Apex Overlay

Automated Email A/B Testing

Visit Website
March 23, 2026 I built a thing because cold email A/B testing is honestly broken

Been running cold email for my dad's HR software company since I was 16. I'm 21 now. Five years of the same frustration: you test 3 email versions, one clearly gets more replies, and your sending tool just... keeps splitting evenly across all three. By the time you notice and manually fix it, you've wasted hundreds of leads on your worst-performing copy.

Instantly.ai does have a built-in auto optimize A/B feature, but it splits sends equally until it has enough data, then hard-switches to one email and kills the rest. That can take days. Every send during that waiting period is a coin flip — and the leads you burn on bad copy are gone forever.

So I built Apex Overlay. It connects to Instantly and starts figuring out which email is getting the best results from the very first sends. As replies and real clicks/opens come in, it automatically starts sending more of what's working and less of what isn't. No waiting, no manual checking, no hard switch. The weaker emails still get a small share of sends so your sending pattern stays natural to spam filters.

It also filters out bot clicks and opens — in our first campaign, 31% of all clicks were from security scanners, not humans. If you're making decisions based on that data, you're optimising on noise.

Two campaigns so far:

StaffCircle — 1,053 HR leads, 3 email versions. By day 12, ~96% of sends were going to the best-performing email. 6% more replies than an equal-split approach. A curiosity-style subject line beat two direct pitches across every persona — a finding we never would have caught with manual testing.

Notetaker — within 24 hours, the engine had already shifted 49% of sends toward the better email. By day 3, 76.8%. The best email hit a 3.2% reply rate while the other two sat at 0%. We got a qualified reply within 3 hours of the first optimised sends — that reply converted to a deal.

14% more replies on day 1 for that campaign - automatically. Not after a week of "testing." Day one.

It's in free beta right now. 3 months free for early users (code FOUNDING), connects to Instantly in about 2 minutes. Full results at apex-scale.com/results.

Happy to answer questions about how the bot filtering works or the algorithm behind it. For the technically curious — it's Thompson Sampling, and this is a prototype. We're testing it on Instantly.ai first. If we keep seeing results like this, the plan is to expand to other platforms and marketing channels where the same "send more of what works" logic applies.

Genuinely curious if anyone else has hit this same problem.


Comment

About

Worked in email marketing from a young age, spent days of my life trying to compare campaigns and which copy engaged better - so I learnt about learning algorithms and how to make email campaigns automatically learn