How we decide when something is not working
The arithmetic behind the weekly read, in full, so you can check it.
Every number on this page comes out of one idea, and it is the idea worth checking first. Detection does not need a fixed number of prospects. It needs a roughly fixed number of expected replies. So the prospects you need is that count divided by your reply rate, and nothing below is doing any hidden work.
Everything here is calibrated against the level we treat as a working campaign, 5% of prospects replying, with a 5% false-alarm budget. That 5% is a bar we declared, not an average we measured: what a campaign has to clear, not a promise about what yours will do. The sequence assumed throughout is one letter and up to 3 follow-ups on LinkedIn or by email. Every figure is exact binomial, not a normal approximation.
The same question, asked of two campaigns
Is this one dead? A letter sent on its own is a single ask, where a working campaign returns 2.67%, so answering takes 289 prospects. A letter plus the follow-ups is the full sequence, where a working campaign returns 5%, and answering takes 153. Same rule in both cases, dead at 3 replies or fewer. Same false-alarm rate, 4.9%. Same power, 87%. Multiply each sample by its rate and both come out at about 8 expected replies. That is the quantity actually held fixed.
So the follow-ups do not buy more trials. The same person ignoring you four times is one outcome, not four. What they buy is a higher rate, and when the event is rare, lifting the rate does the same work as lifting the sample. That is why we sequence, and why a letter sent alone is the expensive way to learn anything.
What each sample size settles
| Prospects | What that settles |
|---|---|
| 30 | Whether anyone is reading it at all. At a 10% scan rate, seeing zero scans out of 30 is under 5% likely, so zero scans is real news. |
| 48 | A standout angle proves itself outright: a true 25% reply rate is provable here, and good news is cheap. Bad news is not available at all. There is no number of replies out of 48, at any threshold, that would let us honestly call a campaign dead. |
| 110 | A collapse to a quarter of a working campaign, 5% down to 1.25%, becomes more likely than not to be caught. Only just: 60%. |
| 153 | The same collapse is caught 87% of the time. This is the honest point where we would tell you. |
| 361 | A milder shortfall becomes catchable: a campaign returning half of a working one, 2.5%, is caught 80% of the time. |
| Beyond reach | Angle A against angle B, and a working rate against a slightly worse one. If half of the healthy rate already takes 361 prospects, two rates that sit close together take many hundreds more, well past anything a quarter of sending produces. We do not attempt it and we will not report a winner we cannot see. |
What this cannot do
It cannot A/B two angles. The table stops where it does: separating two rates that sit close together needs a sample this channel will not reach, so any ranking of angle A against angle B would be noise wearing a percentage sign. We change angles on judgement and on what prospects say back, and we tell you that is what we are doing rather than dressing it up as a test.
It also cannot rescue a small number by discussing it for longer. Before 48 prospects there is no negative call at all, and between 48 and 153 the honest answer to "is this dead" is still too early. That is what the weekly read will say, in those words.
The errors we accept, as numbers
We accept about a 5% chance of calling a working angle dead. What that buys is in the table above: a collapse to a quarter of the healthy rate is caught 60% of the time at 110 prospects and 87% at 153. The two errors trade against each other and you cannot buy both cheaply. That is the point we picked, and you are entitled to hold us to it.
It explains an asymmetry you will feel in the first weeks. Good news is cheap and bad news is expensive, because you are counting something that barely happens either way. A rate well above the healthy bar announces itself inside the first few dozen letters. Silence needs hundreds before it says anything with the same confidence.
Near zero, which is where the first month lives, we report a Wilson score interval rather than the normal approximation from the textbook. Down there the approximation puts the lower bound below zero and hands you a confidence nobody earned. Wilson stays inside the interval it describes and stays usable at 12 letters a week.
If this does not survive your arithmetic
Tell me and I will fix the page. An earlier version of this answer named 110 prospects as the point where an angle could be ruled out. At 110 we would catch that collapse 60% of the time, barely better than a coin toss, so the number moved to 153. That correction is the reason this page exists.