Retail CPAs aren’t trending in a single direction right now. Some brands are maintaining efficiency, while others are seeing sustained pressure on acquisition costs. The typical response is to go straight into the ad account and start pulling levers: tweak bids, reshuffle audiences, reallocate spend.
In isolation, those actions make sense. But they rely on the assumption that the source of inefficiency sits within the ad account itself. In reality, many of the largest drivers of CPA sit upstream or downstream of media: the strength of the offer, the quality of the creative, and the effectiveness of the conversion journey.
Which means if the goal is to reduce CPA, the question isn’t just what to optimise in-platform but where the inefficiency is actually coming from.
You’re probably looking in the wrong place
Nine times out of ten, a CPA problem is a conversion problem in disguise. The ad is doing its job, getting people to the site. What’s failing is everything that happens after the click: a product page that doesn’t convert, an offer that isn’t competitive, a checkout with too much friction. Optimising media spend against a broken funnel just means spending more efficiently on something that isn’t working.
The first things to audit are your product feed and your offer. The feed (your product titles, descriptions, prices, and images) is what platforms use to serve shopping and retail ads, so if that data is incomplete or poorly structured, performance will suffer regardless of how well you bid. If the offer itself isn’t compelling on price or value, better targeting won’t compensate for it.
Bid strategy sits at the end of that list, not the beginning. It’s a useful tool for extracting efficiency from a programme that’s already working, but it can’t create efficiency where the fundamentals aren’t there. The same logic applies to channel diversification: expanding into new platforms because they’re growing or because a rep made a compelling case isn’t a demand strategy. Channel decisions should follow where your audience is and what role that channel plays in the wider system, not trend cycles.
The levers that actually move it
Once the conversion fundamentals are sound, two things tend to move CPA more than anything: creative and audience structure, and both are consistently underinvested.
On creative, the numbers are striking. Meta, citing Nielsen research, attributes 56% of campaign sales ROI to creative quality, and Google puts the figure at 70% of campaign success. Those aren’t marginal gains from asset refresh cycles. They reflect the difference between treating creative as a set-and-forget task and running it as a proper testing system. Meta’s own Cyber 5 data makes this concrete: ad sets mixing image, video, and vertical video formats delivered measurably lower CPAs than single-format sets.
Operationally, we treat this as a rolling test system, not one-off campaigns. We’ll run multiple variants per ad set at a time, each isolating a single variable like the opening hook, format, or offer, so we know what’s actually driving performance. We let each variant spend to a minimum signal threshold rather than a fixed timeline, then make decisions. The top performers get scaled, the middle get iterated, and the bottom performers are cut quickly. The goal isn’t to find one winner. It’s to build a portfolio of working creatives and continuously feed learnings into the next round, so performance compounds rather than resets each cycle. That’s the approach that works.
Audience structure follows the same logic. Running the same ad to everyone, cold prospects and people who have visited the product page five times alike, treats the funnel as flat when it isn’t. The message, offer, and format that works for someone who has never heard of you is different from what works for someone already warm, and the inefficiency compounds at scale.
The fix is to structure audiences in clear layers: at a minimum, cold prospecting, engaged users, and high-intent or existing customers, each with distinct messaging and creative aligned to where they are in the journey. What moves someone between layers should be behaviour, not time. A site visit or video completion shifts someone from cold to warm. Repeated interactions or an add-to-cart moves them into high-intent. That way, discovery-led creative reaches people at the top, proof and product detail does the work in the middle, and strong conversion drivers close at the bottom. You’re not overpaying to push people who just need a nudge.
The measurement problem nobody talks about
Even with the right creative and audience structure in place, most retail advertisers are still making decisions from a distorted picture. Part of that is how CPA gets reported. Part of it is which channels get invested in when pressure builds. Both come back to the same underlying problem: the numbers being used to make decisions don’t reflect what’s actually happening.
Every platform overclaims its contribution. Google takes credit for conversions. Meta takes credit for the same conversions. Reading both dashboards in isolation, without a neutral layer between them, makes double-counting almost inevitable. The more reliable approach is a blended CPA view: total spend across all channels divided by actual conversions from a source you control (your ecommerce platform or analytics tool, not the platforms). That’s the real number.
From there, incrementality testing tells you which channels are genuinely driving those conversions rather than just claiming them. For most retail advertisers, a realistic starting point is a simple geo holdout test: pause spend in a small, matched region, keep everything else constant, and measure the difference in conversions from your own data. Ask your agency for a clear test design and a success metric based on incremental lift, not platform CPA. That distinction matters, because it tells you which channels are driving net new demand versus simply capturing demand that already existed.
The same distortion affects channel investment decisions. When CPA is under pressure, upper funnel spend (awareness, video, broader prospecting) is usually the first thing to get cut, because it doesn’t show an immediate return in the numbers. But that’s the wrong read. Lower funnel channels like brand search, Shopping, and retargeting only look efficient because they’re reaching people who are already close to buying. That pool doesn’t replenish itself. Upper funnel is what keeps new people flowing into it, so cutting it to protect short-term CPA accelerates the problem rather than solving it. Because the effect isn’t immediate, the way to know it’s working is to track what moves first: branded search volume, direct traffic, engagement rates, and new user growth. These are early signals that demand is being built, and they will shift before conversion does. If they’re not moving, lower funnel CPA will eventually deteriorate, no matter how well the bottom of the funnel is optimised.
Bottom line
Sustained CPA pressure in retail is rarely a signal that something in-platform needs adjusting. More often it’s a signal that something earlier in the chain needs attention: the offer, the creative system, the audience structure, or the measurement layer telling you what’s actually working.
Start there. Focus on the parts of the system that matter most, and keep evolving as conditions change.
