Paid social, honestly: what it can and can't do for loyalty

There’s a comfortable story going round that paid social can “do retention” now. Run some retargeting, switch on a few dynamic ads to past buyers, and the platform hands you a 6 to 10x ROAS that looks like loyalty bought with money. I went and checked what that number is actually measuring, because the gap between what paid is credited with and what it genuinely adds is where a lot of small-brand budget quietly leaks. Here’s the honest version: paid social is brilliant at one specific job, mediocre at another it keeps getting praised for, and worth a lot more to you once you know which is which. Founder to founder.

The short version

1. Paid’s real job is acquisition. It buys the first customer and it decides who walks into your retention machine. It does not, on its own, build loyalty. That happens in your owned channels.
2. Most “paid for retention” spend is non-incremental. Retargeting people you can already reach by email reads a flattering platform ROAS (often quoted around 6 to 10x), but far less on a proper holdout (nearer 1.5 to 2.5x), and a meaningful share of those purchases would have happened anyway. Those last figures are modelled and directional, not a fact about your store.
3. There are three narrow seams where paid genuinely lifts lifetime value. Win-back of truly lapsed customers email can no longer reach, value-led lookalikes that raise the quality of who comes in, and ads that acquire people straight into a subscription.
4. Judge paid on the real economics, not the platform number. Blended efficiency, contribution margin and new-customer cost, never the in-platform ROAS the dashboard hands you.
5. Exclude the people you already own. Keep existing purchasers and your email list out of prospecting, and cap how much you spend re-reaching warm traffic.
6. If email and SMS are under roughly a quarter of your revenue, fix that first. That is almost always cheaper growth than another pound on “retention ads.”

---

1. What paid actually does (and where the loyalty myth starts)

Here’s the cleanest way to hold it in your head. Paid social owns customer acquisition and the cost of it, and it selects who enters your world. Your email and SMS own the repeat relationship, because the replenishment, post-purchase and win-back flows run at near-zero marginal cost once they’re built. In Klaviyo’s own numbers, automated flows do around 41% of email revenue off roughly 5.3% of the sends (Klaviyo, 2025). That is the loyalty engine, and it is mostly owned, not rented.

That matters more every year, because the front door is getting more expensive. Median DTC acquisition cost is now cited at roughly $130 to $156 per customer, up about 60% over five years (Rareview, 2025). And targeting got blunter at the same time: Apple’s tracking prompt zeroed the ad identifier for the roughly 75% of users who don’t opt in (ATTN Agency, 2026), so the algorithm now finds buyers through the creative far more than through clever audience-building. The single biggest lever you have on paid performance is the work itself: one analysis puts just under half of incremental sales, around 49%, down to the creative, more than targeting, reach and recency combined (NCSolutions, 2023).

So paid is a top-of-funnel machine that has got pricier and leans harder on good creative. None of that is a loyalty story. The loyalty story starts the moment someone you acquired comes back, and that comeback is overwhelmingly an owned-channel event. The myth creeps in because the ad platform happily takes credit for those comebacks too.

2. The retargeting mirage (read the holdout, not the platform)

Here’s the bit worth slowing down for. When you retarget your own site visitors, past purchasers and email subscribers, the platform shows you a gorgeous return, often in that 6 to 10x range. The problem is what that number counts. A lot of those people were already coming back. They had the cart open, or your welcome flow was about to land, or they’d have searched your name on Friday anyway. The ad got shown, the sale happened, the platform claimed it.

When you actually test it, by holding a chunk of the audience back and showing them nothing, the picture changes. On a proper holdout, that flattering retargeting ROAS tends to land nearer 1.5 to 2.5x, and somewhere around 30 to 40% of the attributed purchases are revenue your owned channels would have closed for free (Stella, 2025; WorkMagic, 2025). I want to be straight about these specific numbers: they are modelled, directional benchmarks from the incrementality world, not a measured fact about your shop. Treat them as a direction of travel, not a promise. The robust, repeatable finding underneath them is simpler and harder to argue with: in-platform ROAS runs roughly 2 to 3x above true incremental, and it overstates worst on exactly the warm, retargeting spend that brands lean on hardest (directional, per the incrementality benchmarks above).

The practical tell shows up around launches. If you let paid retargeting fire at the same moment as your Klaviyo send, the ad will “win” the sale your email was already going to make, and your dashboard will tell you the ads carried the launch. The honest sequence is the other way round. Let owned go first. Your waitlist, back-in-stock and launch email plus SMS fire on the day, convert the people who were always going to convert, and then paid retargeting turns on 24 to 48 hours later as a small, capped harvesting layer for the stragglers. Same spend, very different read on what’s actually doing the work.

3. The three seams where paid genuinely lifts lifetime value

None of this means paid can’t touch loyalty. It means it only does when it reaches someone your owned channels structurally can’t, or when it changes who comes in. There are three seams worth your money, and they only count when you prove them on a real cohort, never on attributed ROAS.

Win-back of the truly lapsed, the ones email can’t reach. This is the strongest case, because it’s the one place with a genuine reachability gap. A customer who lapsed 90 to 180 days ago and no longer opens a single email is, for practical purposes, unreachable in your owned channels. A paid custom audience of those lapsed non-openers, built from Klaviyo or Shopify and explicitly excluding anyone still email-engaged, can bring a slice of them back. One cited case showed around 17% reactivation from unsubscribed customers via paid against about 5% via email, but treat that as illustrative, not a number to forecast off (directional). The discipline that makes it real: exclude your email-reachable customers, or you’re just paying for what a flow does for nothing, and validate it on retained reactivation at 90 and 180 days, not on the ad’s ROAS line.

Value-led lookalikes that raise the quality of who comes in. This one doesn’t retain anyone directly. It changes the intake. If you seed a lookalike (or feed a predicted-value signal to a broad campaign) from your top quartile of customers by lifetime spend, you bias acquisition toward higher-order-value, repeat-prone buyers, so the lifetime value your owned channels go on to realise is structurally higher. Be honest with yourself about the limits, though. The “30 to 40% higher AOV” claims floating around are directional agency numbers with no clean causal test behind them, and the edge has been fading as broad targeting catches up. It also needs real lifetime-value data on a decent base of customers before the seed means anything. It’s a quality-of-intake lever, not a magic audience. Worth remembering here that founder presence and product still do the heavy lifting on who sticks: only about 13% of beauty buyers cite the founder as a key reason to purchase against 39% citing product performance (McKinsey, State of Fashion: Beauty, 2025), so paid changing who comes in works best sitting on top of a product that earns the second order.

Ads that acquire people straight into a subscription. This is the closest paid ever gets to genuinely buying lifetime value, because the recurring relationship is baked into the very first purchase. For a consumable hero with a natural reorder cycle, a subscribe-and-save offer run to cold prospecting, handed off to a Recharge subscription and a replenishment flow, acquires the repeat at the point of acquisition. Rheal Superfoods runs a clean subscribe-and-save at 20% on its consumables with skip and swap kept easy, the kind of model a small brand can switch on today. Wild Nutrition (a B Corp, est. 2013) does the more mature version: 20%-off monthly supplement subscriptions with flexible skip and modify and subscriber consults, which is a sensible north star for where this can go. The guardrail is the same in both cases: gate how much you scale on gross-margin, cohort-based payback, somewhere around 2.5:1 to 4:1 on contribution margin with a sub-six-month payback, not a hopeful forecast off thin early data.

A fourth, lighter case is worth a mention: running your best customer and creator content as paid. It won’t show its loyalty payoff on the ad’s ROAS, because that gets captured downstream in flows and repeat. Tower 28 Beauty is the relatable version of this, where one honest creator clip lit up a hero SKU. Be straight about it though, a Hailey Bieber mention co-drove that particular spike, so the lesson is the mechanic (real content, run as disclosed paid, with the repeat measured downstream), not a guarantee you’ll catch the same lightning.

4. How to actually measure and budget paid

If you take one operational thing from this guide, take this: stop letting the ad platform mark its own homework. Judge paid on three honest numbers instead. Blended MER, which is simply your whole revenue divided by your whole ad spend across everything. Contribution margin, so you know what’s left after product and shipping, not just top-line return. And new-customer cost, so you can see whether you’re buying growth or re-buying people you already had. At sub-£1m scale you can’t run a formal geo-lift test, but you can run a 2 to 4 week all-on, all-off pulse and watch what blended MER actually does.

Then check the settings that quietly decide all of this. Exclude existing purchasers and your hashed email list from every prospecting campaign. It’s the single highest-leverage toggle in the account. One agency benchmark puts the saving at roughly 15 to 25% (MHI Growth Engine, 2025); treat that as directional rather than a promise for your account, but the move itself, not paying to re-reach people you already own for free, is hard to argue with. Cap retargeting at around 15 to 25% of your Meta budget, one tight cart and checkout-abandon set, not a sprawling warm-audience build. And track owned revenue against paid revenue as a trend over time. Rising means each acquisition is compounding into an asset you own. Flat means you’re renting traffic and calling it growth.

One last sequencing rule, because it’s the one that saves the most money. If your email and SMS together are under roughly 25% of revenue, that’s your cheapest next gain, not more retargeting. Klaviyo’s 2025 benchmark puts email at around 27% of ecommerce revenue on average (Klaviyo, 2025), so a brand sitting well below that has a gap in the channel that actually owns loyalty. Fix the flows first. “Retention ads” pointed at people a good welcome and win-back flow would have reached are the most expensive way to buy a result you already had.

What this means for your brand

1. Hire paid for what it’s good at. Use it to acquire and to choose who comes in. Let your owned channels do the keeping. Don’t ask a billboard to be a relationship.
2. Read the holdout, not the platform. When a retargeting campaign shows a 6 to 10x return, assume a big chunk was already yours. The honest number is whatever survives holding the audience back.
3. Spend paid where owned can’t reach. Truly lapsed non-openers, higher-value intake, and acquisition straight into a subscription. Prove each on a real cohort.
4. Exclude the people you own, and cap the warm spend. Keep purchasers and your list out of prospecting, and hold retargeting to a small share of budget.
5. Sequence launches owned-first. Klaviyo fires on the day, paid retargeting harvests 24 to 48 hours later. Don’t let the ad take credit for the email’s sale.
6. Fix owned before you buy “retention.” If email plus SMS is under about a quarter of revenue, that’s the cheaper growth, every time.

A note on the evidence

I’ve been deliberately careful with two kinds of number here. The retargeting figures (that 6 to 10x collapsing toward 1.5 to 2.5x, and the 30 to 40% that would have happened anyway) are modelled, directional benchmarks from the incrementality world (Stella, 2025; WorkMagic, 2025). They tell you the shape of the problem, not the exact size of it in your account, and the only way to know your number is to run your own holdout. The lookalike “higher AOV” claims I’ve flagged as directional too, because they rarely come with a clean causal test. Everything else (the flow share of email revenue, the creative share of incremental sales, the cost-of-acquisition trend) carries a named source and a year, and you can quote it without worrying. The exclusion saving I’d treat as directional as well, since it rests on a single agency benchmark, though the underlying move is sound. Where a figure is soft, I’ve said so, because a framework you can trust beats a benchmark you can’t.

That’s paid social, honestly. It’s a sharper tool than the “retention ads” pitch makes it, and a much more expensive one when it’s pointed at people you already own.

Madison, Mad Social

---

Sources

• Klaviyo, Email Marketing Benchmarks and 2025 Benchmark Report, 2025 (flow share of email revenue; email as ~27% of ecommerce revenue)
• Stella, 2025 DTC Digital Advertising Incrementality Benchmarks, 2025 (retargeting platform ROAS vs holdout; modelled / directional)
• WorkMagic, Incremental Attribution for Meta, 2025 (share of attributed purchases that are non-incremental; directional)
• NCSolutions, Five Keys to Advertising Effectiveness, 2023 (creative as ~49% of incremental sales)
• Rareview, The DTC Reckoning, 2025 (median DTC CAC and the ~60% five-year rise)
• ATTN Agency, iOS Privacy and Meta Ads Optimization, 2026 (ad identifier zeroed for ~75% of non-opt-in users)
• McKinsey with Business of Fashion, The State of Fashion: Beauty, 2025 (13% cite founder vs 39% product performance)
• MHI Growth Engine, Exclusion Audiences for Meta Ads, 2025 (15 to 25% wasted-spend reduction from exclusions; directional)
• Mad Social internal incrementality myth-check and paid-layer playbook, 2026 (in-platform ROAS running ~2 to 3x above incremental; reactivation case figures, directional; subscription LTV:CAC and payback guardrails)

Prepared by Mad Social. Figures are point-in-time (2023 to 2026) and attributed to the named source. Modelled, holdout and lookalike figures are labelled directional in the text and should be confirmed against your own account before you act on them.