Meta Advantage+ Shopping: When It Works, When It Cannibalizes
Independent geo-lift evidence shows Advantage+'s platform-reported ROAS over-states incremental ROAS by ~12 percentage points, and manual prospecting beats ASC on incremental ROAS by ~12% in head-to-head tests. The real question isn't whether to run it — it's where in the account, with what guardrails, and whether you're paying it to find new customers or to take credit for old ones.— Cody Plofker, CMO, Jones Road Beauty (Marketing Operators podcast)
01 / The TensionThe dashboard says one thing. The geo-lift says another.
One Q1 2025 incrementality study on a single account landed on a finding that surprised the room: Meta’s reported 4.2x ROAS for that account tracked closely with its modeled incremental ROAS of 4.4x (channel decomposition was modeled from a simultaneous holdout, not directly measured per-channel — a single-account modeled finding, not a portfolio-level read). Of the three platforms in the test, Meta was the one being honest with itself in the modeled decomposition. Google PMax inflated by more than 2x. Pinterest was undervalued by 3x. Meta, against expectation, was the calibrated one.
That finding doesn’t generalize. The Q1 study covered the Meta account in aggregate — manual prospecting plus retargeting. The same client started layering Advantage+ Shopping (ASC) into the mix in Q2. A year of independent measurement since then — Haus’s 640-test geo-lift portfolio, Tinuiti’s quarterly benchmarks, agency-published commentary from Common Thread Collective, and the Marketing Operators podcast — has converged on a less flattering pattern for ASC specifically. Two distinct findings sit inside the same Haus dataset and need to stay distinct: ASC’s platform-reported ROAS over-states its incremental ROAS by roughly 12 percentage points (the over-reporting gap — “for every $100 the platform reports, manual delivers ~$12 more incremental value”); and in head-to-head geo-lift, manual’s incremental ROAS comes in ~12% higher than ASC’s (the manual-vs-ASC iROAS delta). ASC wins outright in only 42% of those tests.
The instinct is to read those numbers as an indictment and shift budget back to manual. That’s the wrong read. Brands that have run “Meta off” tests through Haus, Northbeam, or in-house holdouts typically see a short-term revenue decline; magnitude varies materially by share of paid mix, brand strength, and seasonality, and the most-quoted ranges (15–25% has circulated in industry conversation, anchored to one Haus executive’s directional estimate) should be read as practitioner heuristic, not a measured cross-account average. The point is not the precise number; the point is that Meta as a channel is incremental, even when ASC as a campaign type over-reports inside it. ASC drives real lift — just less than the platform claims, and less than a competent manual setup in the majority of accounts. The right question isn’t “should we run Advantage+?” It’s where in the account, with what guardrails, and whether the operator-time savings justify the incrementality discount.
02 / The ProductWhat Advantage+ actually is in 2026.
Advantage+ Shopping launched in closed beta in early 2022 and rolled out broadly that September. The pitch: upload creative, set a conversion goal and an existing-customer cap, set a budget. Meta returns spend and attributed revenue. Almost everything in between — creative-to-placement mapping, audience composition, interest signals — is opaque.
The product has expanded since. By February 2025, Meta unified the setup so advertisers no longer toggle “manual vs. Advantage+” — the AI applies across Sales, App Promotion, and Leads as the default, with an Opportunity Score (0–100) rating setup quality. In March 2026, Meta consolidated budget controls further, moving from ad-set budgets to campaign-wide optimization, and shipped product-set optimization (SKU-level), a Buy Now button, performance-based creator commissions, and Manus AI assistance across Ads Manager. The directional theme is consistent: Meta takes the algorithmic levers; the operator gets creative surface area as the only real lever left.
Advertisers still control budget, country, age (in some objectives), creative pool, catalog/product set, and the existing-customer cap. They don’t control detailed interest targeting (removed for Advantage+ Sales Catalog in March 2025), retargeting-audience exclusions (effectively gone), placement-level bidding, or most of the optimization signal.
Two pieces of infrastructure matter for 2026 performance. Andromeda, Meta’s December 2024 retrieval engine on NVIDIA Grace Hopper Superchips, decides which ads even enter the auction — Meta reports +6% recall and +8% ad quality on selected segments at 4x prior efficiency (Meta-reported; internal model-vs-prior-model comparisons, not incremental-revenue lift against a no-ad counterfactual). GEM (Generative Ads Recommendation Model), rolled out Q2–Q4 2025, is the LLM-scale foundation model layered on top — Meta-reported +5% conversion lift on Instagram and +3% on Facebook Feed, again as new-model-vs-prior-model A/B uplifts, not incrementality measurements. Both increase ASC’s dependency on signal quality. Brands without clean Conversions API setups are starving the model and falling further behind every quarter.
03 / The Outside ViewWhat non-Meta sources show — mixed methodology, one direction.
Meta’s own numbers — +22% ROAS for advertisers enabling Advantage+ creative AI, +9% lower CPA, +5% conversion lift on Instagram and +3% on Facebook Feed from GEM — are vendor-published and should be read accordingly: they describe outcomes inside Meta’s own attribution stack, not incrementality against a no-ad counterfactual. The non-Meta picture below is a mix of independent geo-lift (Haus), platform-reported agency benchmarks (Tinuiti), agency-published commentary (CTC), and practitioner accounts. Methodology grade varies — Haus is the only multi-account incrementality dataset in the table.
| Source | Sample | Headline finding | What it means |
|---|---|---|---|
| Haus, “Lessons from 640 Incrementality Experiments” (2025) — independent geo-lift | ~640 geo-lift tests, ~18 months, avg advertiser $14M/yr Meta spend | ASC won 42% of head-to-head tests vs. manual. Platform ROAS 2.4% higher for ASC; geo-lift incremental ROAS 12% higher for manual. Separately, ASC’s platform-reported ROAS over-states its incremental ROAS by ~12 percentage points — “for every $100 the platform reports, manual delivers ~$12 more incremental value.” | Strongest published evidence of two related but distinct findings: (1) ASC’s dashboard inflates incremental returns by ~12 percentage points vs. its own incremental measurement, and (2) manual still beats ASC on incremental ROAS by ~12% head-to-head. ASC drives lift — just less than it claims, and less than manual in 58% of cases. |
| Tinuiti Q1–Q3 2025 Digital Ads Benchmarks — platform-reported agency benchmark | Same-client retail; >$4B under management | ASC retail spend share climbed from ~19% (two years prior) and ~24% (a year prior) to a 38% Q1 2025 peak, then fell to 35% Q2 and 27% Q3 after “unexplained” Q4 2024 over- and under-delivery. | Platform-reported, not incrementality. One of the largest independent U.S. agency portfolios quietly pulled budget out of ASC after a multi-year climb — a peak-and-pullback, not a steady-state collapse. |
| Common Thread Collective, “Meta Advantage+ in 2026” — agency-published commentary | Agency portfolio, DTC-weighted | Recommends treating ASC as a scaling environment, not a base layer, with creative diversity as the surviving lever after March 2026 budget consolidation. | Agency recommendation, not measured finding. Useful as a directional read on practitioner consensus; weight accordingly. |
| Foxwell Digital account audits — practitioner audit set | Audited brand accounts at scale | “1% floor” for existing-customer cap — Meta has stated the AI cannot guarantee zero retargeting overlap. | The cap’s existence is Meta’s tacit admission that ASC, left alone, spends disproportionately on existing customers. |
| Marketing Operators podcast (Plofker, Rekuc) — practitioner accounts | Jones Road Beauty, Ridge | Plofker: Advantage+ “developed in a way that it can very easily become less incremental.” Jones Road’s MMM rebuild flagged spending “beyond the point of efficiency” on Meta. | Practitioner concern that ASC absorbs spend faster than its incremental contribution justifies. |
| AdExchanger interview, Ben Yahalom (True Classic) — practitioner account | Single account-level commentary | ASC reached ~40% of total Meta spend within months of adoption “by going broad by default.” | The default ASC posture concentrates spend rapidly — the same dynamic the cannibalization debate is built on. |
Meta’s vendor numbers and the outside numbers don’t disagree about whether ASC works. They disagree about how much of what it claims is real. Meta measures attributed ROAS against a no-Advantage+ baseline inside its own attribution stack. Haus measures geo-holdout incrementality against the whole business. The gap between those two questions is exactly the cannibalization story.
04 / The Cannibalization MechanicWhy the algorithm structurally over-claims.
The technical complaint is specific and consistent. ASC’s optimization signal favors people most likely to convert in the next 1–7 days — a window that biases delivery toward three groups: existing customers who’d repurchase anyway, warm retargeting pools already converting through email and direct, and branded-search-equivalent intent on Meta’s expanded inventory (people who’d have typed your name into Google).
None of those are zero-incremental. There’s always marginal lift from showing up in front of a warm audience at the right moment. But the marginal lift on a converting-anyway customer is much smaller than the platform reports, because the platform credits the full conversion to the ad. The 12-percentage-point Haus over-reporting figure is those biases compounding across 640 tests — and Plofker’s Jones Road MMM rebuild is the practitioner version of the same finding: spending “beyond the point of efficiency” on Meta because Advantage+ was driving returns from existing customers rather than net-new acquisitions.
What the over-reporting looks like in CFO terms. Use the Haus reading carefully — there are two distinct findings and they translate to dollars differently. The “~12 percentage points” figure is the platform-vs-incremental ROAS gap on ASC specifically (i.e., reported ROAS overstates incremental ROAS by about 12 points on the multiplier scale). The “~12% relative” figure is the manual-vs-ASC iROAS delta from the head-to-head subset. Worked example for the over-reporting gap: a brand at $200K/month with platform-reported 4.2x ROAS and ASC-incremental ROAS of ~3.0x is generating ~$600K/month of incremental revenue, not the $840K the dashboard credits — about $240K/month of attributed-but-not-incremental revenue, or roughly $2.9M a year, depending on the actual incremental ROAS your geo-test produces. Run the math on your own numbers; the published Haus figures set the magnitude, not the per-account answer.
Meta’s product design tacitly acknowledges the bias. The existing-customer cap is a control Meta added so advertisers could push the algorithm toward new customers; the inference that Meta added it because the algorithm “spends disproportionately on existing customers” is plausible but is reading intent into product design — treat the cap as a useful lever, not as a Meta admission. One Top Growth Marketing case study — vendor-published agency commentary, n=1, not a benchmark and not held to the independent-evidence standard the rest of this piece uses — reports that capping existing-customer share lifted a single brand’s new-customer acquisition rate substantially. Treat as directional. As of recent rollouts the cap is hidden or removed in some accounts, pushing operators to enforce the split through separate campaigns instead — which is the more durable structural answer regardless.
One nuance gets lost in most “ASC eats your branded search” takes. Haus’s halo data shows ASC’s omnichannel impact (+51%) exceeds manual’s (+43%). A meaningful chunk of what reads as cannibalization in single-touch attribution is actually ASC-driven discovery converting in retail or other channels and getting credited elsewhere. Pure-DTC brands with no retail exposure see worse ASC incrementality than omnichannel brands for exactly this reason — the halo has nowhere to land.
05 / When ASC WorksThe conditions under which the algorithm earns its keep.
50+ SKUs with clean feed hygiene
Accurate titles, current prices, real availability. Small catalogs starve catalog-side optimization. Product-set optimization (March 2026) makes feed quality more leveraged, not less.
A deep creative pool
ASC optimizes across the asset pool. With 4–5 ads it picks an early winner and rides it. Foxwell flags thin creative as the single most common failure mode in audits. AdExchanger’s interview with Rok Hladnik (Flat Circle) cites 100–150 active variations as a practitioner working sweet spot — a heuristic from one agency principal’s audits, not a cross-account benchmark. Most mid-market brands cannot produce that volume at quality; the directional point is “more than 4–5, fewer than the algorithm needs to discriminate” and the right number for your account depends on creative-production capacity and catalog complexity.
50+ weekly conversions per pixel event
Below this, ASC stays in learning indefinitely and burns budget. 90+ days of consistent conversion history lets ASC inherit pixel learning instead of starting from zero.
CAPI with deduplicated server-side events
Meta reports ~8% CPA improvement for Pixel + CAPI accounts. GEM and Andromeda widen this gap every quarter — brands without clean server-side data fall further behind.
Vertical filter, from Tinuiti’s benchmarks and practitioner consensus: ASC is most consistent in apparel (affordable CPMs, large catalogs, strong creative supply), beauty (high LTV justifies broad-targeting cost), and accessories. Funnel position: prospecting-plus-warm-pool hybrid with the existing-customer cap set 10–30%. Not pure prospecting. Not pure retargeting.
06 / When It BackfiresThe accounts where ASC quietly bleeds budget.
The mirror image of the conditions above:
Small catalogs, low budgets
Below ~$100/day per ASC or ~50 weekly conversions, the algorithm can’t exit learning. The brands Meta reps most aggressively sell on ASC are often the ones least equipped to run it.
HVAC, B2B, expensive durables
Meta’s optimization window is too short for sales cycles measured in weeks or months. If your buying journey is 30+ days, ASC systematically delivers to the wrong people. Manual lead-gen with longer attribution windows wins consistently.
Brands rotating 3–5 ads
Without a deep asset pool ASC has no optimization room. Practitioner heuristics (Foxwell, Hladnik) cite triple-digit active variations as the working sweet spot, with the caveat that this is one-agency audit consensus rather than a cross-account benchmark — the directional point is “well past 4–5 ads,” and the right number depends on your creative supply.
Heavy cart-abandonment dependence
Brands whose model depends on dynamic content paired to specific abandonment behaviors are better served by manual DPA retargeting. ASC’s audience expansion can include audiences your retargeting was already reaching, and creative selection is algorithmic across the asset pool — diluted CVR vs. purpose-built DPA flows, and harder to read attribution.
Two more failure modes from audits. Geographic over-fragmentation: multiple ASCs in the same country cannibalize their own auctions and confuse the algorithm. Foxwell’s rule is one ASC per country. Promotional time-sensitivity: flash sales and short promos don’t give ASC time to learn. Under a 7-day window, manual delivers faster and cleaner.
07 / The Structural RecommendationThree campaigns. Don’t make ASC do all the jobs.
The convergence across Foxwell Digital, Disruptive Digital, Common Thread Collective, Jon Loomer, and the Marketing Operators podcast is striking. The recommended account structure is essentially identical regardless of who you ask:
- One Advantage+ Shopping campaign per country for broad acquisition with a warm-pool boost. Existing-customer cap set in the 10–30% range depending on retention strength.
- One manual prospecting campaign as the test bench for ad-level creative reads. Promote winners into the ASC pool to scale.
- One manual retargeting campaign for cart abandonment, post-purchase upsell, and dynamic creative-by-audience work that ASC can’t do well.
ASC is a scaling environment, not a testing environment. Reading creative performance inside ASC is muddy because the algorithm decides what to serve to whom. Test in manual where the read is clean, then scale through ASC. It’s one of the highest-leverage structural patterns in the practitioner literature — and almost universally ignored by accounts that “let Advantage+ handle it.”
Account-structure walkthrough: cap settings, frequency benchmarks, brand exclusions
Brand-keyword exclusions. Where Meta still allows it, exclude branded-search-equivalent custom audiences — 7-day site visitors, past purchasers — above your existing-customer cap. Forces ASC to actually prospect.
Frequency caps as a cannibalization check. Foxwell benchmarks: full-prospecting campaigns stay under ~2.5x frequency over 7 days; ASCs can run 3.0–3.5x over 7 days proportional to the cap. Climbing past those marks = cannibalization.
Launch runway. Budget as if performance will be flat for 10–14 days. Optimization stabilizes around day 7+. Most common operator mistake: killing the campaign on day 4 because early reads look weak. Early reads are noise.
Catalog hygiene. Accurate titles, current prices, in-stock filters, low-margin SKUs excluded from prospecting. Product-set optimization (March 2026) makes feed quality more leveraged than before.
Consolidation vs. fragmentation. Default to consolidation. Most accounts over-fragment and starve their own campaigns of conversion volume. Fragment only when audiences are genuinely distinct.
08 / The Labor CounterpointThe honest case for ASC.
The strongest defense of ASC comes from inside the practitioner camp, not from Meta. Maurice Rahmey at Disruptive Digital and Andrew Foxwell have both made versions of the same argument: many manual accounts that beat ASC in incrementality tests are run by senior media buyers earning $150K+. Net out the labor cost and the ASC delta narrows substantially, or disappears.
Restated bluntly: ASC lets a $50K/year operator approximate the audience-creative-placement matching of a $150K/year buyer at scale. For brands without senior in-house buying capacity — which is most brands — the right comparison isn’t “ASC vs. perfectly-tuned manual.” It’s “ASC vs. what we’d actually run if ASC didn’t exist.” Under that comparison ASC wins more often than the 42% Haus figure suggests. The counter-counter, for brands with other meaningful channels: that same labor and capital can be redeployed into Google PMax prospecting, TikTok, retail media, or OOH where the incremental return may be higher per dollar.
The labor-arbitrage math also assumes you catch the inflation early. Brands without a routine MMM or geo-test cadence can run ASC as the broad-acquisition layer for 12–18 months, then discover via measurement that a meaningful share of attributed Meta revenue is repurchase-shifted rather than incremental. The recovery isn’t free: a quarter or two of CAC repricing, a board conversation about why prior-period CAC was understated, and often a forecast revision. The labor savings get paid back — sometimes with interest — when the gap finally surfaces. Operator hours are the easy cost to model; over-investing on inflated returns for a year is the harder one.
The Plofker MMM at Jones Road is a useful warning against overcorrection in the other direction: the same model that flagged inflated branded search also confirmed Meta was incremental at the topline. Net the discount down in your math; don’t gut the line.
09 / What to WatchThe next twelve months.
- Existing-customer cap deprecation. The cap is already missing from some accounts. If Meta removes it broadly, separate campaigns become the only enforcement mechanism — making the three-campaign structure non-optional.
- GEM model maturation. The signal-quality gap between CAPI-equipped and CAPI-poor accounts will widen through 2026 as GEM trains on richer data. If you don’t have a clean server-side setup, you’re losing incremental performance every week you wait.
- The next Haus update and Northbeam Incrementality. Northbeam launched automated lift testing in April 2026 with Meta as a testbed option. Paired with the next Haus refresh, this should give the first multi-account read on whether GEM and Andromeda have closed the ASC-vs-manual gap or widened it.
- Meta API consolidation toward Advantage+ structure. Meta’s API roadmap has been steadily reducing surface area for legacy campaign types in favor of Advantage+ field shapes. We have not seen a hard sunset announcement for the broader Marketing API as of this writing — operators on Triple Whale, Northbeam, or Motion should confirm integration support against current Meta release notes before assuming any specific cutover date.
Advantage+ isn’t a replacement for thinking about your account. It’s a reasonable hire — as long as you supervise its work, give it the right scope, and don’t let it take credit for revenue it didn’t actually create.
Bitcadet · April 2026
Synthesized from independent measurement and account audits across nine retail and DTC programs.
