AI Did Not Lower Your CAC. It Hid Where It Is Rising.

A founder told me his acquisition numbers were great.

“Our CAC has been very stable for the past few weeks. Amazing!”

I asked him one question. What did you change to keep it stable?

“Nothing. I just assumed it’s the status quo since nothing changed.”

So we opened the data. Plenty had changed. The channel allocation had moved a lot. More budget was flowing to retargeting and brand searches. Less was going to cold audiences. His blended customer acquisition cost (CAC) sat perfectly flat the whole time.

The machine had changed what he was buying. It never told him.

The consensus view says the algorithm optimises better than humans, and a stable CAC proves it is working. I disagree with the second half. My argument: AI-driven bidding did not lower your CAC. It removed your ability to see where CAC is rising. This article shows how the blind spot works, and what you can do about it while staying on the platforms.

1. A stable blended CAC is an average, and averages absorb movement

Blended CAC is total spend divided by total new customers. It is one number stretched over very different purchases.

A conversion from a brand search is cheap. The customer already knew you. A conversion from a retargeting ad is cheap. They were already on your site. A conversion from a true cold audience is expensive. That last one is the only one that grows your business.

Shift the mix toward the cheap conversions and the blended number holds steady. It can even fall. Meanwhile, the cost of a genuinely new customer climbs, unseen.

A flat CAC line does not mean nothing changed. Sometimes it means something changed exactly enough to keep the line flat.

2. The blending is by design, and the platform grades its own homework

Google’s Performance Max (PMax) puts Search, Shopping, YouTube, Display, Gmail and Maps into one campaign. It reports one result. Meta’s Advantage+ pools prospecting and retargeting into a single budget.

This is not a bug. The black box is the product.

Optmyzr studied 503 accounts in February 2025. In 91% of them, PMax overlapped with the advertiser’s own Search keywords. Search-term overlap hit 97%. Branded queries are the cheapest conversions in the account, so PMax happily takes them.

Here is the detail that should bother you most. The researchers could not measure the cost impact at all, because Google did not report cost data for PMax search terms. The people auditing the blind spot were blinded by it.

Meta runs the same play. One agency audit of 42 ad accounts found that in 38 of them, Advantage+ spent up to 45% of the daily budget on returning customers. Return on ad spend (ROAS) looked great. New customers were missing.

💡 Key Takeaway: The platform decides where your money goes, reports its own results, and hides the breakdown. That is not measurement. That is marking your own exam.

3. Reported CAC and incremental CAC are different numbers

The gap between them is the expensive part.

At Uber, Kevin Frisch turned off roughly $100 million of a $150 million performance budget. App installs barely moved. The installs the platforms claimed were mostly happening anyway.

Airbnb cut performance marketing by $541 million in 2020. Traffic came back to 95% of 2019 levels without it.

These are extreme cases at extreme scale. The mechanism is not extreme at all. Automated bidding hunts for the cheapest reported conversions. The cheapest reported conversions are the ones most likely to happen anyway. So the machine is pulled, by design, toward spend that looks efficient and adds little.

Your dashboard CAC measures what the platform claimed. Your incremental CAC measures what your money actually caused. Only one of them pays your bills.

4. The algorithm is not the villain, and leaving is not the advice

Let me concede the strongest counterargument, because it is true. Value-based bidding fed with good first-party signals works. Google’s own data shows a 14% median lift in conversion value for advertisers moving from target CPA to target ROAS. I have seen automated bidding beat manual media buying many times.

And you cannot leave. For most founders in Singapore and Southeast Asia, Meta and Google are the distribution. Boycotting the black box is not a strategy.

So the answer is not less automation. It is a different relationship with it. Feed the machine your best first-party data, because it genuinely bids better with it. Then refuse to let it be the only measure of its own output.

5. What you can do this quarter

The good news: the platforms have quietly returned some controls. Use all of them.

On Google:

• Turn on brand exclusions in PMax, so brand demand stops subsidising the reported number
• Add campaign-level negative keywords, now up to 10,000 per campaign
• Read the PMax search terms report monthly, not never
• Watch the channel performance report. It shows where budget flows, even though it still refuses to let you steer it

On Meta:

• Define existing customers and set the existing-customer budget cap, so Advantage+ must actually prospect
• Split every result by audience type: new, engaged, existing

Then build the measure the platforms will never give you. Track CAC by channel and by audience from your own data, and run a holdout test each quarter: pause a channel in one market and watch what actually drops.

This is not theory for me. At a video over-the-top (OTT) streaming platform where I led digital acquisition, I built a measurement-led growth framework outside the ad platforms. It delivered a 2.5x cost-per-acquisition (CPA) improvement, from the same channels everyone else was using.

💡 Key Takeaway: Feed the algorithm your data. Never let it grade its own homework. Keep one CAC table the platforms cannot touch.

Final Thoughts: You cannot manage a number you can no longer see

Automated bidding is genuinely good at buying ads. It is genuinely bad at telling you what it bought. A flat blended CAC can hide a rising cost of real growth for months, and the bill arrives when you try to scale, or when an investor runs the numbers you did not.

You do not need to quit the platforms. You need your own scoreboard: CAC by channel, CAC by audience, and a periodic test of what your spend actually causes.

If your CAC has been suspiciously stable while your growth has not, that is worth a conversation. Book a discovery call or connect with me on LinkedIn.

---

A note before you close this tab. The fact that you read this far tells me something. You already sense that the way you’ve been thinking about growth might be incomplete. That instinct is worth following.

Mervyn Chua is a growth-transformation consultant helping founders and CEOs build the strategic clarity and systems to grow in an AI-first world. If this raises questions worth exploring for your brand, let’s talk.