Meta Ads Performance Dropped? What Andromeda Changed and How to Recover in 2027

A founder in Melbourne sent us a screenshot last month. Same budget, same products, same ad sets that had been printing money since January. Then, over about ten days, cost per purchase went from A$38 to just over A$60 and stayed there. "Did Meta break something?" was the whole message.

We get a version of that question every week. A Meta ads performance drop feels like a bug. Most of the time it isn't. The system that picks who sees your ads has changed since late 2024, Meta counts conversions differently in 2026, and the auction got pricier. Old habits now quietly work against you.

What actually changed inside Meta's ad system?

Three separate changes landed close together, and they're easy to confuse. One is about how Meta chooses ads. One is about how it reports results. One is about price.

Andromeda: the retrieval step that now picks your shortlist

Before Meta ranks ads in an auction, it has to narrow the pool. Meta's engineering team explained in a December 2024 post that Andromeda, its new retrieval engine, takes "tens of millions of ad candidates" down to "a few thousand" for each person. Meta reported a 6% recall improvement and an 8% ads quality improvement on selected segments.

Ranking got an upgrade too. In November 2025 Meta described GEM, its Generative Ads Recommendation Model, and credited it with a 5% increase in ad conversions on Instagram and 3% on Facebook Feed in Q2 2025.

Why would a better system hurt your account? Because it rewards different things. In an April 2025 Business News post, Meta wrote that "the focus has shifted from niche targeting to creative diversification as the best lever" (as quoted by Jon Loomer). In February 2025 Meta also dropped its old advice to run about six ads per ad set.

The latest signal came on 28 September 2026. Marketing Brew reported agencies being told that two videos of the same person, with only the overlay text changed, would be treated as the same creative. Meta's spokesperson didn't confirm the mechanics, saying the guidance is "deliberately outcome-based (not model-specific)". And in August 2026 Meta added a "Creative diversity" column (Columns > Customize columns), marked as in development, that rates a campaign's visual variety Low, Medium or High.

The reporting changes that look like a performance drop

Some drops aren't real. In March 2026 Meta changed click-through attribution for website and in-store conversions so it only counts conversions after a link click. Conversions after likes, shares, saves and other non-link clicks moved into a new "engage-through" window (Marketech APAC, 5 March 2026; Jon Loomer, 10 March 2026). Meta said billing wasn't affected. If your reporting columns only show click-through results, your purchase count can fall with no change in actual sales.

Earlier, on 12 January 2026, Meta stopped returning 7-day and 28-day view-through data in the Ads Insights API (PPC Land). Dashboards built on those windows dropped for that reason alone.

Then there's Apple. Safari already strips click IDs such as fbclid from links opened in Mail and Messages, and Safari 26 extended Advanced Fingerprinting Protection to all browsing by default in September 2025 (PPC Land, September 2025). Less browser signal means the Pixel alone matches fewer conversions, so the Conversions API matters more each year.

The auction simply got more expensive

Meta's own earnings tell the price story. Average price per ad rose 12% year on year in both Q1 and Q2 2026, while ad impressions grew 19% and 14%.

It's October, too. Black Friday falls on 27 November this year, Cyber Monday on 30 November, and Boxing Day sales follow in the UK, Canada and Australia. CPMs climb through the holiday season every year, so a 20% CPA rise in late October needs a calmer read than the same rise in March. For the full cost picture, see our breakdown of why Facebook ad costs keep rising.

How do you diagnose a Meta ads performance drop?

Start with the maths. Cost per result is CPM divided by (CTR × conversion rate), roughly. If CPA doubled, one of those three moved. Find which before you form an opinion.

Compare the two weeks before the drop with the two weeks after (tick "Compare" in the Ads Manager date picker), then work through this table.

Then walk this order. It finds the cause fastest in our audits.

1. Rule out real-world causesDid prices, stock, shipping fees or the checkout change? Did a competitor launch a sale? Ask whoever runs the store first.
2. Reconcile with your back endCompare Meta-reported purchases with Shopify, WooCommerce or CRM orders for the same dates. If the back end is flat and Meta fell, it's measurement. Go to the attribution and tracking fixes.
3. Split CPA into CPM, CTR and conversion rateWhichever moved most tells you where to look: auction, creative or post-click.
4. Check spend distribution at the ad levelIf spend has piled onto one or two ads, and the rest share a creator, hook or layout, you have a diversification problem.
5. Read the Delivery columnLook for "Learning limited", "Creative fatigue" and "Creative limited". Opportunity score suggestions are a checklist, not orders.
6. Check signal qualityIn Events Manager, look at Event Match Quality for Purchase or Lead, and the Diagnostics tab for deduplication warnings.

If tracking turns out to be the culprit, our guide to fixing Pixel and Conversions API mismatches walks through deduplication and match quality step by step.

Why did my Facebook ads stop working? Root causes, ranked

This ranking reflects how often each cause is the main driver in accounts we audit. It's practitioner experience, not a Meta statistic.

1. Too little real creative variety

The big one. The typical account has 12 to 20 ads that are really three ideas: same creator, same product shot, same opening line, different headlines. Under Andromeda, that reads as a narrow set of options. Delivery concentrates, frequency climbs on a small pocket of people, and CPA rises.

Motion's 2026 Creative Benchmarks show how lopsided spend already is, even in healthy accounts.

The takeaway isn't "make 50 ads". Motion found volume gives you more chances at a winner while the odds per creative stay the same. Different ideas are extra tickets. Variations of one idea mostly aren't.

2. Measurement changes and weak signal

The most expensive one to misread, because people "fix" campaigns that were fine. The attribution change, the API window removals and Safari's privacy features all shrink what Meta sees and reports. If Event Match Quality for Purchase (Meta scores it out of 10) is low, the system is optimising on thin data.

3. Auction pressure and seasonality

CPM rises with nothing else moving: BFCM in the US and Canada, Boxing Day in the UK and Australia, end of financial year sales in Australia each June. You can't fix the auction, but you can plan budgets around it.

4. Fragmented structure

Meta's learning phase needs about 50 optimisation events per ad set within seven days of the last significant edit. Ten ad sets splitting 120 purchases a week means none of them gets there. You'll see "Learning limited" everywhere. See our guide to ads stuck in Learning limited for the delivery side of this.

5. Post-click problems

A slower page after a theme update, a dropped Afterpay option, an expired coupon. CTR holds and conversion rate falls. Meta gets blamed for website problems more often than you'd think.

6. Over-editing

Daily budget swings, audience changes and new ads dropped into a live ad set every other day keep resetting learning. The account never settles long enough to judge.

How to recover: a step-by-step fix with real settings

Do these in order. New creative judged on broken tracking teaches you nothing.

1. Fix the reporting view firstIn Ads Manager, open Columns > Compare attribution settings and add the click-through, engage-through and view-through windows next to each other. Compare total results with your store or CRM. Agree on one source of truth, and stop comparing click-only numbers with last year's all-click numbers.
2. Repair signalIn Events Manager, confirm Purchase or Lead arrives from both the browser and the server, with a shared event_id for deduplication. Send hashed email, phone, external ID, and the fbp and fbc parameters where consent allows. On Shopify, the Facebook & Instagram app's data sharing setting should be on "Maximum".
3. ConsolidateMerge ad sets that share an objective and country. For most accounts under about $15,000 (or £12,000) a month, one prospecting campaign with Advantage+ audience, Advantage+ placements and Advantage+ campaign budget is plenty. Our Advantage+ campaigns guide covers when to keep manual controls.
4. Rebuild creative around conceptsPlan four to six concepts that differ in angle, format and person. One founder story video, one product demo, one customer testimonial, one static comparison, one carousel of use cases, one creator-style Reel in 9:16. Each concept gets one or two executions, not ten.
5. Use Flexible format where it helpsAt ad level, Flexible format lets you upload up to 10 images or videos plus several text options and lets Meta assemble them. Use it to hold executions of the same concept, and keep different concepts in separate ads so you can read results.
6. Set sane attribution and optimisationFor most e-commerce, optimise for Purchase with the default 7-day click-through and 1-day view-through. Lead gen accounts with a CRM should push qualified-lead events back via the Conversions API and optimise for those, rather than raw form fills.
7. Change budgets gentlyMeta treats large budget changes as significant edits. Many buyers raise budgets around 20% every few days. That's a rule of thumb, not a Meta rule.
8. Wait, then judgeGive the rebuilt setup 7 to 14 days, or about 50 conversions per ad set, before calling it. Judge on back-end CPA, not on the first three days in Ads Manager.

Mistakes that make a Meta ads drop worse

Honestly, the reaction to a drop often does more damage than the drop.

• Duplicating the "old winner". A copy of the same ad competes with the original for the same people. Under Andromeda that's even less useful than it used to be.
• Narrowing targeting. Stacking interests to "reach better people" usually raises CPM and starves the algorithm. Meta's own guidance points the other way.
• Killing ads after 48 hours. Early numbers swing wildly, especially with a 7-day click window still filling in.
• Chasing in-platform ROAS after the attribution change. If you rebuilt a profitable campaign because click-through purchases fell in March, you may have broken something that was fine.
• Blaming Meta for a landing page problem. Test the page on a mid-range Android phone. Our roundup of high-converting landing page examples shows what good looks like.

How to prevent the next drop: a monitoring routine

Drops are easier to fix when you catch them in week one, not week four. This is the cadence we run for clients.

Set automated rules (Ads Manager, Rules > Create new rule) to email you when an ad set's cost per result rises more than 30% above its 7-day average, or when frequency passes a level that's high for your account. Don't let rules pause things automatically until you trust them.

Keep a change log too: date, what changed, who did it. It's the first thing we ask for, and most accounts don't have one.

If you run Google Ads too, read both together. A Meta drop alongside rising branded search can mean Meta still does the awareness work while Google takes the last click. Our comparison of Google Ads vs Meta Ads explains how the two split the job, and our Google Ads team can read both sides of that picture.

Worked scenario: a skincare brand in Melbourne

A direct-to-consumer skincare brand in Melbourne spends about A$40,000 a month on Meta and sells through Shopify, shipping across Australia and New Zealand. From January to August, cost per purchase averaged A$38. In September it climbs to A$63 in ten days, just as the team is planning its Black Friday push.

The account has seven prospecting ad sets split by interest, each with 14 to 18 ads. Two thirds of the ads feature the same creator in the same bathroom, with different text overlays. At ad level, two ads take 74% of spend. Frequency on those two is 4.1 over seven days. Every ad set shows "Learning limited".

Reconciling with Shopify shows something else. Shopify orders from Meta-tagged UTMs fell 22%, while Ads Manager's click-through purchases fell 38%. Part of the gap is the March link-click change. Event Match Quality for Purchase is 5.4, because the server events aren't sending phone numbers.

The rebuild takes two weeks. Seven ad sets become one Advantage+ prospecting campaign with Advantage+ audience. The team produces five concepts: a dermatologist explaining the active ingredient, a customer's 30-day routine diary filmed on her phone, a static price comparison with salon facials, a carousel of three skin concerns, and an unboxing Reel by a different creator in Brisbane. The server events start sending hashed phone numbers, and EMQ rises to 7.6.

By week five, the back-end cost per order is A$45. Not the old A$38, because pre-Black Friday CPMs are real, but a long way from A$63. Spend on the top two ads drops to 48%.

The drop had three stacked causes. Fixing only one would have looked like it "didn't work".

Lead-gen accounts, like a dental group in Manchester or a clinic chain in Toronto, hit the structure and signal problems hardest because volumes are lower. Our dentist PPC guide covers lead quality, and our PPC trends for 2026 explain why automation rewards better inputs.