Outbound experiments: change one variable

A team decides outbound needs a shake-up. In one energetic week they load a new list, rewrite the opener, move the calling blocks to the afternoon, and add a step to the cadence. Two weeks later connects are up and meetings are flat, and the retro is a shrug: something worked and something did not, and there is no way on earth to say which was which. The effort was real. The learning is zero. The program is now different, not better, and the next shake-up will be guesswork too.

The discipline in one sentence

Change one variable, hold everything else, give it a fixed window, and name the number that will judge it before you start. That is the whole method. It feels slow to teams in a hurry, which is exactly why so few run it, and why the ones who do end up with an outbound motion that visibly improves each quarter while everyone else redecorates theirs.

Each piece exists for a reason. One variable, because attribution is the entire point: an experiment that cannot tell you what caused the result is a lottery ticket. A fixed window, two weeks for most call-side tests, because open-ended experiments end when someone gets bored, which means they end on noise. And the number named in advance, because a result judged after the fact will be judged by whatever moved in a flattering direction. If the test is a new opener, the number is conversations that survive thirty seconds. If it is a list cut, it is connect rate. Decide before, not after.

What is actually worth testing

Not everything moves the funnel equally. The variables with real amplitude, roughly in order: which accounts you call, the first two sentences a buyer hears, when you call, and how you follow up. A new list segment can double a connect rate; a polished closing line moves almost nothing. Test the big levers first, and mine the levers from evidence you already own: the recordings, the reply patterns, the segments your weekly funnel numbers keep flagging. Mid-July, with H1 results fresh and the fall still ahead, is the natural season for this: the H1 postmortem generates hypotheses, and the quiet weeks are cheap laboratory time. The experiments you run now compound in September; the sequencing logic we laid out in the September start applies to message tests as much as to programs.

Respect the sample size

The most common failure after changing too much is concluding too fast. Outbound samples are small. A rep runs a few hundred dials and a few dozen conversations in a two-week window, and at those volumes a good week and a bad week can be pure variance, the trap we covered in when not to react. Practical guardrails: judge rates, not counts; require the difference to be large before you crown a winner; and when a result looks miraculous, run the window again before rolling it out. A real improvement will survive a second look. Noise will not.

Also keep a log. One line per experiment: what changed, when, the number before and after, the verdict. Without it, the same tests get rerun every two quarters by whoever was not there the first time.

This discipline is built into how we run engagements: one change at a time to a client's track or list, judged against the weekly report everyone already reads. It is unglamorous, and it is why month four consistently outperforms month one. If your outbound gets redecorated every quarter and never quite improves, book a strategy call. Bring the last three changes you made, and we will work out together which one, if any, did the work.