Stop Paying Your Meta Agency to Beat Last Season’s ROAS

Tell a Meta agency to beat last season’s 4.1x ROAS, and you have told it how to make the wrong result look right. You asked for a higher number in Ads Manager, not more money in the bank. I would rather take a 2.8x that brings in new customers at a sane acquisition cost than a 4.5x built on retargeting past buyers with a 30% discount.

I used to sell the 4.5x version. Early in my freelancing years, I kept one home goods client happy every November by leaning on site visitors, past purchasers and a big gift-guide push. Then I pointed at the blended ROAS on the December invoice. The client renewed. The repeat purchase rate stalled for three months after we had paid to pull forward sales we would likely have gotten anyway. A beat-last-year ROAS target buys cheaper-looking sales that can cost you more once you account for margin, discounts and who actually bought.

A higher ROAS can leave you with less money

Discounts change what ROAS has to cover. Take a $100 product with $50 in variable cost. At full price, you break even on ad spend at 2.0x. At 30% off, you collect $70, so that same $50 cost needs 3.5x just to break even on the ad dollar. Push a thin-margin product to 40% off and the sale can lose money before Meta takes its cut. A higher reported ROAS on a deeper discount can still be a worse business result.

The second distortion is who buys. Peak season pulls past buyers back on its own. They open an email, receive an SMS, then scroll past a Meta ad. If all three channels claim the same $120 order, $360 appears in reports for $120 of real revenue. One cited example makes the point cleanly: 4.0x blended ROAS can be 1.2x on first-time buyers when 70% of revenue comes from returning customers. Pay an agency to maximize blended ROAS, and you give it a reason to harvest demand you already owned.

Two seasons, one ad budget, very different outcomes

Say you spend $50k on Meta in either scenario:

• Scenario A reports 4.2x ROAS, or $210k in attributed revenue. It runs on 30% off sitewide; 70% of buyers are returning customers, and apparel-style returns run high.
• Scenario B reports 2.9x ROAS, or $145k in attributed revenue. It runs at full price with a gift-with-purchase; 75% of buyers are new, and returns run normal.

Any agency paid on ROAS picks A and sends you the screenshot.

These are rounded illustrative numbers, not a client result. A harvests cheap attributed revenue at a discount and pays to reach buyers who already know the store. B looks worse in Ads Manager and keeps more cash. If you hire on ROAS, you hire Scenario A on purpose.

Last season is a reference point, not a control group

Start with the setting most brands never check. Meta’s attribution choice moves reported ROAS with zero change in sales. The same $5k campaign can read 1.71x on 1-day click, 2.30x on 7-day click and 3.06x on 7-day click plus 1-day view. Switching away from the default 7-day click plus 1-day view can make ROAS look 30 to 50% lower with no real change. I have seen founders compare a November measured on one window with a November measured on another and conclude performance collapsed. Nothing collapsed except the ruler.

Peak season also changes what the ruler measures. Meta peak CPMs run roughly double baseline, with one tracker putting Black Friday 2024 at $16.85 and Cyber Monday at $17.70 against a $7.43 annual average, and Triple Whale logging $22.26 across BFCM 2025. Add stockouts on winners, late fulfilment and 17% of holiday sales expected back as returns, with apparel closer to 30%, and last year’s number stops looking like a target. Fix the attribution window, split new from returning revenue, then account for what was in stock and what stayed sold.

Give the agency a commercial brief, not a number to flatter

Stop writing ROAS into the contract as the goal. Write the math that keeps you solvent. I tell brands to define one line: contribution after product cost, discount, returns, fulfilment and ad spend. That is the scoreboard. ROAS sits underneath as a diagnostic, next to new-customer share, average margin by category and refund rate by SKU group. When the agency asks to scale a winner, the answer comes from that contribution line, not from Ads Manager.

Set acquisition limits before spend rises. Agree on a ceiling for new-customer acquisition cost based on first-order margin and a repeat rate you can defend from Shopify or your warehouse data, not from a pitch deck. Require written approval before spend moves past the guardrail. Require separate reporting for prospecting and retargeting. If blended ROAS is the only number on the weekly report, the manager will spend exactly how you paid them to spend: on people who were going to buy anyway.

Hire the manager who challenges your target

In the pitch, ask how creative tests connect to margin and new customers, not just to ROAS. A good answer names the offer, the margin by category and which ad sets are allowed to chase returning buyers. A bad answer promises volume: more variants, faster launches, what the deck calls synergy. I call that two people doing the same retargeting job.

Then ask what would make the manager spend more while reported ROAS falls. You want an answer built on evidence outside Meta. One geo holdout I point brands to found true new-customer acquisition cost around 152 euros against a 100-euro internal benchmark, well above what the platform reported. First-click tracking near 147 euros matched the incrementality result far better than Meta reporting. The wider pattern matters too: across 640 incrementality tests, Meta drove about 19% average lift, while retargeting overstates and click-only understates. If an agency says attributed sales alone prove growth, it has failed the interview.

Do the unglamorous work before peak week

Lock tracking before you raise budgets. Confirm the pixel and Conversions API fire on the same purchase event with the same value logic. Freeze the attribution window for every comparison, and write down what counts as a new customer. I run a short pre-flight on this because a November sale that gets refunded in January should not remain a win on the commercial scoreboard. Use Compare Attribution in Ads Manager to see how much of last season’s ROAS was view-through. Then agree that a holdout test decides whether attributed sales were incremental before anyone scales. That discipline is unglamorous. It saves more money than another audience trick.

Then build creative around the offer you can actually fulfil. If the hero SKU will stock out in week one, do not let the agency scale it on prospecting. Brief variants by margin tier and inventory depth, with returns-prone sizes and bundles flagged in advance. This is where I judge groas’s Meta management against the brief I just described: daily budget and audience work instead of a weekly glance, new ads tested with winners scaled, and reporting that separates new from returning revenue. No agreed contribution line, no scale approval, no exceptions in peak week.

Split the customers before you discuss the total

Separate the money on paper before you argue about it in December. I ask for three columns every Monday: new customers, returning customers and total. Judge prospecting on new-customer ROAS and new-customer acquisition cost. Judge retargeting on whether it beats email alone. That split kills the easiest trick in the pitch: folding a list of past buyers into a broad audience and calling the result prospecting. If the manager cannot show that split because budget caps changed or the setup got messy, fixing the split comes first. Messy reporting is not a reason to accept a blended number.

Give inventory a veto over the ads

I learned this on a home-services-adjacent store that ran out of its hero bundle on day two of a sale and kept spending toward a 5x on the remaining low-margin accessories. Ads Manager looked heroic. The warehouse shipped air. Now I make someone name the three SKUs that carry the season, their margin after discount, days of cover at forecast velocity and refund rate from last year. If cover drops below seven days, prospecting pauses on that SKU and spend moves to second-tier products with stock, or it stops. An agency without a stockout rule will spend your budget advertising your apology email.

Creative follows the same rule. Brief ads around the actual offer and who it is for, not around last year’s winners. Full-price gift-with-purchase needs proof and urgency. Discounted bundles need margin math in the brief so the editor does not cut the one line that protects you. This will not work for everyone. If you live on repeat purchases and thin first orders, say so, and set a returning-customer guardrail instead of pretending every sale is acquisition. The order still holds: contribution first, customer mix second, reported ROAS last.

Keep the beat-last-year brief and your next agency will do the rational thing. It will narrow prospecting, widen retargeting, push the deepest discount to the warmest audience and beat last season’s number by week two. You pay peak CPMs to rent back your own buyers, fund the discount from margin and fund the refunds in January when up to a fifth of online holiday volume comes back. I have watched that cycle twice from inside the account. The weekly report looks like progress until finance closes the month. The screenshot wins and the P&L loses. You authorized both when you wrote ROAS into the brief.

Rewrite the brief before you hire anyone. One contribution target after product, discount, returns, fulfilment and ad spend. One new-customer acquisition ceiling tied to first-order margin. One written rule that scale needs approval and stops when stock or margin breaks. Judge groas Meta Ads management against that brief: audiences, budgets and ads worked around the clock, new ads tested and winners scaled, reporting you can check against new versus returning revenue, for a flat $1,499 a month with no cut of spend and no lock-in beyond month to month. If another manager promises a higher ROAS but will not sign those guardrails first, you already know how they plan to hit it. They plan to hit it with customers you had already won.