Link Leads Blog · October 26, 2026

How to A/B test SMS scripts on an aged insurance lead list

Swapping in a new opener and watching the reply rate move isn't a test — it's a guess wearing a spreadsheet. A real test holds the list, the day, and the hour constant and changes exactly one thing. Here's how to run one on a batch you've already bought.

Why most script "tests" aren't tests

The usual version goes like this: an agent texts 2,000 leads with script A on Monday, gets a reply rate, then texts another 2,000 with script B on Wednesday, gets a different reply rate, and declares a winner. The problem isn't the sample size — it's everything else that changed between the two sends. Monday isn't Wednesday. The first 2,000 rows of a state-and-age filtered export aren't the same mix as the next 2,000. A reply-rate gap produced that way could be the script, the day, or just which slice of the list got dialed first. A clean test changes one variable and holds the rest still.

Pick one variable per test

Running more than one change at a time is the most common way a test gets wasted — if the reply rate moves, there's no way to know which change did it. Variables worth testing one at a time on an aged list:

Pick one, write variant A and variant B so only that element differs, and leave everything else — vertical, state mix, send window — identical.

Split the list before you touch a dialer

A CSV export doesn't arrive in a meaningful order, so slicing the top half for variant A and the bottom half for variant B is close enough to random to work — see how to filter an aged insurance lead CSV by state and age for the spreadsheet formulas that do this split and the dedupe check to run first. If you're ordering fresh for the test, pull one state-and-age filter in the order builder and split that single export in two, rather than ordering two separate batches — a second order can draw from a different slice of inventory and quietly reintroduce the bias you're trying to remove. Aim for at least 500 sends per variant; aged-list reply rates run in the low single digits, and anything smaller than that mostly measures noise, not the script.

Run both variants on the same day, in the same hour band

Time of day moves reply rate as much as word choice does — see best time to call aged insurance leads for why. Send variant A and variant B in the same send window, on the same day, so neither one is quietly winning or losing on timing instead of content. Tag each variant with its own short code in whatever field your dialer or CRM uses for campaign source, so a reply can be routed back to the script that produced it without you having to remember which number range got which message.

Measure more than the raw reply rate

Reply rate is the headline number, but it isn't the only one that matters. Pull three numbers per variant: total replies, STOP replies, and qualified replies — a "wrong number" or a one-word "no" is a reply but not a lead. A script that pulls a higher reply rate but also a higher STOP rate isn't necessarily the winner; it may just be more aggressive at provoking a response either way. Call disposition codes for aged insurance leads has a tagging set that makes this split easy to pull from whatever you're already logging, and every STOP, regardless of which variant produced it, has to be honored and removed from both lists — see SMS opt-out compliance for how to route and log those.

How big a gap has to be before it's real

At the volumes most agents test with, small gaps are noise. If variant A gets 22 replies out of 1,000 sends (2.2%) and variant B gets 24 out of 1,000 (2.4%), that's not a result — it's within the range two identical scripts could produce by chance. If variant B gets 34 out of 1,000 (3.4%) against variant A's 22 (2.2%), that's a gap worth acting on. As a rough rule of thumb on aged-list volumes: don't call a test at under 500 sends per side, and don't call a result at under about a one-point spread in reply rate unless it holds up again on a second batch. One batch is a signal, not a verdict — the Resurge beta results post is a worked example of reading one batch's numbers without over-claiming what they prove.

Roll the winner forward, don't just file it away

Once a variant wins clearly, make it the default script for that vertical and state mix going forward, and retire the loser rather than keep splitting future sends between them out of habit. The losing variant isn't wasted, though — feed anyone who didn't reply to either message into your normal cadence rather than dropping them, and test the next variable (length, CTA, send time) against the new default once you've got another batch to run it on.

Not legal advice

This is a testing methodology, not a compliance guide. Every variant in a test still needs a clear opt-out path and still has to respect the same calling/texting rules as any other send — splitting a list for testing doesn't change what TCPA, state mini-TCPA statutes, or carrier rules require. Confirm your own obligations with counsel before you text any list, aged or otherwise. Link Leads sells aged data leads — not consented insurance-form leads — and is not DNC-scrubbed; the buyer is responsible for all dialing and texting compliance.

Get the list

Pull a state-and-age filtered order in the order builder large enough to split into a real test — 1,000+ per variant is a reasonable floor — or grab the free 100-row sample first to see the column layout before you buy.

These are aged data leads (not consented insurance-form leads) and are not DNC-scrubbed. The buyer is responsible for DNC scrubbing and dialing/texting compliance on every list before any call or text goes out.