has the fanciest AI. The senders at the top are not writing ten times better sentences than the average; they are reaching the right people with a relevant message. Targeting and relevance, not raw writing horsepower, separate the top from the middle.
The opener carries the result
Here is the statistic that should reorganise your whole approach: 58 per cent of all replies come from the very first email. Most sequences are built on the assumption that persistence wins, that the fourth or fifth follow-up wears the prospect down. The data says the opposite. The bulk of your replies are decided at the opener, by whether that first message is relevant enough to earn a response. This means the marginal hour you spend lengthening a sequence is worth far less than the hour you spend sharpening the first email to a specific person. Sequence length is a minor variable; the opener is the main event.
What the AI can and cannot do
This is where the honest boundary of an AI lead gen tool sits. The AI writes faster than any human, drafts a hundred variations before you have finished your coffee, and personalises at a scale no manual operator could match. What it cannot do is invent relevance. If your ICP is wrong, if you are emailing people who have no reason to care, the AI will write a fluent, well-structured, perfectly personalised message to the wrong person, and it will fail. Garbage ICP in, garbage replies out, regardless of how good the prose reads. The AI is an amplifier, and an amplifier turns up whatever you feed it, including the noise.
The lean allocation rule
Which gives you a clean rule for where to spend the AI's leverage. Spend it on personalisation at the front of the funnel, on making the first message to a narrow, well-chosen ICP genuinely relevant. Never spend it on cranking volume, because volume past the point of relevance does not lift your replies, it lowers your average and threatens your deliverability at the same time. The temptation the tools dangle is to send more; the move that actually works is to send better to fewer, chosen more carefully. A narrow ICP hit with a sharp, personalised opener is the entire game. The AI just lets you play it faster.