How to Improve Google Ads ROAS Without Optimizing the Wrong…

I previously wrote a straightforward list of ways to improve Google Ads ROAS: choose the right bidding strategy, refine targeting, use ad extensions, test ads, improve landing pages, and keep optimizing.

Those are still useful operating areas. The list begins too far downstream.

Before trying to improve ROAS, I want to know whether the account is measuring a return the business should actually optimize. If the conversion value is incomplete, duplicated, disconnected from margin, or based on an unqualified lead, the bidding system can improve the reported ratio while moving the business in the wrong direction.

The first ROAS optimization is therefore not a bid change. It is a definition audit.

What Does Google Ads ROAS Measure?

The platform relationship is simple:

ROAS = reported conversion value ÷ advertising cost

If Google Ads records $50,000 in conversion value against $10,000 in ad cost, reported ROAS is 5.0, often displayed as 500%.

But “conversion value” is not guaranteed to mean collected revenue. Depending on the account, it might represent:

ecommerce transaction revenue;
revenue before refunds, discounts, shipping, or tax adjustments;
gross or contribution profit;
an assigned value for a lead;
predicted lifetime value;
offline sales imported from a CRM;
several conversion actions added together.

Google defines conversion value per cost as total conversion value divided by the cost of ad interactions. Its conversion-reporting documentation explains the calculation, but the advertiser still defines which conversions and values enter it.

That definition boundary is where many ROAS problems begin.

Decide Whether ROAS Is the Right Constraint

ROAS is useful when the values being reported are comparable and the organization has a real return requirement. It is incomplete when leadership actually needs to control:

contribution profit;
customer acquisition cost;
new-customer volume;
qualified pipeline;
payback period;
capacity or inventory;
lifetime value;
incremental revenue.

A campaign can raise ROAS by spending less on uncertain prospects and concentrating on branded or returning demand that was already likely to convert. The ratio improves. Incremental growth may shrink.

It can also raise revenue ROAS while profit declines. Products with different margins should not be treated as equally valuable merely because their checkout totals are similar.

Use ROAS as one operating constraint, not a synonym for profitability.

Audit the Numerator Before Changing Bids

List every conversion action included in the campaign’s primary goals. For each one, record:

the trigger and counting method;
whether it is primary or secondary;
the source of its value;
when that value becomes available;
attribution model and conversion window;
whether it represents a new or existing customer;
whether refunds, cancellations, duplicates, and offline outcomes are reconciled;
whether the action is used by bidding.

For lead generation, trace sample records from the click through the form, CRM, qualification, opportunity, and realized sale. A lead-value model should reflect observed progression and economics. Assigning every form fill a large static value can create excellent platform ROAS and terrible sales conversations.

For ecommerce, reconcile transaction IDs and values with the order system. Check duplicate purchase events, currency handling, discounts, canceled orders, returns, tax, shipping, and margin differences.

Google’s conversion-value guidance supports reporting revenue or profit-related values. The important part is choosing the value the business wants the bidding system to pursue and maintaining that definition over time.

Set a Business-Based ROAS Boundary

Historical platform ROAS is a reference, not the business target.

For a simplified ecommerce example, break-even revenue ROAS can be approximated from contribution margin before advertising:

Formula // inspect the definition

Break-Even Revenue ROAS

Simplified relationship Break-even revenue ROAS = 1 ÷ contribution margin rate

Use contribution margin before advertising. This is a boundary—not automatically the business target.

If $100 of revenue leaves $40 after variable product, fulfillment, payment, and service costs—but before advertising—the contribution margin rate is 40%. The simplified break-even revenue ROAS is 2.5, or 250%.

That is not automatically the target. The business may need room for overhead, returns, growth, cash flow, repeat behavior, or profit. Lead generation needs a different chain connecting lead value to qualification, close rate, deal margin, and lag.

Write down the assumptions. Do not copy a target from the platform recommendation or another advertiser.

Choose the Bidding Strategy After the Value Audit

My older guidance listed Maximize Conversions, Target CPA, and Target ROAS as interchangeable options to consider. Each strategy asks the system to solve a different problem.

Maximize Conversions: pursue conversion volume within the budget.
Target CPA: pursue conversions around a cost constraint.
Maximize Conversion Value: pursue reported value within the budget.
Target ROAS: pursue reported conversion value while trying to meet an efficiency constraint.

Google’s current bidding guide recommends value-based strategies when conversions have different business values. That only works if those differences are reported reliably.

Target ROAS is not a command to create profit. Google says it predicts conversion value from the values supplied through conversion tracking and adjusts auction-time bids to pursue the target. Google’s Target ROAS documentation also warns that an overly high target can limit traffic.

This creates a familiar tradeoff:

increasing the target may improve efficiency while reducing scale;
decreasing the target may admit more auctions and value while lowering the ratio;
changing values or goals can make the historical target temporarily misleading;
thin or delayed conversion data can make value optimization unstable.

Evaluate total conversion value, cost, profit, and customer volume alongside the ratio.

Refine Search Intent Before “Audience Targeting”

The earlier version emphasized demographics, interests, and behaviors. Those controls can matter, but Search campaigns begin with the demand expressed in queries and the matching system that admitted them.

Start with:

actual search terms;
keyword and match behavior;
brand, nonbrand, competitor, and returning-customer separation;
geography and service eligibility;
device and time patterns when they change the experience;
audience observation data;
landing-page and offer alignment;
negatives and routing rules.

The goal is not to remove every lower-ROAS segment. Some segments are acquiring new customers while branded demand is harvesting existing awareness. Compare their roles before consolidating them into one average.

Review marketplaces exposed another version of the same attribution problem in a separate SaaS implementation. Analytics received clean campaign parameters from the marketplace clicks, but the CRM did not preserve that source on the resulting leads. When those prospects later returned through search, the CRM credited the outcome to the last visible Google interaction. The marketplace spend feed worked; the outcome-attribution handoff did not. As a result, the source was undercounted in both the CRM report and the aggregate model that depended on it.

That is why I do not call a channel measurable merely because its spend imports successfully. Reconcile at least one path from click or placement through analytics, form or product event, CRM record, and final eligible outcome. If the identity breaks between systems, ROAS can reward the channel that collected the last observable touch instead of the channel that introduced the prospect.

A costly search term may be irrelevant. It may also expose broken tracking, weak qualification, delayed sales, or a poor page. Use the negative keyword sweep before blocking demand merely because the visible ROAS is low.

What I Found After Separating Brand From Nonbrand

I worked on a global SaaS account serving English-language markets where paid search appeared to be one of the strongest conversion channels. The first warning was visible before I opened a report: it was nearly impossible to search for the company name without seeing a paid ad.

That did not prove the ads were wasteful, so I built a GA4 event to track clicks to the product login and used heat-mapping evidence to inspect the same behavior. Among the branded paid clicks that could be tracked through this path, more than 94% went to login. The denominator matters here: this was not 94% of every branded click, and a login click does not identify every visitor perfectly. It was still a strong enough operational signal to show that the campaign was paying for a large amount of existing-customer return behavior rather than only acquiring new prospects.

The account was therefore not simply generating efficient new demand. Branded campaigns were also paying to bring existing users back to the site. Looking only at channel or source would have hidden the distinction; the campaign dimensions, destination behavior, GA4 event, and heat map made it visible.

The first intervention was measurement, not bidding. I cleaned up the UTM structure, removed the large unassigned bucket, and separated brand from nonbrand activity so the acquisition and return paths could be inspected independently. Brand terms had also leaked into campaigns intended to capture nonbrand demand, so I excluded those queries instead of assuming the campaign label described the traffic accurately. I then compared the February and April attribution paths for both leads and customers who converted to paid.

Path position for converted customers	February	April
Early touchpoints	1.96%	8.82%
Mid touchpoints	4.89%	14.41%
Late touchpoints	93.15%	76.77%

That shift did not prove that one channel suddenly caused more sales. It showed that the cleaned setup was preserving more of the journey before the last interaction. The earlier report was overwhelmingly late-touch because the account could not reliably see how many customers first arrived.

The second intervention was spend allocation. I ended the branded campaigns and excluded brand terms that were still entering nonbrand campaigns. Paid-search leads declined afterward, which looked negative inside a paid-search-only report. Organic leads grew enough that paid and organic leads combined increased by approximately 30% while the business spent less. The cumulative customer-acquisition cost fell, and payback and ROAS improved when the two acquisition paths were evaluated together rather than defending the paid-search subtotal.

The organic change continued beyond the immediate channel reclassification. Over the following months, the site grew from roughly 600 first-page keywords to more than 1,200. I have seen similar brand-to-organic reallocations precede compounding organic growth in other accounts, and I have a working hypothesis that stronger branded organic click-through behavior may contribute a useful search signal that later supports nonbrand visibility.

That is a hypothesis, not the result of a controlled ranking experiment. The account was also undergoing measurement cleanup, campaign changes, and broader marketing work. I would not attribute the keyword growth to branded organic CTR alone or present the sequence as proof of a Google ranking factor.

R. Garcia’s field observation: Branded search can make ROAS look excellent while charging the business to reacquire attention it already owns. In this account, a purpose-built GA4 login-click event showed that more than 94% of the trackable branded paid clicks in the inspected path went to login. Clean the attribution path, separate brand from nonbrand, and identify returning customers or login behavior before calling that revenue incremental.

The result was a more credible marketing mix, not merely a prettier paid-search ratio. Organic search became the leading late touchpoint for leads, more early and mid touchpoints survived into the converted-customer path, and the team could see where paid search was introducing demand versus collecting it at the end.

What This Case Still Does Not Prove

I would not turn this into a universal rule to pause branded search. A brand campaign may still earn its place when competitors are bidding aggressively, the organic result is weak, the ad controls an important offer or destination, or a properly designed incrementality test shows that paid coverage creates additional profit.

The comparison also included measurement cleanup, brand exclusions, and other campaign changes during the same period. The exact observation window for every outcome was not preserved in this publishable record. That makes it strong operational evidence and a useful diagnosis, but not a clean causal experiment. The next level of proof would be a geo holdout, time-based brand test, or another controlled comparison that measures total conversions and profit—not only the conversions Google Ads kept for itself.

Read the Account as a System, Not a Collection of Ratios

A later cross-market engagement reinforced the same lesson under a different operating pattern. After measurement repairs, query cleanup, and controlled account changes, advertising spend stayed roughly flat while new-customer volume rose by about one quarter and cost per customer fell by about one fifth.

Those are intentionally rounded, privacy-safe comparisons rather than raw account totals. More importantly, they are not proof that one bid setting caused the result. The work combined several interventions, markets matured at different speeds, and some cohorts had more time to convert than others.

The useful evidence was the agreement among several views:

platform spend and click data;
governed conversion actions;
downstream customer outcomes;
like-for-like reporting periods;
market and campaign maturity;
change logs and live configuration readback.

One intervention also produced an honest tradeoff: a landing-page migration improved governance and consistency while short-term paid-search efficiency worsened. That did not make the migration automatically wrong or the campaign automatically bad. It created a new question: did the cleaner operating system justify the acquisition cost, or did the page need another iteration?

That is what an ROAS review should preserve. A polished ratio is less useful than a traceable explanation of what changed, which populations are comparable, and which cost moved somewhere else.

Use Ads and Assets to Qualify, Not Just Attract

“Ad extensions” are now called assets in Google Ads. Sitelinks, callouts, structured snippets, images, calls, locations, prices, and other assets can give a searcher more information and more ways to act. Google recommends using assets relevant to the business goal.

More clicks are not automatically better. Ads and assets should help the right person understand:

what the business actually offers;
who it is for;
price or qualification boundaries when useful;
location, timing, delivery, or availability;
meaningful product or service differences;
the correct next page or action.

An asset that routes a visitor to a useful category page can improve the path. An asset that attracts curiosity without purchase intent can add cost while weakening ROAS.

Test the complete message package. Responsive search ads, display campaigns, and video campaigns are not simply interchangeable ad formats inside one Search test. They serve different inventory and can play different roles in the customer journey.

Improve the Landing Page Around the Decision

The landing page must continue the promise made by the query and ad. Google describes landing-page experience in terms of relevance, usefulness, navigation, and whether the page meets the expectations created by the ad. Google’s landing-page definition is a useful floor, not a complete conversion program.

Inspect:

message and offer continuity;
mobile speed and interaction stability;
price, terms, availability, and proof;
form length and validation;
checkout or scheduling friction;
accessibility and readability;
consent behavior and tracking loss;
what happens after the submitted lead enters the CRM;
whether the page filters or encourages the intended customer.

Do not optimize the form-completion rate while ignoring qualified rate, close rate, refunds, or customer value. A page can generate more conversions and lower business ROAS if it makes the wrong action easier.

Protect ROAS From Attribution and Customer-Mix Confusion

Google Ads can assign conversion credit across eligible ad interactions using its configured attribution model. Google’s attribution documentation explains that data-driven attribution distributes credit using account data, while last click assigns credit to the final clicked ad and keyword.

Attributed value is not the same as incremental value. It answers how credit is distributed under the model, not whether the sale would have happened without the advertising.

At minimum, separate or annotate:

brand and nonbrand demand;
new and existing customers;
prospecting and remarketing;
online and imported offline outcomes;
short and long conversion lags;
revenue and contribution profit;
observed value and modeled value.

This is also why ROAS should not be compared casually across channels. Each platform may claim value using different windows, identity signals, view-through rules, and models.

Make One Governed Change at a Time

Continuous optimization does not mean continuously changing everything.

For each material change, record:

the hypothesis;
primary business metric and guardrails;
campaign and customer scope;
start date and expected conversion lag;
original and changed settings;
tracking or value changes during the period;
decision date and result.

Use campaign experiments when the platform supports the comparison. Preserve enough time for delayed conversions and bidding adaptation. Avoid changing budgets, targets, conversion actions, ads, pages, and targeting simultaneously and then crediting the outcome to whichever change is easiest to explain.

A Better ROAS Improvement Sequence

Use this order:

Reconcile value: prove what enters the numerator.
Define the business boundary: decide whether revenue ROAS, profit, CAC, or another outcome governs the decision.
Check acquisition mix: separate brand, existing customers, remarketing, and genuinely incremental demand.
Choose bidding: match the strategy to trustworthy goals and sufficient data.
Improve query admission: govern match behavior, routing, exclusions, and eligibility.
Improve ads and assets: make the offer clearer and qualify demand earlier.
Improve the destination: remove friction without lowering customer quality.
Test and wait: respect conversion lag and isolate material changes.
Reconcile downstream: compare platform value with CRM, orders, margin, and finance.

My Updated Rule for Improving ROAS

My earlier checklist was a useful account-optimization list. The updated rule is more demanding:

Improve the value definition first, then ask the bidding system to improve the ratio.

The fastest way to raise reported ROAS may be to narrow spend toward demand the business already owns. The most valuable decision may be accepting a lower short-term ratio to acquire more profitable new customers. The account cannot choose between those outcomes until the marketer defines them.

The SaaS example above is why I now treat brand separation, return behavior, and path completeness as part of a ROAS audit. It is field evidence, not a promise that one brand policy or bid target improves every account.

Use the paid search systems hub to connect ROAS with queries, conversion tracking, attribution, and CRM outcomes. Use the blended CAC calculator when advertising spend alone is too narrow, and the lead-to-revenue funnel calculator when a form value needs to be challenged against qualified pipeline and closed revenue.