I haven't seen A/B Testing itself (not just the change introduced) hurt CAC — not to mention that it's very hard to prove. Meta is a very complex beast, and attributing a spike in CAC to a launched test ignores the hundred other variables in play at the same moment. And even if that's the case — that's the whole point of Experimentation! Changes are going to be introduced either way, so why not try to understand their effect, learn, and iterate? What is a valid concern is that CR/ARPU uplifts don't necessarily translate into better CACs or ROAS. In theory, increasing CVRs by 10% should decrease CPAs by ~10%, but in reality, the effect is hard to isolate and probably not linear — especially if the effect is negative. Recently, I ran an experiment that generated a 10% increase in ARPU (statistically significant, correctly powered), driven by a 25% increase in AOV and a 15% decrease in CVR. In theory, it worked. But when we pushed the change to production, CACs — instead of increasing 10% — increased 25%. My theory is that Meta penalizes your site for converting worse by charging higher CPMs, so you end up getting hit twice. An Intelligems A/B test can never catch that, because it happens before the randomization occurs. There's also a conspiracy theory going around that if your ARPU improves because of a test, Meta will detect it and — since you're now capable of paying more — it will raise your CACs. I want to believe that's not the case, but I have my doubts. So yeah, observing how CAC/ROAS behave after a big experiment is sent to production is important, but again, hard to prove causality — especially when the effect is small (3–5%).