Key finding
Our analysis of 82 same-workspace consecutive-month pairs of LinkedIn outreach found the median month-over-month acceptance-rate swing was 5.69 points, 54.9% of month pairs swung 5 points or more, and 31.7% swung 10 points or more.
What we measured
Every operator has lived this meeting: acceptance fell from 29% to 24%, someone declares the messaging broken, and a rewrite is commissioned. The unexamined assumption is that a five-point move must have a cause you control.
This study measures the assumption. We took every workspace in the dataset with substantial sending in consecutive calendar months and asked: when nothing about the measurement changes, how much does a team's own acceptance rate move from one month to the next, as a matter of course?
Methodology
Dataset window. Connection requests sent between 8 January 2025 and 30 June 2026, pulled 27 August 2026, restricted to requests matured at least 30 days and to complete calendar months. That is 18 full months inside the 196,696-request matured cohort.
The unit is a same-workspace month pair. For each workspace and month with at least 300 matured requests, we computed the acceptance rate. Every case where the same workspace also cleared 300 requests in the next calendar month forms one pair, and the statistic is the absolute difference in acceptance rate across the pair. There are 82 such pairs. Comparing a workspace with itself is the point: cross-workspace differences never enter the number.
Threshold robustness. At 300 requests per month, a true rate of 27% has a sampling standard error of about 2.6 points, so part of any swing is sampling noise. Raising the bar to 1,000 requests per month (44 pairs) shrinks sampling error below 1.5 points; the measured swing does not shrink with it (the median edges up to 6.13), which is how we know the movement is mostly real variation, not counting error.
Provenance. This is an analysis over platform data, not a platform measurement: the pairing rules and thresholds are ours.
The findings
Absolute month-over-month acceptance swing, same workspace, 82 pairs at the 300-request threshold:
| Statistic | Swing |
|---|---|
| 25th percentile | 2.32 points |
| Median | 5.69 points |
| 75th percentile | 11.15 points |
| 90th percentile | 16.61 points |
54.88% of month pairs swung 5 points or more. 31.71% swung 10 points or more.
At the 1,000-request threshold (44 pairs): median 6.13 points, 90th percentile 14.84, with 61.36% of pairs at 5 points or more. The swing survives the stricter cut essentially intact.
For scale, the platform-wide monthly acceptance rate across the 18 full months ranged from 14.52% (June 2025) to 34.38% (March 2025). The monthly aggregate itself moves through a 20-point range across a year and a half.
What this means
A five-point move in monthly acceptance is not a signal. It is the median experience of a team that changed nothing about how the metric is computed. One month pair in three moved by ten points or more.
This has a corollary for every timing and tactics claim in the genre, including ours. The entire day-of-week effect in this dataset spans 1.86 points, a third of the median monthly swing; any tactic whose claimed effect is smaller than the noise floor cannot be validated from a single team's dashboard, ever. The differences that do clear the floor in this dataset are compositional: target industry spans 11.13 points and target company size spans 7 points.
Two honest boundaries on the claim. First, the data cannot separate ambient drift from a workspace's own changes: teams edit targeting and templates month to month, and some of the measured swing is those edits working or failing. The swing is therefore an upper bound on pure noise, but it is exactly the number an operator experiences as "my rate moved". Second, variance is not futility: a sustained shift that holds for several months against this backdrop is real information. One month is not.
In practice
A founder whose acceptance dropped from 29% to 24% between April and May is inside the median band of this table. The defensible response is to change nothing and read June against this table: a five-point move is the median of what month pairs did anyway, and one month cannot distinguish a real shift from the ambient swing.
A RevOps lead evaluating an agency or a template rewrite on one month of before-and-after is running a test whose measurement noise (median 5.69 points, 90th percentile 16.61) is larger than most effects they are hoping to detect. The honest evaluation window in this data is a quarter, or a controlled split run in parallel.
Limitations
The swing conflates noise with unmeasured changes. Workspaces edit targeting, lists, and copy continuously. This analysis measures the month-to-month movement an operator observes, not a clean null distribution.
Sampling error contributes at the 300 threshold. Roughly 2 to 3 points of swing at that threshold is countable-noise floor; the 1,000-request robustness cut bounds this and the finding holds.
Pairs are not independent. A workspace active for many months contributes many pairs, and heavy senders contribute more. The dataset is concentrated (largest workspace 22.88% of matured requests, top ten 78.50%).
Calendar months are arbitrary units. Weekly rates would swing more, quarterly rates less. The month was chosen because it is the unit teams actually review.
Acceptance only. Reply rates ride on a smaller accepted denominator and swing harder; treating this study's thresholds as applying to reply rates would understate reply-rate noise.