Why is my ROAS dropping even after switching creative production tools?
ROAS usually keeps dropping after a creative tool switch because the tool was not the broken part, and the switch itself reset things that were working: proven ads were paused, tracking and naming broke, and the new tool's output looks more alike than you think. Break ROAS into its four parts (cost per thousand impressions, click-through rate, conversion rate and order value), find the one that moved, and fix that. Creative only controls some of them.
What does ROAS actually depend on?
Return on ad spend is revenue divided by spend. Written out, it is the product of four things:
ROAS = (click-through rate × conversion rate × average order value × 1,000) ÷ CPM
Each part has a different owner:
CPM (cost per thousand impressions) is set by the auction: competition, season, audience size and how well the platform rates your ads.
Click-through rate is mostly creative: does the ad earn attention and a click from the right people?
Conversion rate is mostly the landing page and offer, plus whether the ad's promise matches what the page delivers.
Average order value is pricing, bundles and offer design.
A creative production tool can move click-through rate, and through promise-match it can nudge conversion rate. It cannot fix an auction, a slow page, a price rise or broken tracking. So the first question is not "which tool?" It is "which of the four moved?"
What does switching tools quietly reset?
Teams often see ROAS fall in the weeks after a switch for reasons that have nothing to do with the new tool's quality:
Proven ads were retired. Migrating to a new workflow often means pausing the old library and launching fresh ads. New ads start without delivery history, and the old winners that carried the account stop spending.
Naming and tracking broke. New export formats, file names and landing URLs drop UTM parameters or your naming convention, so results stop joining to the creative behind them. Some "drops" are partly reporting.
The output is more uniform. Template-driven tools produce variants that share a layout, voice and structure. Meta's retrieval system picks a few thousand candidates from tens of millions of ads for each request, and Meta says it is designed to surface a more varied set (Meta Engineering). Fifty similar ads do not buy fifty times the reach.
Volume spread spend thin. More ads on the same budget means less spend per ad, slower reads and more ads that never get enough delivery to prove anything.
The promise drifted. A new tool writes in its own voice. If the ads now promise something your page does not show, click-through rate can rise while conversion rate falls, and ROAS drops.
The tool optimises for its own score. Many tools score creatives before launch. Check what the score predicts. A model that predicts clicks will happily favour curiosity bait that does not buy.
How do I find the part that moved?
Work through these in order. Each step rules out a cause before you spend money on the next.
Check measurement first. Is the pixel firing on every new landing page? Are server events still flowing? Meta's Conversions API sends events from your server alongside the pixel, which protects you when browser tracking fails. Compare ad platform revenue against your store or CRM for the same weeks. If the platform shows a drop your store does not, you have a reporting problem.
Check CPM account-wide. If CPM rose across every campaign, including ones the new tool never touched, the auction moved. Look at seasonality and competitor activity before you blame creative.
Split the funnel. Compare click-through rate, conversion rate and order value for the four weeks before the switch and the four after, using sums over each period rather than averages of daily rates.
Click-through down: the new creative is weaker or too similar.
Click-through steady or up, conversion rate down: promise mismatch, page or offer.
Order value down: the new ads attract a different buyer, or bundles and upsells changed.
Check distinctness. Lay the last twenty ads side by side. If you can describe most of them with one sentence, you are running one idea many times.
Check structure and changes. Look at the account's change history for budget jumps, new campaigns and edits made during the migration. Restructuring often coincides with a switch and resets learning.
What should I change first?
Fix the cheapest confirmed cause first:
Measurement: repair tracking, naming and UTMs before judging anything else. You cannot optimise what you cannot see.
Retired winners: relaunch the old top performers alongside the new work. Let results, not the tool change, decide what stays.
Uniform output: brief different angles, not different templates. Give each ad a distinct argument, buyer and proof.
Promise mismatch: rewrite ads so the first claim the buyer sees is the first thing the page proves. Keep a list of claims you can support and check every ad against it; Nagent's HOOK does this with a claim gate, covered below.
Thin spend: run fewer ads with more budget each, and set a kill condition for each before launch.
Auction or offer: hold creative steady and work the bid, budget, page or offer instead.
What should I ask a creative tool vendor before switching again?
Most tools are good at making assets. Ask what happens after the asset ships:
Does it read my ad results, and from which platforms?
How does it decide a creative won, and against what baseline: my own account or an industry benchmark?
Does it separate a weak idea from a weak execution of a good idea?
Can it check the claims in an ad against proof I supply?
Does it keep a record of what won so the next brief starts there?
What does its predictive score actually predict?
How is it priced, and what happens to cost when I make more variants?
How do the main creative tools differ on learning from results?
Per their own sites, checked 24 October 2026:
Omneky: generation plus Insights that show spend, ROAS, CTR and conversions against the period before, with an AI Creative Score (Omneky Insights). Plans from $29 a month (Lite) to $249 a month (Pro), billed monthly (pricing).
AdCreative.ai: generation plus Creative Insights from connected ad accounts and creative scoring (AdCreative.ai). Starter from $39 a month billed monthly.
Pencil: multi-model generation with testing before you scale spend, and launch and tracking across Meta, TikTok, YouTube and others (Pencil). Core from $14 a month.
Motion: analytics rather than production, with attribution integrations such as Northbeam and Google Analytics on its Pro tier (Motion pricing). Starter is $999 a month for up to $50k of monthly spend.
Foreplay: Lens analytics with benchmarks for CPA and ROAS against other advertisers, plus an ad library (Foreplay Lens).
If your ROAS problem is in CPM, conversion rate or order value, none of these will fix it alone. If it is in click-through rate and variety, the ones that read your results and tell you why something won will help more than the ones that only make more.
Where does Nagent fit?
We make Nagent. Two of its Marketing team leads touch this problem, and it is worth being exact about which does what.
HOOK, the creative strategy lead (Beta), works on the creative parts. It writes distinct angles, each with the claim it rests on; its GATE seat checks every claim against the proof in your brand pack, which keeps the ad's promise in line with what you can show; and its PULSE seat ranks every ad against your own account's ads over 90 days, flags fatigue by a fixed rule and separates a weak angle from a weak execution. It reads a connected Meta account nightly, or any platform's export as a .csv or .xlsx file.
NIA, the performance marketing lead, works on campaigns: on Google Ads she drafts campaigns from a brief and proposes live changes to budgets, bids, keywords and ad copy; on Meta ads (Beta) she can pause, re-budget or resume a campaign, or create an empty one that arrives paused. Every live change waits for a person to approve.
What neither does: HOOK cannot pause an ad or change a budget. NIA does not fix your landing page, your price or your tracking. Work is metered in credits; a campaign build is about 40 credits, and the pricing page has the plans. Nagent is multiplayer AI for growth: people and AI agents in one shared context, delivering the number. The people still own the decisions.
Frequently Asked Questions
Why are my AI generated ad creatives not improving ROAS?
Usually because they improve volume, not variety or promise-match, and ROAS depends on CPM, conversion rate and order value too. We cover that pattern in why more AI ad creative has not moved your ROAS.
Should I switch back to my old creative tool?
Not before you diagnose. Relaunch your old winners next to the new work and compare them on the same budget. If the old ads win, the issue is briefs and variety, not the tool.
How long should I wait before judging a new creative tool?
Long enough for each new ad to get meaningful delivery and for conversions to settle, usually several weeks. Judge on settled data and on ads that actually received spend.
Can a better creative tool fix a high CPM?
Only partly. Strong, distinct creative can earn better delivery, but account-wide CPM rises are usually auction or season.
Want a second pair of eyes on a ROAS drop? Book a demo and we'll show how HOOK reads your account.