Sales Experiments: 20 Tests You Can Run This Quarter

Why most sales teams stop experimenting after week 3

A VP reads a Gong report. Decides the team is going to test a new cold email opener. Monday morning: kickoff. Week one: three reps try it. Week two: two reps try it. Week three: one rep — the one who was never going to hit quota anyway. The "experiment" is dead, nobody called it, and nobody knows whether the new opener would have worked.

The test never had a chance. It had no hypothesis, no kill rule, no sample size, and no forum to read the result. It was a directive disguised as an experiment. Most sales experiments die this way — the team agrees to "try something different," leaves unclear what counts as different and what counts as success, and by week three the urgency of the quarter swallows the discipline of the test.

The sales experiments that actually change a quarter have a shape. A single variable. A sample big enough to read. A kill rule that fires when the variable underperforms. A forum — usually the weekly pipeline call — where the result gets read out loud in 90 seconds.

This post is that shape, applied to 20 experiments. Pick three. Run them for 10 business days. Read the result in the pipeline review. Kill the losers, scale the winner, queue the next three. The point is not testing for its own sake — it is how a team learns its own market faster than the competition learns it.

The anatomy of a sales experiment that ships

A sales experiment that ships has six parts. Skip any one and the thing turns into a directive nobody tracks. The rest of the post — and every one of the 20 experiments — is written against this shape.

1. 1

   The hypothesis

   One sentence, written in "If X, then Y" form. "If we lead with the signal in the opener, reply rate climbs from 6% to 10%." Not "we should try signal-led openers" — that is a suggestion, not an experiment.

2. 2

   The single variable

   One thing changes. Not the opener AND the CTA AND the send time. If three things change and replies move, you do not know which one moved them. Discipline here is what separates learning from noise.

3. 3

   The control

   The old version runs beside the new version on the same kind of prospect. If the control runs on Tier-2 accounts and the variant runs on Tier-1, the result is a firmographic difference, not a copy difference. Split the same segment and send both the same week.

4. 4

   The sample size

   Small enough to run in 10 business days; big enough to read. A sensible floor: 200 prospects per variant for reply-rate tests, 20 calls per variant for call-stage tests, 10 deals per variant for win-rate tests. Below that, you are reading variance.

5. 5

   The kill rule

   The condition that makes you stop the experiment before the sample fills. "If the variant replies under 2% after the first 75 prospects, kill it." A kill rule protects the pipeline from a bad variant that runs unchecked for three weeks.

6. 6

   The learning

   Written in one paragraph, shared in the pipeline review. Not a slide deck — a paragraph. "We tested signal-led openers against pattern-led on 240 prospects split evenly. Signal-led replied at 9.2%, pattern-led at 5.8%. Rolling signal-led into the Growth segment; retiring pattern-led."

If the shape sounds obvious, good — running it every time is the thing teams skip. The gap between "we tested this" and "we know this" is the shape. Reps write down what they learned, managers compound the learnings into the playbook, the next new rep inherits a sharper starting point than the last.

The experiments in the rest of this post are all written to fit this shape. Each has a hypothesis, a variable, a sample floor, and a kill rule. Pick the ones that match the quarter; ignore the rest for now and drop them in the queue document for next quarter.

How to pick the 3 experiments to run this quarter

Twenty experiments is too many to run at once. Three is the right number for a team of four to seven reps. Two for a team of three or fewer. Picking well is the one decision that makes the whole post useful — the other 19 go in a queue, not in the rearview.

The prioritization matrix. Score each experiment on two axes: (1) likely effect size at the quarter level, 1–5; (2) cost to run in rep hours, 1–5 where 5 is cheapest. Multiply. Run the top three.

A team with a reply-rate problem using this matrix runs experiments 3 and 13 first, queues 17 for next quarter, and deprioritizes 06. The math is coarse on purpose — the point is to break ties, not to rank 20 tests to three decimals.

The three-pick rules:

• · No two experiments share the same variable. If opener AND send time change this quarter, neither result reads cleanly.
• · At least one experiment hits the part of the funnel underperforming the team average. Do not test what is already best.
• · At least one experiment is cheap enough for one rep to run alone — so the team is not blocked if an experiment breaks mid-sprint.

The mistake most teams make is picking the three experiments they want to run. Pick the three that answer the question the team cannot answer today. "Is our opener broken?" is a better experiment prompt than "let us try that new opener I read about."

The 7-day sales experiment playbook

An experiment does not need a Gantt chart. It needs seven days, written out once, and a forum to read the result. Most outreach tests actually run 10 business days (two work weeks) — the playbook below maps to that cadence.

1. Day 0

   Friday before. Write the hypothesis on one line, the variant copy on another, the sample size on a third, the kill rule on a fourth. Slack it to the team. If anyone cannot describe the experiment in one sentence after reading, rewrite.

2. Day 1

   Monday launch. Split the prospect list 50/50. Launch both variants in parallel. For outreach, the sequence goes live in both versions; for calls, reps draw from a shared pool and record which variant they ran.

3. Days 2–4

   Watch the rates. Track reply, open, and book rates daily. If the variant trips the kill rule, kill it. If the control collapses (a sign that the data is off, not that the control is bad), pause and debug — do not rescue with more sample.

4. Day 5

   Halfway checkpoint. Are both variants on track to hit sample size by Day 10? If not, add prospects or extend the window by 3 days. Do not shrink the sample to fit the calendar — small samples are why teams reach wrong conclusions.

5. Day 10

   Freeze and read. Sample size hits. Freeze both variants. Pull the numbers from the CRM or outreach tool. No peeking early — the temptation to read it at Day 7 is what produces false positives.

6. Day 11

   Write the paragraph. Four sentences max: what was tested, what was seen, what the team is doing next, what to test next. Share in the pipeline review on Monday — 90 seconds on the screen, not a deck.

The forum matters more than the timeline. An experiment read out loud in a pipeline review, with numbers on the screen, is how teams build muscle around decisions. An experiment buried in a Notion doc never gets believed. Give it 90 seconds in the weekly ritual — that is all. If the result is inconclusive (variants too close to call), write that too. Inconclusive is a valid finding; it means the variable mattered less than expected, which is a learning in itself.

The 6 failure modes that kill sales experiments

Every experiment that fizzles fails in one of six ways. Spot the pattern before launch and the quarter is saved. Miss it and the team burns 10 business days on a test that produces no learning worth acting on.

1. 1

   No kill rule

   The variant underperforms by week one, but nobody calls it. The rep keeps running it to avoid looking like a quitter. Three weeks later the pipeline is thin and nobody learned anything. Fix: write the kill rule on Day 0. Post it in the channel. Let the number fire the decision, not the human.

2. 2

   Two variables changed at once

   The opener AND the send time got swapped. Replies move, and nobody can say which change caused it. Fix: one variable. If the team wants to test two, run two experiments in sequence, not one double-barreled one.

3. 3

   Sample too small

   50 prospects per variant feels like enough until the variance is 3 percentage points and the "winner" was noise. Fix: 200 prospects for reply-rate tests, 20 calls for call-stage tests, 10 deals for win-rate tests. Below those floors, call it a directional signal, not a result.

4. 4

   No control

   The team tested the new opener against "the old opener from Q1" — except nobody ran Q1's opener this week, so the comparison is against a ghost. Fix: the control must run in parallel, same segment, same week.

5. 5

   Result never written up

   The experiment ran, numbers came in, and the quarter closed before anyone wrote the paragraph. The learning evaporated. Fix: a one-paragraph writeup is mandatory. It goes in the shared queue document. The next experiment cannot start until the last one is written up.

6. 6

   Wrong funnel stage

   A team at 1% reply rate runs a close-phrasing test. Reply rate is the bottleneck; close phrasing is the wrong place to spend the hours. Fix: always experiment against the constraint. Identify the weakest stage in the team funnel first.

The failure mode underneath all six is treating experiments as inspirational rather than structural. An experiment is a rep-hour investment with an expected return; when the team treats it as "let us try this cool thing," the discipline falls apart. Treat each test as a small, serious bet. The compounding follows.

How Gangly gives reps a clean testbed for sales experiments

Experiments need two things reps rarely have inside the stack: parallel variants running in the same tool, and result data that is not buried in CRM exports. Gangly runs the workflow so the first experiment is the cheapest one to launch — every one after runs on the same rails.

• Workflow Sequencer — runs A/B variants of a cadence on split prospect lists inside one sequence, tagged as variant A or B. The rep sees the live reply rate per variant; the manager reads the result in the dashboard instead of a CSV export.
• Outreach Writer — drafts two versions of an opener from the same signal, flagged by variant. Reps review and approve both. Gangly does the typing; the rep owns the experiment.
• Post-Call Notes — tags every call with the experiment variant, so a call-format test produces a clean 20-call sample without the rep remembering which call was which.

The point is not that Gangly replaces the experiment — the rep and the manager still write the hypothesis, pick the variants, and read the result. Gangly removes the admin that otherwise makes experiments feel heavier than they should. If a team is running fewer than three experiments a quarter, the likely reason is not a lack of ideas — it is the overhead of setting one up inside a tool that was not built for parallel variants.

Start with one cadence test and read the result in 10 business days. The compounding is what matters: a team that ships 12 experiments a year turns 12 opinions into data. A team that ships none is still arguing about opener structure in Q4 with the same conviction it had in Q1 — and the same lack of evidence.

Related reading: our sales battle cards post covers the structure winning experiments get turned into, and the cold email reply rate study is the data set most of the outreach experiments above are calibrated against. The sales admin time study quantifies the rep-hour cost of the process experiments in section six.

Frequently asked questions