Our operational database currently holds 3,215 LinkedIn conversations and just over 5,000 individual messages, across our own pipeline and client campaigns. That is a small dataset by SaaS-benchmark standards and an enormous one by "we actually read these conversations" standards. Here is what it says, with the caveats left in.
Finding one: the niche premium is real and it is huge
The single largest performance gap in our data is not messaging, timing or profile quality. It is how narrow the audience is.
| Audience | Acceptance rate | Reply rate |
|---|---|---|
| Quantity surveyors, UK construction | 57% | 29% |
| Marketing managers, broad UK | 31% | early data |
| Recruitment owners, broad UK | 30% | 19% |
The construction campaign outperforms everything else we run, and it is not close. Two reasons, we think. Nobody else is running thoughtful outbound to quantity surveyors, so there is no fatigue; a marketing manager's inbox is a warzone by comparison. And a narrow niche lets the first message be genuinely specific, because the sender demonstrably understands the recipient's world.
The practical rule: if you can choose between a 2,000-prospect broad list and a 400-prospect narrow one, take the narrow one every time. The broad list feels safer and performs worse.
Finding two: acceptance is a list metric, replies are a message metric
When acceptance is low, the problem is nearly always upstream of the copy: wrong people, dormant profiles, or a sender profile that does not look credible to that audience. When acceptance is healthy but replies lag, the first message is the suspect. Diagnosing the right layer saves weeks. We watch the two numbers separately for exactly this reason, and the dashboard we give clients reports them separately too.
Finding three: the reply cliff is brutal and merciful
Most conversations that die, die instantly: the connection is accepted and the first message never gets an answer. This sounds bleak and is actually the merciful version, because silence costs nothing but the request. What our data says loudly is that a reply, any reply, changes the odds completely. Of the conversations that got a first response in our pipeline, a large fraction progressed to a real exchange, and 161 are currently tagged interested with 43 meetings booked all-time. The lesson: engineering the first reply is worth more than perfecting message four. Ask something the prospect can answer from their chair, about their world, with no link and no pitch.
Finding four: hostility is rare, and handling it is the whole game
Across thousands of messages, genuinely hostile responses are a rounding error, a handful in total. But each one matters more than a hundred silences, because outbound in a niche is outbound in a village: people talk. Our system hard-stops the moment a prospect signals annoyance: the conversation is paused permanently, no follow-up fires, and a human reviews it same-day. If you are evaluating any AI outreach tool, ask one question first: what happens when someone says stop? The vendors with a bad answer are the reason the whole category has a reputation problem.
Finding five: the platform sets the ceiling, respect it
LinkedIn allows roughly 150 connection requests per account per week, and it moves the throttle around without notice. One of our own campaigns sits rate-limited as we write this. That is not a failure of the system, it is the operating environment, and any architecture that does not assume throttling will eventually convert a marketing problem into an account problem. The volume ceiling is also why everything else in this article matters: when you only get 150 shots a week, list quality and first-message quality are not optimisations. They are the campaign.
All of this runs approval-first on your own account in AI Sales, £499/month + VAT, and we owe you 1 client in 90 days or it's free until you get one. Want the field manual instead? The free skills behind it are on the tools page.