Sticky A/B variant on first pass vs re-randomize: same path forever vs new coin flip each visit

Sticky-first-pass assigns a variant on the customer’s first arrival at the A/B split and reuses it on later passes (and often across devices until identity merges); re-randomize draws a new variant each time they reach the node.

What is sticky A/B vs re-randomize?

Inside a journey A/B node you choose a philosophy:

1. Sticky on first pass: stable experience. If they saw subject A first, later re-entries usually keep A. Cleaner measurement, less cross-variant contamination.
2. Re-randomize every time: each visit to the node is a new coin flip. The same person may see A one day and B the next. Sometimes useful for exploration, worse for interpreting lift.

The classic FAQ—“What if I change the 50/50 ratio mid-test?”—matters: under sticky, prior assignments usually stay and only new arrivals follow the new ratio—unless the platform resets everyone. Read the docs.

Contamination risk table

Don’t confuse A/B splits with Audience vs Action routing—that’s state vs behavior: Audience Path vs Action Path.

Iranian example: subject-line test + Kavenegar SMS

A home-goods store tests two cart-abandon email subjects, then sends a reminder SMS in the same offer frame.

• Sticky: whoever got subject A keeps that narrative on re-entry; opens and purchases attribute cleanly.
• Re-randomize: email A then SMS framed like B wrecks “which subject won.”

Filter the audience before the split (e.g. cart-7d segment), not after assignment. For first-match vs ranked action windows see action-path first-match vs ranked window.

Adaptive vs rule-based journeys

When paths can change mid-flight, sticky experiment assignment matters even more so metrics don’t smear—see adaptive vs rule-based journey.

Truthful Leadara mapping

With Leadara journeys, segments, and email/SMS you can run subject/CTA experiments on nurture paths. Before scale, write down: sticky or re-roll each pass? How do you merge multi-device identity? Leadara AI Chat is an internal team assistant, not a customer live-chat bot. Don’t invent WhatsApp connectors.

When sticky vs re-randomize

Practical steps

1. One-sentence hypothesis + primary metric;
2. Audience filter before the A/B node;
3. Choose sticky vs re-randomize in the brief;
4. Lock the ratio; no casual mid-flight edits;
5. Build two test profiles that re-hit the split;
6. After identity consolidate, review assignments;
7. Report on first-assignment if sticky—not last-seen variant.

Common mistakes

• Live ratio edits without knowing prior assignment fate;
• Filtering losers after the split;
• Comparing variants when one person saw both (uncontrolled re-randomize);
• Mixing A/B with Audience Path;
• Ignoring mobile/desktop merge.

FAQ

What if I change the 50/50 ratio mid-test?

Under sticky, old assignments usually stick and new arrivals follow the new ratio—unless docs say reset. Safest: close the test and open a new one.

Can the same person see both variants on phone and laptop?

Before identity merge, yes. After unify, platform-specific; sticky usually keeps one winning assignment.

Filtering before or after the split?

Before. After skews the sample.

One assignment across SMS and email?

For a multi-channel experiment, share one sticky assignment so the message frame stays coherent.

When is re-randomize OK?

When the goal is experiential variety, not clean lift—and you label reporting accordingly.

Where does holdout fit?

Define holdout/control beside or before the split; don’t casually mix it with sticky without a brief.

Does an adaptive journey break sticky?

If the path changes but the experiment variant must stay fixed, keep sticky explicit in the design.

Leadership summary?

“A/B sticks on first pass; we don’t edit live ratios; filters sit before the split.”

Next step

Lock a cart subject-line test as sticky, create two re-entry test profiles, and sign a close-and-reopen protocol before any live ratio change.

Mid-flight test-close protocol

If business pressure wants a live ratio change, sign this first:

1. Announce when new assignments stop;
2. Close results to that moment on first-assignment;
3. Open a new test with a new variant name—don’t hack the same node;
4. Don’t re-randomize prior contacts unless that’s a separate goal;
5. Keep before/after windows separate in reporting.

Without this, every live ratio edit poisons lift interpretation for weeks.

Align email and SMS copy

If you test an offer frame (not only a subject), share sticky assignment across channels so Kavenegar SMS matches the email story. Otherwise customers hear two narratives and support gets confused.

One-page brief template

• Hypothesis: subject A beats B on cart-7d
• Primary metric: open; secondary: 48h purchase
• Assignment: sticky-first-pass
• Filter: before split; active cart, no purchase in 24h
• Ratio: 50/50 locked to end
• Duration: min N conversions or 14 days
• Exit: purchase → leave nurture

Related reading

• Audience Path vs Action Path routing: who they are now vs what they do in a window
• Action-path first-match vs ranked window: advance on first event or wait and pick the top path
• Adaptive journey vs rule-based journey: in-flight re-route vs fixed if/then