Most advice on how to increase ROAS starts and ends with bidding. Lower your target, raise your target, switch to Maximise Conversion Value, wait a fortnight, repeat.
That advice fails for a simple reason: ROAS is not a setting. It is an output. It is what you get when traffic quality, conversion rate, average order value and cost interact, and a bid adjustment only touches one of those four.
This guide covers twelve tactics that move the underlying inputs, ordered the way we actually run them. If you are still working out what a healthy number looks like for your category, read our breakdown of what a good ecommerce ROAS is in 2026 first, then come back here.
The short answer. To increase ROAS in 2026: fix conversion tracking before touching bids, set targets from product margin rather than blended revenue, validate with a geo holdout, exclude brand traffic from Performance Max, prune the bottom of your product feed, suppress audiences that would have converted anyway, raise creative volume instead of budget, fix the feed rather than the campaign, lift AOV with bundles and thresholds, report prospecting separately from retargeting, reallocate budget on a rolling weekly window, and let email and CRO carry the conversions paid media cannot buy profitably.
First: the four levers ROAS actually sits on
Every tactic below pulls one of these four. If a tactic you are considering does not pull one, it will not move your ROAS.
- Traffic quality. Are you paying for people who intended to buy?
- Conversion rate. Do they buy once they land?
- Average order value. How much do they spend when they do?
- Cost. What did you pay to get them there?
Notice that two of the four sit on your website, not in your ad account. This is why paid media teams working in isolation hit a ceiling, and why the last tactic in this list matters as much as the first.
Fix the measurement before you optimise anything
1. Repair your conversion tracking first
Browser-side pixels lose a meaningful share of conversions to ad blockers, ITP, consent rejection and iOS privacy controls. If the platform cannot see a third of your sales, its bidding algorithm is optimising against an incomplete picture, and your reported ROAS is understated at the same time.
Server-side tracking through Meta’s Conversions API and Google’s enhanced conversions closes most of that gap. Releva reports that ecommerce brands typically lose 30 to 40 percent of conversion data to client-side-only setups, and SignalBridge’s 2026 benchmark report puts recovery from server-side implementations at up to 30 percent of previously missed conversions. Chatterbuzz reports that Meta accounts reaching an Event Match Quality score above 8.0 see materially lower CPAs than accounts below 4.0 on identical budgets.
Do this: implement CAPI with hashed email and phone alongside the click ID, turn on enhanced conversions for Google, and check your Event Match Quality weekly. Treat any score under 7.0 as a live bug.
What this looks like in practice. In one D2C account we took over, conversion tracking had never been installed at all. The 2023 spend recorded zero conversion value, so nobody could tell which ads produced sales or whether the account was profitable. We installed Google and Meta tracking through Tag Manager, added server-side tracking, and set UTM and event conventions before changing a single bid. Platform ROAS then moved from 0.26x in 2023 to 1.49x in 2024, 2.35x in 2025 and 2.54x in 2026. Be clear about what that first jump represents: a large part of it is conversions the account was always producing but could not see. The later growth is genuine gain. Both matter, and conflating the two is how tracking work gets oversold.
2. Set targets from margin, not revenue
A 4x ROAS on a 60 percent margin product is excellent. The same 4x on a 25 percent margin product loses money once you account for shipping, returns and payment fees. Optimising blended ROAS across a catalogue with mixed margins quietly pushes budget toward your least profitable SKUs, because they are usually the cheapest to sell.
This became urgent in 2026. Google changed how target-based bid strategies behave on budget-limited campaigns on 17 August 2026. Campaigns that had been overachieving their stated target now deliver closer to it, which means an account quietly outperforming a lazy target will drift back toward that target unless someone resets it. Optmyzr and Adthena both advised advertisers to review their targets before the change landed. Google’s own guidance is to reset the target to match recent actual performance; using product margin rather than a blended figure to decide what that target should be is our recommendation, not Google’s.
Do this: pass gross profit as a custom conversion value instead of revenue, then set tROAS against profit. If your platform cannot support that, segment campaigns by margin band and run a separate target for each.
3. Validate with a geo holdout before you trust the number
Platform-reported ROAS is not incremental ROAS. Both Google and Meta claim credit for purchases that would have happened anyway, most visibly on brand search and retargeting.
Estimates of the size of that gap vary, and you should treat any single figure with caution, but practitioners including Common Thread Co and Eightx put typical over-attribution in the 20 to 60 percent range. That would mean a campaign reporting 3x could be delivering closer to 1.5x incrementally. Geo holdout testing, running a channel in matched test regions while going dark in control regions, is the practical way to find out.
Do this: run one four-week geo holdout per quarter on your largest channel. You do not need enterprise tooling to start. Matched-market comparison in a spreadsheet will expose a 40 percent over-attribution problem perfectly well.
Cut the spend that was never earning
4. Exclude brand traffic from Performance Max
Performance Max will happily buy your own brand searches, traffic that would have converted at near-zero cost through organic. It inflates campaign ROAS and hides the real performance of your prospecting spend.
Google expanded PMax controls through 2026: campaign-level negative keywords, brand exclusions, first-party audience exclusions and channel-level reporting across Search, Shopping, Display, YouTube, Discover, Gmail and Maps, as covered by Search Engine Journal and PPC Land. There is far less excuse for treating PMax as a black box than there was two years ago. Our Performance Max optimisation playbook walks through the setup in detail.
Do this: apply a brand exclusion list, add campaign-level negatives for your brand and common misspellings, then compare PMax ROAS before and after. The drop you see is the honest number.
5. Prune the bottom of your product feed
In most catalogues a small share of SKUs earns the return and a long tail quietly consumes budget at a loss. Shopping and PMax will keep serving that tail because it still technically converts, just not profitably.
Do this: pull a 90-day product-level report, sort by spend, and identify every SKU spending above your break-even cost per acquisition without clearing it. Exclude them with a feed rule rather than pausing them manually, so the exclusion holds as the catalogue changes. Re-run monthly.
A note on effort. On one Shopify catalogue of 300 to 400 products that turned over weekly, with new photography and new products landing constantly, feed hygiene was a daily job rather than a setup task. Holding a 5x return through a seasonal scale-up depended on that more than on any bid change made in the same period. If your catalogue churns, treat feed maintenance as an ongoing line of work, not a project.
6. Stop paying to retarget people who were coming back anyway
Retargeting posts the best-looking ROAS in almost every account, which is exactly why it survives audits it should not. A large share of that revenue is people who had already decided.
Trimming windows and excluding your highest-intent segments will lower reported ROAS while raising real profit, which is why this tactic is unpopular and why it works. Route those customers to owned email and SMS flows instead, where the cost per conversion is a fraction of paid.
Do this: cut retargeting windows from 30 days to 7 to 14, exclude recent purchasers and active email subscribers, and hold a control group so you can measure what you actually gave up.
Raise the return, not just cut the cost
7. Increase creative volume before you increase budget
Meta’s Andromeda update, which finished rolling out at the end of 2025 and has shaped how accounts perform through 2026, shifted the platform’s optimisation away from audience configuration and toward creative. In practice the algorithm needs a wide, varied pool of assets to find pockets of demand, and it exhausts a small pool fast.
Agency analyses of Andromeda-era accounts, including Jetfuel and Skaleit, report a consistent pattern: brands testing 20 or more new ads a month materially outperform those testing fewer than ten, and the highest-performing accounts run several hundred live assets at once. Treat the specific percentages in those write-ups as directional rather than precise, but the direction is not in dispute. Our guide to Meta ads for DTC brands covers the production process.
Do this: set a creative quota, not a creative budget. Twenty new assets a month across three distinct angles beats four highly polished ones. Kill on cost per result, not on how much you like the ad.
8. Fix the feed, not the campaign
For Shopping and PMax, the product feed is the targeting. Your title is your keyword list. Weak titles, missing GTINs, thin descriptions and poor imagery cap your ROAS regardless of how the campaign is structured, because they cut how many auctions you enter at all.
Do this: front-load titles with brand, product type and the attribute people actually search on, such as size, colour or material. Fill every optional attribute. Test lifestyle imagery against plain product shots. Our UK Google Shopping setup guide covers feed structure end to end. In most accounts we audit, feed work returns more than any bid change made in the same quarter.
9. Lift average order value
This is the most underused ROAS lever, because it does not live in the ad account. AOV sits in the numerator of the ROAS calculation, so a 15 percent lift in AOV is a 15 percent lift in ROAS at identical spend and identical conversion rate. No bid change delivers that reliably.
Do this: add a free-shipping threshold set roughly 20 to 25 percent above current AOV, build two or three genuine bundles rather than random cross-sells, and test a single post-purchase upsell. Measure AOV weekly alongside ROAS. If you only watch ROAS you will miss which of the two moved.
Structure and discipline
10. Report prospecting and retargeting separately
Blended ROAS is the single most misleading number in most ecommerce dashboards. It averages cheap bottom-funnel conversions with expensive prospecting and produces a figure that describes neither.
Do this: report three lines every week, being prospecting ROAS, retargeting ROAS and blended, plus new customer acquisition cost. Make budget decisions on the first and last. A brand whose blended ROAS is climbing while prospecting ROAS falls is shrinking, not scaling.
11. Reallocate on a rolling week, not daily
Daily budget shifting is the most common self-inflicted ROAS problem we see. Every material change resets the learning phase, and an account that never leaves learning never reaches stable performance.
Do this: review on a seven-day rolling window, cap changes at 20 to 30 percent of budget per adjustment, and leave campaigns alone between reviews. If you cannot resist daily edits, restrict your own account access to read-only on the days between.
12. Let CRO and email carry what paid cannot buy profitably
There is a point where the next unit of paid traffic cannot be bought at a profit, and in 2026 the pressure is coming from conversion rates rather than cost. smec’s Market Observer, which tracks over €650M in European ecommerce ad spend, found Search CPC inflation flattened to 0 percent year on year in Q1 2026, while Shopping remained the fastest-inflating channel at around 11 percent. The more alarming number is conversion rate: European Search conversion rate fell from roughly 5.5 percent in Q4 2025 to about 4.5 percent in Q1 2026, against more than 6 percent the year before, and mobile conversion rate dropped from near 5.5 percent to roughly 2.5 percent across the quarter.
Read that carefully. Traffic is not getting much more expensive. It is converting worse. That is a site problem, and no bid strategy will fix it.
Do this: run conversion rate optimisation and lifecycle email as part of the same programme as paid, not as separate projects with separate owners. Our 30 ecommerce CRO tactics and eight email flows that drive revenue are the two places to start. A 10 percent conversion rate improvement raises ROAS across every campaign you run, at once, for free.
What this looks like in practice. On the same D2C account described in tactic one, email went from no measurable contribution to roughly a quarter of total sales, with 94 percent of that coming from automated flows rather than manual sends. Cart recovery improved 22 percent and open rates held at 68 to 70 percent as the list grew. None of that required a change to the ad account. It is revenue paid media would otherwise have had to buy at auction prices, and it lifts blended return without lifting spend.
Three things that will not increase your ROAS
- Lowering your target ROAS overnight. It restricts delivery, starves the algorithm of data, and usually costs you volume without improving profit.
- Adding more campaigns. Splitting the same budget across more campaigns fragments conversion data and keeps everything in learning.
- Chasing a competitor’s ROAS figure. Their margins, AOV, attribution window and return rate are all different from yours. The number is not comparable.
A sensible 30-day order of operations
- Week 1. Audit tracking. Fix CAPI and enhanced conversions. Establish the real baseline before changing anything else.
- Week 2. Apply brand exclusions to PMax and prune the bottom of the feed. Fastest wins in the list.
- Week 3. Rebuild targets on margin rather than revenue. Split prospecting and retargeting reporting.
- Week 4. Launch the creative quota and set the AOV threshold. Start the first geo holdout.
Do not attempt all twelve at once. If you change everything simultaneously you will not know what worked, which means you cannot repeat it.
Frequently asked questions
Tracking fixes and feed exclusions can show up within two to three weeks. Bid target changes need a full learning phase, so allow three to four weeks before judging them. Creative volume and AOV work compound over a quarter. Anyone promising a step change in a week is either reallocating budget into retargeting or changing the attribution window.
No. You can push ROAS up by cutting prospecting and spending only on retargeting and brand terms, and reported ROAS climbs while revenue falls. Judge ROAS alongside total contribution margin and new customer acquisition cost, never on its own.
POAS, meaning return on ad spend measured against gross profit rather than revenue, is the better metric for any brand with variable margins across its catalogue. ROAS remains useful as a fast directional signal and is what the ad platforms optimise against natively, so most brands run both.
It depends entirely on your gross margin and your customer lifetime value. At a 60 percent margin, 4x is comfortable. At 25 percent it is close to break-even once shipping and returns are counted. We break the benchmarks down by channel in our guide to Google Ads ROAS for ecommerce.
Where this usually goes wrong
Most brands we speak to have already tried three or four of these tactics in isolation. The reason they did not hold is that paid media, tracking, CRO and email sat with different owners who each optimised their own number. ROAS is a whole-funnel output, and it responds to whole-funnel ownership.
If you want a second opinion on where your ROAS is actually leaking, book a strategy call and we will walk your account with you. Or read more about how Engage360 handles paid, email, SEO and CRO as one system.
