How to find out why Meta Ads ROAS dropped

Step 1

Confirm the drop is real

Do not diagnose from a single day. Compare the last 7 days against the previous 7, and against the trailing 28-day average, pulling spend, revenue, purchases and ROAS together. A one-day dip on a conversion-optimised account is usually noise or attribution lag, not a problem.

• Pull daily data rather than one aggregate, so you can see when it turned — the inflection date is the single most useful fact in the whole investigation.
• Judge an ad set on at least 3–5 conversions per week. Below that you are reading noise, and the honest answer is that there is not enough signal yet.
• Do not make optimisation decisions on an ad set still in learning. The gate we use is spend of at least 2× AOV or 4× CPA, whichever is higher.

The trap: rate metrics are not additive. You cannot average daily ROAS, CTR, CPM or frequency to get a weekly number — you have to query at the grain you intend to report. Averaging row-level rates is one of the most common ways a drop gets invented that was never there.

Step 2

Split spend from revenue

ROAS = revenue ÷ spend. Only one of those needs to move. If revenue held flat and spend rose, you have an allocation question. If spend held and revenue fell, you have a conversion or tracking question. Write down which one it is before going further — it halves the search space immediately.

• Spend rose, revenue flat → look at budget changes, newly launched campaigns, and CBO reallocating toward an ad set that converts worse.
• Spend flat, revenue fell → split revenue into purchases × AOV. A purchase-count drop and an AOV drop have completely different causes.
• Both moved → take the larger percentage change first.

Check this before you trust the revenue number: Meta returns several overlapping purchase actions — omni_purchase, purchase, offsite_conversion.fb_pixel_purchase, onsite_web_purchase and more. Use exactly one. We check omni_purchase first and fall back to purchase. Summing them double-counts revenue and inflates ROAS, which means the drop you are chasing may be last month's number having been wrong rather than this month's being bad.

Step 3

Decompose into the rate metrics

Now walk the funnel and find the broken link. Each metric fails for a different reason, so the one that moved tells you which team owns the fix.

• CPM up — you are paying more for the same reach. Auction pressure, a narrower audience, a placement shift, or a cost cap the algorithm cannot fill at.
• CTR down — the ad is landing worse. Creative or audience.
• CPC up without CPM or CTR moving — usually a mix shift between placements rather than a real change.
• CVR down — traffic arrived and did not convert. Landing page, offer, checkout, stock. Rarely the ad itself.
• AOV down — discounting, product mix, or a promotion that shifted the basket.

A cost cap combined with rising CPM has two different causes that need separating before you act: either the algorithm cannot find conversions at your cap price, or the original audience is saturating and delivery is leaking into worse segments. Frequency, reach trend and audience size tell you which.

Step 4

Find where it actually happened

Account-level numbers hide the answer. Drill down one level at a time — campaign, then ad set, then ad — and quantify each child's contribution to the parent. The goal is to move from "ROAS dropped 20%" to "three campaigns account for the entire decline."

• Sort ads by spend descending and keep the ones making up the first 90% of spend. Those are the ads that matter; the long tail cannot move an account-level number.
• Always filter to active entities. The insights API returns anything that spent in the window, including things you have since paused — which quietly pollutes every aggregate.
• Check at least two breakdowns — placement, device, or age and gender. Out-of-target delivery eating budget is common when Advantage+ audience expansion is on.
• Hold segment conclusions to at least 100 impressions and 3 conversions. Below that, say so rather than recommending on it.

A pattern that looks like a problem but isn't: under CBO, one ad set holding 80%+ of the budget is usually the algorithm working correctly — it found the best converter. Do not pause the dominant ad set because smaller siblings look better on small samples; they generally fail at scale.

Step 5

Investigate the creative properly

Creative fatigue is real but over-diagnosed. It has a specific signature, and a single metric is not it.

• Fatigue is frequency rising AND CTR falling, sustained over 14+ days. Both, together, over time.
• High frequency on its own is not fatigue. A small retargeting pool runs at high frequency permanently and converts fine.
• For video, split the funnel: hook rate is 3-second views ÷ video plays, hold rate is thruplays ÷ 3-second views. A hook problem and a hold problem need different fixes.
• Check whether one new creative is dragging the average, rather than the whole set decaying.

The thresholds we work to: hook rate at or above 40% is good and under 25% is poor; CTR at or above 1.25% is good and under 0.65% is poor. These are our working numbers from the accounts we run, not an industry standard — calibrate against your own account's trailing 90 days before treating them as targets.

Step 6

Rule out tracking and attribution

Not every ROAS drop is a performance drop. If the funnel metrics look broadly normal but revenue fell, suspect measurement before you touch budgets. This check saves more wasted optimisation than any other step here.

• Pixel and CAPI firing — compare purchase event volume against its own daily baseline.
• Attribution window changes. Moving from 7-day click to 1-day click will drop reported ROAS without a single thing changing in the account.
• View-through share. A campaign whose ROAS is mostly view-through is over-credited, and the true incremental number is lower.
• Two campaigns targeting the same audience can double-count conversions between them.
• Advantage+ Shopping aimed at existing customers takes credit for purchases that would have happened anyway.

If the account optimises for a non-purchase event — COMPLETE_REGISTRATION, LEAD, ADD_TO_CART — do not convert that count into revenue. One common configuration routes real purchases through the registration event because a third-party profit tracker sends them that way, which means the event means the opposite of what the name suggests. Confirm what the event represents in this account before reading anything into it.

Step 7

Check what changed in the account

Finally, look at the change log over the window. Most ROAS drops have a human cause sitting in plain sight, and finding it takes a minute.

• Budget changes and newly launched campaigns
• New creatives entering rotation
• Audience, targeting or exclusion edits
• Bid strategy or cost cap changes
• Landing page, offer, pricing or stock changes
• Anything that reset learning