When ROAS disappoints, the reflex is to touch the budget: scale winners harder, cut losers faster, shift spend between platforms. Sometimes that is right. But budget moves only redistribute the efficiency you already have. The levers below improve the efficiency itself, and none of them require spending more. One caveat before the list: these tactics interact with your niche, price point and funnel shape, so treat them as a checklist to test against your own data, not a formula that works identically everywhere.
1. Feed the algorithm better conversion signals
Ad platforms optimise toward the events you send them. If those events are incomplete, blocked by ad blockers, cut off by app switches, missing the conversions that happen off-site, the algorithm learns from a biased sample and buys you more of the wrong people. The highest-leverage fix on this list is usually invisible in Ads Manager: make the conversion feed accurate and complete before optimising anything downstream of it.
Concretely, that means server-side event delivery with proper deduplication and strong match quality, so the platform sees the real buyers, including the ones who converted a week later in another app. Our Meta Conversions API guide covers the mechanics; FlowTracker's attribution layer automates them. Teams are routinely surprised how much delivery improves when the platform finally sees the full conversion picture.
2. Exclude the junk placements
Automatic placements spread your spend across every surface the platform sells, and those surfaces do not perform equally for every offer. Low-quality in-app inventory in particular can generate clicks, and even shallow conversions, that never become customers. The tell is a placement with unusually cheap clicks and near-zero downstream revenue.
Run a placement breakdown against your deepest conversion event, not against clicks or leads. Exclude the surfaces that consistently spend without producing buyers, then re-check quarterly, because placement performance drifts as the platform's inventory changes. Do not slash to a single placement out of superstition; you pay for over-narrowing with higher auction prices. Cut what the data convicts, keep the rest.
3. Put creative iteration on a cadence
Creative fatigue is the quiet ROAS killer: the same audience sees the same ad until frequency climbs, CTR sags and CPMs effectively rise. The fix is not one great ad, it is a production rhythm. A workable baseline for a small team:
- Ship a small batch of new variants on a fixed schedule, weekly or biweekly, rather than when someone feels inspired.
- Vary one meaningful element per variant, hook, format, angle, proof, so wins teach you something.
- Watch frequency and CTR trend per ad, and retire creatives on evidence, not affection.
The cadence matters more than the volume. A steady trickle of tested variations beats a quarterly creative "refresh" because it keeps the learning loop short.
4. Optimise on the right funnel event
There is a real trade-off between event volume and event quality. Optimise on a shallow event (lead, registration) and the algorithm gets plenty of signal but learns to find form-fillers. Optimise on the deepest event (purchase, FTD) and the signal is truer but may be too sparse for stable delivery. The practical approach is to work down the funnel as volume allows: start where you get enough weekly conversions for the platform to learn, then move the optimization event deeper as accounts and audiences mature. If your funnel runs through broker integrations or chat channels, this only works when those deep events actually reach the platform, which loops back to lever one.
FlowTracker reports every campaign against real deposits and orders, not proxy events, so these levers stop being guesses.
5. Write kill rules before you need them
Losing ad sets do their damage in the gap between "this looks bad" and "fine, pause it". Decide the thresholds in advance, in writing: at what multiple of your target cost per acquisition, over what window, does an ad set get cut? A rule like "pause anything that has spent three times target CPA with zero conversions" is crude, but it is applied consistently, which beats a sophisticated judgment applied whenever someone gets around to it.
Two nuances keep kill rules honest. First, respect conversion lag: if your typical click-to-purchase delay is four days, judging an ad set on 48 hours of data kills winners. Size the window to your funnel's real lag. Second, kill at the right level, cut the ad set or creative, not the whole campaign, so you preserve what the campaign has learned.
6. Bid toward lifetime value, not first order
Two campaigns with identical first-purchase ROAS can have wildly different real returns if one attracts one-time buyers and the other attracts repeat customers or heavy depositors. Where the platform supports value-based optimization, feeding actual customer value, including repeat orders and redeposits, teaches it to find high-LTV customers rather than cheap first conversions.
The honest caveats: value optimization needs meaningful volume of valued events to work, and it needs back-end data plumbed into the ad platform, which most teams have never connected. If your revenue arrives through Stripe, Shopify or broker systems, an integration layer that forwards those values automatically is what makes this lever usable at all. Even before you change bidding, simply reporting campaigns on LTV instead of first-order value changes which "winners" you scale.
7. Monetise the audience you already paid for
ROAS has a numerator too. Every lead your ads produced who did not buy is inventory you already own: no new spend required to reach them again on owned channels. Segmented follow-up, WhatsApp and Telegram sequences for engaged non-buyers, win-back sends for lapsed customers, turns yesterday's spend into today's revenue and lifts the return on the original campaigns. The economics of paid messaging channels have their own rules, covered in our WhatsApp pricing guide, but done with decent segmentation, broadcast campaigns are usually the cheapest revenue in the building.
The honest summary
None of these levers is magic, and their ranking depends on your situation: a team with broken tracking should start at lever one and stop reading; a team with clean data and stale ads should live in lever three. What the seven share is a direction, more accurate signals in, more disciplined decisions on top, and more value extracted per lead, which is what improving ROAS without new budget actually means. Fix the measurement first, then let every other lever compound on top of numbers you can trust.