What should you A/B test first in a cold email?
Test in this order, and keep the winner fixed before moving on:
| Order | What to test | Why | A high-contrast test |
|---|---|---|---|
| 1 | Trigger (why you're writing now) | The biggest lever in our data | New role vs company growth |
| 2 | Opener | Decides whether the rest is read | Observation about them vs a question |
| 3 | Proof | What makes it credible to a Gulf buyer | Regional client example vs global one |
| 4 | Call to action | Turns interest into a meeting | Meeting ask vs offer to send a short summary |
| 5 | Subject line | Smallest effect on meetings | Their company name vs the topic |
In our audit of 242 campaigns, the trigger mattered more than any wording change: campaigns built on a job-change trigger produced more opportunities per email than those built on hiring, industry pain or funding news. See who you email matters more than what you write.
How many emails do you need to A/B test? (Sample size rules)
At least 150-200 people per version, and never call a winner below 100.
Textbook statistics need far more: often well over a thousand contacts per version to confirm a small difference. Most GCC segments don't have that many people. A list of finance leaders at mid-size companies in Dubai may be a few hundred people in total.
So in a small market:
- Treat results as directional. A clear gap at 150-200 sends per version is a signal to act on, not proof.
- Look for big differences. Test bold alternatives (a different trigger or offer), not small word changes.
- Repeat the winner on the next batch before rolling it out everywhere.
When not to split at all: if the whole segment is only a few hundred people, don't divide it. Send one version to the first batch, learn, and change one thing for the next batch. Sequential batches teach you more than two half-sized groups.
Which metric decides the winner?
Real replies and meetings, not opens or raw replies. Open tracking is unreliable and can hurt deliverability. Raw reply rate counts automatic replies and polite no's. In our audit, some copy got high reply rates but few genuine opportunities. Measure:
- Real reply rate: unique replies minus automatic replies. See real reply rate vs reply rate.
- Positive replies and meetings booked, once you have enough of them to compare.
Wait until the whole sequence has finished for both versions before deciding; late replies to the follow-ups count too. Count replies that arrive on LinkedIn or WhatsApp against the version that person received.
Stop early for deliverability. If one version starts bouncing or landing in spam, pause it straight away; don't wait for 150 sends.
How to set up a cold email A/B test (5-step checklist)
- Change one thing only between version A and version B.
- Keep a control: the current best version keeps running unchanged.
- Split randomly by company: two people at the same company should get the same version.
- Match the sending setup: same list, same days, same sending inboxes for both versions.
- Test openers as separate versions, not inside rotating text (spintax): rotating text spreads the difference across every email, so you can't measure it.
How is cold email A/B testing different in the UAE and GCC?
Smaller segments, connected buyers, and a calendar that shifts results.
Reputation risk in a small market
Burning through a segment with weak test versions costs you buyers you can't replace quickly, and people talk. Only test versions you'd be happy to send to everyone.
Local vs global proof
A regional client example against a global one is often a bigger lever than wording.
Formal vs direct tone
A warmer, slightly more formal version can beat very casual copy with senior Gulf buyers. Test it rather than assuming.
Calendar and working week
Compare versions sent in the same weeks; Ramadan, Eid and the summer distort any comparison across periods. UAE and Saudi working weeks differ, so don't mix the two countries in one test.
