Prospecting Call Script A/B Test
Purpose: find out which wording works best on each of the three prospecting calls, using results we record ourselves rather than third-party benchmarks. The scripts are in Prospecting Cold Call Scripts. This doc says which variant to use, what to record, and how a winner is decided.
Test window: 2026-10-06 to 2027-01-04 (12 weeks), or until every decision below is made, whichever comes first.
Closed on 2026-11-04, before any decision. The S3 control changed underneath the test: the pre-call send became Email 1A, which carries the Results-First offer, and every call now closes on that offer. A variant dialled before that date and one dialled after it would differ by the email and the offer, not only by the twenty seconds this test compares. No results had been entered in the tables below. The scripts doc now carries one form per situation, and the variant options in
opener_usedare no longer assigned. The method below stays as the template for the next test.
Why Test This Way
There is almost no data yet. One prospecting call had been made when this test started (2026-08-31, hung up). Every rate in the scripts doc is a third-party benchmark.
Volume was small when this was written, and is no longer. The original premise was 25 to 50 prospects a month reaching the cadence, or roughly 10 to 20 conversations. A dedicated caller dialling 40 to 55 a day changes that by an order of magnitude: a four-day block is 160 to 220 dials, which at the 15 to 22 percent connect rate this ICP shows is 25 to 48 conversations in a week. The decision rules below were calibrated for the old number and are now conservative rather than optimistic — a screening decision that was going to take a quarter can take a fortnight. Florida, Ohio and Pennsylvania are still never dialed. Booked meetings remain too rare to compare scripts on, so each test is still judged on something that happens earlier in the call.
So the results are directional, not statistically significant. A winner here means a clear practical difference, backed by what the call recordings show. It is not proof. Say so whenever a result from this test is quoted.
More than one caller, from 2026-10-29. This test was designed around a single caller, so that the voice never varied between variants and the script was the only thing that did. That assumption no longer holds, and it is the one that can invalidate a result: a variant dialled mostly by one person and a variant dialled mostly by another differ by voice, accent, pace and experience as well as by wording. So the CALL note must record who dialled, and no variant may be compared across callers without saying so. Where the counts allow, compare within a caller first.
The Variants
Each situation is its own test with its own control. Every challenger changes one thing. The full wording is in the scripts doc.
| Situation | Variant | What it says differently | Hypothesis |
|---|---|---|---|
| S1 cold, phone only | S1-A (control) | Opener, finding, then "Can I send it over? What is the best email?" | |
| S1-B | Asks for the email first, as permission, then gives the finding | Asking for the address before the finding gets more reports sent | |
| S2 follow-up | T3-A (control) | "…Did you get a chance to open it?" | |
| T3-B | "…Is that something you already knew about?" | The open question invites "no", and "no" ends the call. Asking about the finding removes that dead end | |
| T3-C | "Have you got thirty seconds for the one thing in it worth fixing first?" before the finding | Asking permission first cuts early hang-ups | |
| S3 send, then call | S3-A (control) | "Have you got two minutes while it is open?" | |
| S3-B | Gives the finding first, then "Want me to show you where to check that?" | Leading with value keeps more people on the line than asking for time |
Everything after the ask is shared: the forty-five-second block, the qualifiers, the twenty-minute walkthrough ask and the objection handling. So each test compares only the first twenty seconds or so.
Assignment
- Rotate within each situation in the order prospects reach it. In S2, the first prospect gets T3-A, the next T3-B, the next T3-C, then back to T3-A. The caller never chooses.
- Set
opener_usedbefore the first dial. A prospect keeps their variant on every redial. - A dropped variant leaves the rotation, and the rotation continues with what is left.
- Exclusions apply as normal: FL, OH and PA are never dialed, and the stop rules in the call execution SOP still bind.
What Counts
A conversation is a dial where the decision maker picks up and does not decline recording. Gatekeepers, voicemail and no-answer are logged but are not conversations.
| Situation | Primary measure (the call's one job) | Guardrails |
|---|---|---|
| S1 | An email address captured, or the report requested | Hang-up within 20 seconds; any do-not-call request |
| S2 | The call reached the twenty-minute walkthrough ask | Same |
| S3 | They stayed on and walked through the finding live | Same |
Recorded on every call, as secondary measures: walkthrough booked, connected-interested, and call length.
How it is recorded: the one-line CALL note from Step 6 of the call execution SOP, plus opener_used on the contact. The note must name the caller, per the multi-caller caveat above. No other fields.
Decision Rules
Applied per situation, at the weekly review.
- Screening, after 10 conversations per variant. Drop a variant if its 20-second hang-up rate is double the control's, or if it hit its primary measure zero times.
- Head-to-head, until 20 conversations each. The best surviving challenger against the control. The challenger wins if it beats the control on the primary measure by at least 15 percentage points and is no worse on either guardrail. Otherwise the control stays.
- After a decision, the winner becomes the new control in the scripts doc. If the situation is to keep testing, a new challenger is drafted from what the recordings and the objection list show.
- Hard stop on 2027-01-04. Any situation without a decision is decided on the counts so far plus the listening notes, and the reasoning is recorded below.
A do-not-call request on any variant is reviewed the same week, whatever the counts.
Before the First Live Call
- Role-play every variant against three typical owners drawn from the bucket taxonomy: a busy roofer, a skeptical landscaper, and a gatekeeper. Reword anything that sounds wrong out loud.
- Read each variant aloud. It should reach the ask in about twenty seconds.
- Set up
opener_used. It needs the optionsS1-A,S1-B,T3-A,T3-B,T3-C,S3-AandS3-Balongside the existing A, B and C. See the structure inventory. - Log one test dial per situation on the test contact. Confirm the weekly count picks all three up.
- Check whether call length and recordings are stored against the conversation when calls go through the platform phone. If they are, take call length from there and use those recordings for the weekly listen.
Weekly Review (About 15 Minutes)
- Count. Pull every contact note starting
CALL |from the past week. Count conversations, primary hits, 20-second hang-ups and do-not-call requests by situation and variant. Add them to the results table. - Listen. One recording per active variant, against five questions: did they stay past 20 seconds, did they confirm the finding, did they ask a question, did the call reach the ask, and what was the objection?
- Write down every objection in the objection log below, in the prospect's own words where possible. New objection-handling content comes from this list.
- Apply the decision rules.
Results
| Week ending | Situation | Variant | Conversations | Primary hits | 20-sec hang-ups | Do-not-call | Booked |
|---|---|---|---|---|---|---|---|
Objection Log
| Date | Situation | Variant | What they said |
|---|---|---|---|
Decisions
| Date | Situation | Decision | Counts at decision | Why |
|---|---|---|---|---|
| 2026-11-04 | All | Test closed, no winner declared. S2 keeps the T3-B question as its only form, on judgement rather than counts: it has no "no, have not looked" dead end. S1 and S3 keep their controls, reworded for the offer | None recorded | The pre-call email and the offer changed for every variant at once |
After This Test
- S2 against S3. Once each situation has a winner: is calling minutes after the email better than calling a week later, for the same kind of prospect? That is a cadence change, so it needs its own test and, if it wins, a change to the prospecting pipeline SOP cadence.
- This doc is the template for the next test. Copy it, change the variants and the window, and keep the measures and decision rules unless there is a reason to change them.
Related Documents
- Prospecting Cold Call Scripts: every variant's wording
- Prospect Phone and Call Execution SOP: call blocks, dispositions, the
CALLnote - Prospecting Objection Handling: shared by every variant
- SC Outbound Calling Compliance Standard: what is permitted, by state
Document History
| Version | Date | Changes |
|---|---|---|
| 1.2 | 2026-11-04 | Test closed before any decision, because Email 1A and the Results-First offer changed every situation at once. Closure recorded in Decisions; the doc stays as the template for the next test |
| 1.1 | 2026-10-28 | Two founding premises corrected rather than left to mislead the first result. Volume: a dedicated caller at 40 to 55 dials a day makes the 25-to-50-prospects-a-month basis obsolete, so the decision rules are now conservative and a screening decision lands in a fortnight rather than a quarter. Single caller: no longer true from 2026-10-29, which is the assumption most able to invalidate a comparison, so the CALL note must record who dialled and cross-caller comparisons must be declared. No variant wording changed here; S1's control lost the honesty interrupt in the scripts doc |
| 1.0 | 2026-10-06 | Initial version. Three situations, seven variants, measures that suit low call volume, screening and head-to-head decision rules, weekly review, 12-week window |