Acceptance held at 28.5%, the connection note collapsed 37% — the model moves the persuasion elsewhere

The published LinkedIn dataset is unusually large and unusually blunt. Acceptance sits at 28.5% across 13.2 million requests and has held. The message that follows gets a reply 10.4% of the time. The note attached to the connection request gets 3.0%, down from 3.5% a year earlier — a 37% fall in twelve months. Computer software senders underperform the platform on both acceptance and reply, which is unsurprising given that software buyers receive more of this mail than anyone else on the network. Meanwhile the paid route to the same audience prices a lead at $193 through lead gen forms and $346 through landing pages.

Nearly three in ten people accept the request. Roughly one in ten of those reply to the message. The constrained step is not the list.

1. 01

   Acceptance is not the constrained step, so optimising the list is optimising the wrong thing

   28.5% accept and 10.4% of follow-up messages get a reply. Targeting improvements move the first number, which was never the problem. The model spends its effort on the first message after acceptance, which is where roughly nine in ten conversations end.

2. 02

   The connection note is a declining asset and should be treated as one

   3.0% and falling 37% a year is a channel closing, not a channel underperforming. Whatever the note used to do has to move to the profile, because that is what a recipient actually opens before deciding.

3. 03

   Software senders are below average on a network saturated with software outreach

   27.5% acceptance and 8.8% reply for computer software against 28.5% and 10.4% platform-wide. The implication for copy is counterintuitive: a software sender's message needs to sound less like good software marketing, not more like it.

4. 04

   The denominator decides the number, and most published rates do not state theirs

   The same programme can honestly report a 27.6% reply rate against accepted connections or 7.47% against requests sent. Any model that quotes one figure without its denominator is reporting an opinion. This model reports both at every stage.

5. 05

   The published funnel matrix says LinkedIn's weakness is qualification, not closing

   In the channel-by-stage benchmark LinkedIn has the strongest visitor-to-lead rate of the five channels at 2.2% and the weakest MQL-to-SQL rate at 30%, while closing at 39%. LinkedIn produces conversations that qualify badly and close well, which is an argument for a qualification step before the meeting rather than for more sends.

Acceptance **28.5%** · follow-up reply **10.4%** · connection-note reply **fell 37%** in 12 months (3.5% → 2.2%) · across **13,218,869** requests. Software senders sit below platform average on both.

The model front-loads the two things the benchmarks say actually move: the profile, because it is what gets opened before the message is read, and the first message after acceptance, because that is where the conversation is lost. Connection requests are sent without a note or with a note that does no selling. The first message carries one observation and one question and no link. A qualification step sits between the reply and the meeting, because the published matrix says LinkedIn-sourced leads leak at exactly that stage. Paid runs alongside as a cost comparison rather than as a channel: at $193 per lead through lead gen forms, paid gives the outbound programme a price to beat on the same audience.

1. 1

   Weeks 1–2

   Audit and denominators

   Current rates recalculated against both denominators; profile audited per sender

   Owner: OmniFlow

2. 2

   Weeks 2–4

   Profile and first-message rebuild

   Sender profiles rebuilt as landing pages; one observation, one question, no link

   Owner: OmniFlow

3. 3

   Weeks 3–10

   Send and log

   Paced manual sending with a per-session log; note used for nothing that matters

   Owner: Sender + OmniFlow

4. 4

   Weeks 5–12

   Qualification step

   A defined qualification gate between reply and meeting, with a written disqualifier list

   Owner: OmniFlow + client sales

5. 5

   Week 12

   Review

   Both denominators reported; cost per qualified conversation against the paid benchmark

   Owner: OmniFlow

At the published rates, 1,000 connection requests produce roughly 285 accepted connections and roughly 30 replies to the follow-up message, and a software sender should model the lower end of that. The model reports cost per qualified conversation next to the published paid cost per lead of $193, because a manual outbound programme that cannot beat a lead gen form on the same audience is a programme that should be a lead gen form. Every published rate here is replaced by the sender's own from week three.

Modelled. Inputs: Expandi, LinkedIn Outreach Benchmarks 2026 (13,218,869 connection requests); Metadata, 2026 B2B Paid Media Benchmark ($57.6M ad spend, 153 advertisers, 211,000 leads); First Page Sage, B2B SaaS Funnel Conversion Benchmarks (June 2025); 6sense, B2B Buyer Experience Report 2025 (~4,000 respondents). Modelled outputs are not a forecast or a guarantee of results. Cited: Expandi *LinkedIn Outreach Benchmarks 2026* (13.2M requests, May 2025–Apr 2026).

Every figure below is published research, not a client result. They are the inputs to the arithmetic above, listed so it can be checked rather than taken on trust. The bracketed number points to the full citation at the end of this page.

These are the surfaces this engagement is run and measured from, shown with representative figures built around the benchmarks cited on this page. Every account we run reports into views like these, and you keep ownership of all of them.

Each figure on this page, the system it is read from, and the definition and window it is measured over.

Not applicable — this is a modelled engagement. Its weakest input is that the acceptance and reply benchmarks are drawn from accounts using an automation tool, which is a self-selecting population: they send far more, target more loosely, and look nothing like a hand-worked list of two hundred people. The rates should be read as a floor rather than as a target, and the model deliberately does not promise to beat them, only to report against them honestly.

A modelled capability, not a client account. Figures illustrate what the model produces and are labelled as modelled wherever they appear. See all our work

Full citations for everything cited on this page, with the sample and period each study covers, so you can go and read the original.