Most Google Ads optimisation advice begins with a list.

Add negative keywords. Test new headlines. Adjust bids. Review locations. Improve the landing page. Change the campaign structure. Add audiences. Increase the budget. Try broad match. Launch Performance Max.

Every item might be sensible. That is the problem.

When everything is a possible improvement, nothing is a priority. A manager makes six changes, performance moves, and nobody can say which change caused it. The account becomes busier but not necessarily better.

At River Stone, we take a different view: Google Ads optimisation is not the practice of changing as many things as possible. It is the discipline of finding the constraint that is preventing the business goal, then removing that constraint without damaging the rest of the system.

That definition changes the work.

If the conversion tracking is wrong, better ads help the algorithm chase the wrong outcome faster. If lead quality is poor, buying more traffic creates more bad leads. If the sales team cannot handle the leads it already receives, increasing the budget creates operational strain rather than profitable growth. If the landing page converts well but impression share is low, another redesign is productive-looking procrastination.

The account does not need every possible improvement today. It needs the right improvement next.

This guide explains how we find it.

The short answer: how to optimize Google Ads in the right order

If you need the process before the explanation, use this sequence:

1. Define the business outcome, target, time frame, and non-negotiable efficiency limit.
2. Verify that conversion and revenue data are accurate enough to make decisions.
3. Build a simple metric tree connecting spend to clicks, conversions, qualified outcomes, revenue, and profit.
4. Diagnose the first meaningful shortfall across five areas: Measurement, Business, Conversion, Traffic, and Creative.
5. Separate the visible symptom from the underlying constraint.
6. Choose one high-impact intervention and state the hypothesis before changing the account.
7. Give the change enough time and data to produce a readable result.
8. Keep essential maintenance running, but do not let maintenance replace constraint removal.
9. Review the result against business economics, not Google’s optimisation score.
10. Once the constraint moves, diagnose again. The next constraint will usually be different.

This is not as exciting as applying ten recommendations in an afternoon. It is much more useful.

First, what “optimisation” should mean

Google Ads is an auction and an automated decision system, but an advertising account sits inside a larger business system.

The platform can create impressions, visits, calls, forms and purchases. It cannot decide whether a lead fits the sales team, whether the gross margin can support the acquisition cost, whether stock will remain available, whether a sales rep follows up quickly, or whether the company has capacity to fulfil another 100 orders.

For that reason, a platform-only definition of optimisation is incomplete.

We define useful optimisation as:

Improving the business outcome produced by Google Ads while respecting the company’s real constraints.

Those constraints might include a maximum customer acquisition cost, a minimum contribution margin, a required volume of qualified leads, a payback period, stock, geography, sales capacity, or cash flow. An account that reports a beautiful ROAS but cannot produce enough profit or volume is not optimised. Neither is an account that produces cheap leads nobody can close.

This is why the first question is not “What can we change in Google Ads?” It is “What outcome must the business create, and what is stopping it?”

Why common Google Ads optimisation checklists fall short

A checklist is valuable for quality assurance. It helps a manager catch a broken URL, a disapproved ad, a feed issue, an accidental setting, or a conversion action that stopped recording. We use checklists for exactly that purpose.

But a checklist cannot decide strategy on its own.

Suppose two advertisers both have a 4% landing-page conversion rate.

• Advertiser A sells an urgent local service, where comparable high-intent traffic converts at 12% and the sales team has unused capacity.
• Advertiser B sells a complex B2B product with a nine-month sales cycle, where 4% is healthy but only 6% of form submissions become qualified opportunities.

The same platform metric points to different work. Advertiser A may have a landing-page constraint. Advertiser B probably has a qualification, offer, targeting, or measurement problem. Telling both to “improve conversion rate” hides the actual diagnosis.

Lists also encourage simultaneous changes. If a manager changes match types, targets, budgets, ads and landing pages together, the account may improve, but the team has learned very little. If it deteriorates, the team does not know what to reverse.

The right role for a checklist is to tell you whether the system is healthy. The role of diagnosis is to tell you what deserves focused improvement.

Step 1: turn the business goal into an equation

“Get more leads” is not a usable goal.

Neither is “lower CPA,” “increase ROAS,” or “scale the account.” Each phrase allows someone to declare victory while missing what the company actually needs.

For example, a manager can lower CPA by pausing every campaign except branded search. The chart improves. Growth disappears.

A decision-ready goal includes four things:

• An output: profit, revenue, qualified opportunities, new customers or another business result.
• A target: the number the business needs.
• A time frame: the period in which it must happen.
• A guardrail: the cost, margin, volume or payback condition that cannot be ignored.

A lead-generation goal might be:

Produce 80 sales-qualified leads per month from Google Ads at no more than $250 per qualified lead.

An ecommerce goal might be:

Generate $60,000 in monthly contribution profit from paid search while maintaining enough stock and a contribution margin above 30%.

Then express the result as simple maths.

For lead generation:

Revenue = Leads × Lead-to-qualified rate × Win rate × Average deal value

For ecommerce:

Contribution profit = Orders × Average order value × Contribution margin − ad spend

For SaaS:

New MRR = Trials × Activation rate × Trial-to-paid rate × Average revenue per account

The equation prevents platform metrics from floating free of commercial reality. It also shows where the shortfall lives.

Imagine a company needs 20 deals per month. Its average win rate is 25%, so it needs 80 qualified opportunities. It currently receives 150 leads but only 12 become qualified. The bottleneck is not necessarily traffic volume. The account already creates leads. The dramatic loss occurs between the form and qualification.

Buying another 100 leads before fixing that gap may simply create more work for sales.

At River Stone, this is what we mean when we say campaigns should be built around unit economics rather than platform recommendations. The goal is not a higher score inside Google Ads. The goal is a healthier economic output outside it.

Step 2: verify measurement before trusting the diagnosis

Measurement sits before strategy because every later conclusion depends on it.

If a purchase tag fires twice, ROAS is overstated. If a lead form counts page views as conversions, automated bidding learns to buy visitors who load a page rather than prospects who submit. If every phone call is valued equally, a 20-second wrong-number call can look identical to a qualified sales conversation. If imported offline conversions arrive inconsistently, the platform receives a distorted sample of the customers the business actually wants.

Before changing bids, keywords or ads, answer these questions:

Are the right actions primary conversions?

Primary conversions influence bidding. Secondary actions are useful for observation, but they should not compete with revenue-producing outcomes unless that is intentional.

A lead-generation account should not optimise equally toward a brochure download, a contact form, a qualified opportunity and a sale. Those events represent different business value.

Does each action fire once, at the right moment?

Test forms, calls, purchases and booking flows. Look for duplicate tags, thank-you pages that can be reloaded, missing transaction IDs, cross-domain breaks, payment-provider redirects and consent behaviour.

Do Google Ads and the source of truth reconcile?

They will rarely match perfectly because attribution models, time zones and reporting windows differ. But the differences should be understood. Compare Google Ads with the CRM, ecommerce platform, call-tracking system or finance data. An unexplained gap is a decision risk.

Is revenue actually revenue?

Check taxes, shipping, refunds, cancellations, repeat orders and currency. For ecommerce, revenue alone may still mislead if product margins vary widely. A campaign can improve ROAS by selling more low-margin products and reduce profit at the same time.

Can the system distinguish quality?

For lead generation and SaaS, the best signal is rarely a raw form submission. Import qualified stages, closed outcomes or values when the data and volume support it. The closer the bidding signal gets to real business value, the less the platform must guess.

Measurement does not have to be philosophically perfect. It has to be trustworthy enough for the decision in front of you. If it is not, measurement is the active constraint and optimisation begins there.

River Stone field-note opportunity: Insert a short, anonymised example in which a tracking error changed the apparent winner. Include the date range, the incorrect event, how it was discovered, and how the conclusion changed after correction. Do not publish client-identifying data without permission.

Step 3: build a metric tree instead of staring at a dashboard

A dashboard shows metrics. A metric tree shows how they create the outcome.

For a simple lead-generation account:

Revenue = Impressions × CTR × Landing-page CVR × Qualified-lead rate × Win rate × Average deal value

For ecommerce:

Profit = Impressions × CTR × Purchase CVR × Average order value × Margin − ad spend

The exact model will vary, but the principle is stable: begin with the business output and work downward through volume, efficiency and value.

This helps explain why derived ratios such as CPA and ROAS are symptoms, not diagnoses.

CPA = Cost ÷ Conversions

CPA can rise because CPC increased, conversion rate fell, traffic mix changed, tracking lost conversions, branded demand declined, or the auction became more competitive. The sentence “CPA is too high” names the pain. It does not identify the cause.

The same is true of ROAS:

ROAS = Revenue ÷ Ad spend

ROAS can fall because conversion rate declined, average order value changed, low-margin products consumed more spend, tracking lost value, demand shifted, or the campaign expanded into less efficient incremental traffic. Each cause implies a different action.

Build the tree using the business’s own historical data before reaching for industry benchmarks. Use a stable period, segment blended numbers where necessary, and mark each component as healthy, uncertain or materially below what the goal requires.

Three cautions matter:

1. Averages hide extremes. A 5% account conversion rate may combine one campaign at 11% and another at 0.7%.
2. Strong metrics can compensate for weak ones. A great sales team can temporarily hide poor lead quality.
3. Revenue lags. This month’s closed deals may come from leads created months ago, while the current pipeline is weakening.

The tree is not a decorative report. It is a map for finding the first place where reality breaks from the plan.

Step 4: diagnose across five constraint areas

Once the numbers are trustworthy and the goal is clear, inspect five areas in order.

1. Measurement: can we trust the numbers?

Look for broken or duplicate conversion actions, missing values, inconsistent CRM imports, consent loss, attribution changes, untracked calls, feed errors and unexplained differences between platforms.

Measurement is a prerequisite. If it is materially wrong, other optimisation decisions become unreliable.

2. Business: can the economics and operation support growth?

Review margin, customer value, close rate, qualification, capacity, inventory, offer strength, price and sales follow-up.

This is where many “Google Ads problems” turn out not to be advertising problems. If the business loses money on the first order and has no reliable repeat-purchase economics, raising ROAS targets inside the platform does not repair the commercial model. If sales contacts leads after four days, another headline test will not recover the lost intent.

3. Conversion: can the destination turn the right interest into value?

Inspect message match, offer clarity, mobile experience, speed, proof, friction, forms, checkout, calls to action and the steps between click and revenue.

Do not ask only whether the page converts. Ask whether it converts the right people. A shorter form may increase lead volume and reduce lead quality. That can lower platform CPA while increasing cost per qualified lead.

4. Traffic: are we reaching enough of the right demand?

Review search terms, match types, negatives, geography, schedules, audience settings, product coverage, impression share, budgets and bid constraints.

Traffic has two failure modes: wrong traffic and insufficient right traffic. Cutting waste helps the first. Expanding high-intent coverage, resolving rank limitations or increasing a justified budget helps the second.

5. Creative: does the message attract and pre-qualify?

Evaluate whether ads make the offer clear, match the query, give a reason to choose the business and discourage poor-fit clicks where appropriate. For visual campaigns, review concepts and messages, not only individual assets.

Creative is not merely a CTR lever. The ad sets an expectation. A sensational message may raise CTR and send less qualified visitors. A more precise message may attract fewer clicks but produce more profitable outcomes.

The order matters. There is little value in polishing creative while the account records the wrong conversion. There is little value in buying more traffic while the offer cannot convert it. And there is no durable scaling plan if the economics fail at the business level.

Step 5: separate the symptom from the constraint

Good diagnosis moves through three layers:

Symptom → Constraint → Intervention

Consider “lead volume fell.” That is the symptom.

Possible constraints include:

• demand fell and impressions declined;
• budget became restricted;
• a target CPA became too aggressive;
• ads lost eligibility;
• search terms shifted away from high intent;
• landing-page conversion rate declined;
• the form or tracking broke;
• leads still arrived, but CRM imports failed.

Each would produce a version of the same headline problem. Only one or two may explain the current account.

The intervention comes last. If the constraint is lost impression share due to rank, relevant actions might include improving ad relevance, landing-page experience, bids or Quality Score inputs. If the constraint is a broken form, none of those actions belongs at the top of the list.

Three diagnostic mistakes appear repeatedly.

Mistake 1: treating the metric as the cause

“CPA is high” is not a cause. “Landing-page conversion rate fell after the mobile form update while CPC and traffic mix remained stable” is much closer.

Mistake 2: confusing correlation with causation

Performance fell after a bidding change, so the bidding change becomes the automatic suspect. But the same week may include a holiday, a price increase, a stock issue, a tracking change or a new competitor. Use change history, annotations and segmented data before assigning blame.

Mistake 3: diagnosing the area you prefer to work in

Media buyers find bidding problems. Copywriters find messaging problems. Developers find page problems. The account does not care what the team enjoys. Walk the entire system.

Step 6: choose the single best next move

Once the constraint is named, rank possible interventions by four questions:

1. How much could this move the business goal?
2. How confident are we that it addresses the constraint?
3. How quickly can we learn from it?
4. What is the cost and risk of being wrong?

The best next action is not always the biggest project. A quick repair to an accidental location setting may outperform a landing-page rebuild. Conversely, an account with persistent poor lead quality may need offline conversion tracking and CRM discipline rather than another round of keyword pruning.

Write the hypothesis before executing:

We believe poor lead quality is caused mainly by Google Ads optimising toward every form submission equally. If we import qualified opportunities and make that signal available to bidding, the share of qualified leads should improve without pushing cost per qualified lead beyond our agreed limit.

Then define:

• the metric expected to change;
• the guardrail that must not break;
• the audience, campaigns or pages affected;
• the observation window;
• the decision rule for keep, revise or reverse.

This is how optimisation creates knowledge rather than motion.

Step 7: let changes produce evidence

Google Ads data has delays. Conversions occur after clicks. Sales outcomes may arrive weeks later. Automated bidding reacts to new inputs. Seasonal demand changes the baseline.

If the team makes another major change before the first has had a fair opportunity to settle, the evidence becomes muddy.

There is no universal waiting period. Use the account’s conversion lag, conversion volume, sales cycle and the size of the change. A high-volume ecommerce campaign may provide directional evidence quickly. A high-ticket B2B account with ten qualified opportunities per month cannot support daily strategy changes.

Ask “Do we have enough information to make this decision?” rather than “Has a week passed?”

Useful evidence includes:

• enough relevant conversions or qualified outcomes;
• coverage of the account’s normal weekly pattern;
• completion of a typical conversion lag;
• no unrelated event that makes the comparison invalid;
• a result large enough to matter commercially.

Sometimes the correct conclusion is “not enough evidence yet.” That is not inactivity. It is protection against paying to learn the wrong lesson.

Maintenance and optimisation are different jobs

Focusing on one constraint does not mean ignoring account health.

Maintenance prevents avoidable damage. Constraint work improves the limiting part of the system.

Daily or frequent maintenance

• urgent disapprovals and policy issues;
• broken URLs or forms;
• severe spend or conversion anomalies;
• exhausted budgets where demand is time-sensitive;
• Merchant Center suspensions or major feed failures;
• tracking outages.

Weekly diagnosis

• performance against the business goal and guardrails;
• search-term quality and material waste;
• budget allocation and pacing;
• changes in traffic mix, conversion quality or product mix;
• bid-strategy health;
• progress of the active constraint.

Monthly and quarterly review

• profitability and unit economics;
• sales or CRM outcomes;
• structure and campaign roles;
• creative and landing-page learning;
• audience, market and competitive changes;
• the next constraint and improvement roadmap.

Maintenance keeps the aircraft safe. Constraint work changes where it can fly. An account can be immaculately maintained and still fail to grow.

Where Google’s optimisation score fits—and where it does not

Google describes optimisation score as an estimate of how well an account is set to perform. The score runs from 0% to 100%, and recommendations show a potential score uplift.

That can be a useful prompt. It is not the business objective.

Google does not know every margin, sales-capacity limit, inventory priority, lead-quality rule, test design or strategic reason behind an account structure. A recommendation to expand targeting or increase a budget may be appropriate. It may also conflict with the active constraint.

At River Stone, we treat recommendations as inputs to review, not instructions to obey.

For every recommendation, ask:

• Which business goal does this support?
• What evidence suggests it addresses the current constraint?
• What new risk or traffic does it introduce?
• Can we measure the result cleanly?
• Would we make this change if no score were displayed beside it?

Google allows recommendations to be applied or dismissed. Review the details, make a deliberate decision, and document the reason. Do not confuse completion of a platform task with improvement in profit.

Auto-apply deserves particular caution because strategic changes can occur without a manager evaluating their fit. Review those settings and check change history. Repairs may be low risk; changes to bidding, targeting, budgets, conversion goals or creative deserve business context and human accountability.

A worked example: high CPA is not one problem

Consider a fictional home-services advertiser.

The account spends $30,000 per month and reports 300 leads at a $100 CPA. The company needs a maximum cost per qualified lead of $400. Only 45 of the 300 leads qualify, so the actual cost per qualified lead is $667.

The surface conclusion is obvious: CPA is too high.

But lowering raw-lead CPA from $100 to $85 would not necessarily solve the business problem. If qualification stays at 15%, cost per qualified lead only falls to about $567. Better, but still above the limit.

Walk the system:

1. Measurement: Form and call tracking work, but all leads carry the same value and qualified outcomes are not returned to Google Ads.
2. Business: The sales team marks qualified opportunities reliably in the CRM. Capacity is available. The maximum acceptable cost per qualified lead is documented.
3. Conversion: The landing page converts at an acceptable rate, but the form asks almost nothing about job type, location or timing.
4. Traffic: Search terms reveal research queries, small repair jobs outside the company’s minimum project value, and service-area leakage.
5. Creative: Ads promise a broad service and do not state the type of project the company accepts.

The active constraint is qualified demand, not raw lead volume.

A coherent intervention could include:

• exclude clearly irrelevant and out-of-area demand;
• make minimum project fit clearer in ads and on the page;
• add a small number of qualification fields without creating unnecessary friction;
• import qualified opportunities into Google Ads;
• evaluate bidding toward qualified value when volume and data quality support it.

The important metric becomes cost per qualified lead. Raw CPA may rise because fewer low-intent people submit. That can still be a commercial win.

Suppose the changes produce 210 leads at a $119 CPA, but 75 qualify. Cost per qualified lead falls to $333. Someone watching only CPA would call the account worse. The business would call it healthier.

That is why optimisation must follow the economics.

The River Stone Google Ads optimisation checklist

Use this as a decision sequence, not a bag of tactics.

Goal and economics

• Is the desired output a business result rather than a platform metric?
• Is there a numeric target and time frame?
• Is there a cost, margin, volume or payback guardrail?
• Do we know the role Google Ads must play in the wider revenue plan?
• Can the business fulfil the additional demand?

Measurement

• Do primary conversion actions match the outcomes we want bidding to pursue?
• Do actions fire once and at the correct event?
• Are values, currencies and transaction IDs accurate?
• Are calls, forms, bookings or purchases tested end to end?
• Do platform results reconcile reasonably with the CRM, store or finance source?
• Can we measure lead quality, sales outcomes, margin or another deeper value signal?

Diagnosis

• Have we mapped the goal into a metric tree?
• Have we separated volume, efficiency and value?
• Have we checked segmented data rather than relying on blended averages?
• Have we named the symptom without pretending it is the cause?
• Have we checked Measurement, Business, Conversion, Traffic and Creative?
• Can we state the active constraint in one sentence?

Intervention

• Does the proposed change directly address the active constraint?
• Is there a written hypothesis?
• Is there one primary success metric and at least one guardrail?
• Is the scope small enough to interpret but large enough to matter?
• Have we recorded the change and date?
• Have we avoided bundling unrelated changes into the same test?

Evaluation

• Has the normal conversion lag passed?
• Is there enough volume to support a decision?
• Did an external event distort the period?
• Did the business outcome improve, not merely the platform score?
• Should we keep, revise or reverse the change?
• What is the next constraint now?

How AI should—and should not—fit into optimisation

AI can accelerate analysis, summarise search terms, generate hypotheses, find anomalies, organise creative themes and help document tests. It can also produce enormous quantities of confident, generic activity.

The useful boundary is straightforward: AI may assist the work; it should not replace the business context, source data, evidence or accountable judgement behind the decision.

Google’s published guidance on AI-generated content follows a similar principle for Search: appropriate automation is not inherently against its guidelines, but using automation primarily to manipulate rankings is. Google says creators should focus on original, high-quality, people-first content regardless of how it is produced.

The same standard is sensible in advertising operations. Use AI to increase the quality or speed of a real process. Do not use it to manufacture the appearance of expertise, invent client experience, or make account changes nobody has verified.

For River Stone content, that means a human expert should review technical claims, add genuine experience, link to primary sources, and own the final recommendation. A 5,000-word article is not helpful because it is long. It is helpful if the reader can make a better decision without returning to Google for the missing answer.

What to put in the optimisation log

An optimisation process becomes far more valuable when another person can reconstruct the decision six months later. A good log does not need to be elaborate. For every material change, record:

• the date and owner;
• the business goal and active constraint;
• the evidence used for the diagnosis;
• the exact account, campaign, asset, page or tracking element changed;
• the hypothesis and expected direction of movement;
• the primary metric and guardrails;
• the earliest sensible review date;
• the final decision and what the team learned.

This matters because Google Ads accounts have memory, even when teams do not. A target may look strangely conservative because it was set during a stock shortage. A negative keyword may appear unnecessary because nobody remembers the irrelevant query pattern it blocked. A campaign split may look like clutter until someone finds the regional sales-capacity reason behind it.

Change History tells you what changed inside Google Ads. It does not capture the full commercial context, website release, sales-process change, inventory event or human reasoning behind it. The optimisation log closes that gap.

It also makes honest reporting easier. Instead of presenting a collection of completed tasks, report a chain of decisions:

We identified qualification rate as the constraint. Search-term analysis showed a concentration of low-value repair enquiries, and the CRM confirmed those leads rarely became opportunities. We tightened the message and irrelevant-query coverage while keeping qualified-lead cost as the guardrail. Qualification improved; raw lead volume fell; cost per qualified opportunity moved in the right direction. We are keeping the changes and now testing whether offline qualification signals can improve bidding.

That paragraph is more useful than “added 42 negative keywords and updated three ads.” It connects evidence, action and business outcome. It also makes failure productive. If the predicted metric does not move, the team can reject or revise the hypothesis instead of quietly replacing it with another unexplained task.

Frequently asked questions

How often should I optimize Google Ads?

Monitor urgent health issues frequently, diagnose performance on a regular weekly rhythm, and review economics and structure monthly or quarterly. The right cadence depends on spend, conversion volume, conversion lag and business volatility. High volume supports faster learning; low-volume accounts need longer windows. More frequent changes do not automatically create faster improvement.

What should I optimize first in Google Ads?

Start with the business goal and conversion data. If measurement is materially wrong, fix it first. If the data is trustworthy, identify the largest constraint between the current result and the business target. Do not begin with the easiest setting to change.

Is Google Ads optimisation score important?

It is a useful review prompt, not a substitute for profit, qualified leads, revenue or another business KPI. Evaluate each recommendation against your strategy. A higher platform score does not prove a better commercial result.

Should I change several things at once?

Only when the changes form one coherent intervention or an urgent repair requires it. Unrelated simultaneous changes make cause and effect hard to interpret. For ordinary improvement work, isolate the main hypothesis wherever practical.

How long should I wait after making a change?

Long enough to cover the relevant conversion lag and collect sufficient observations for the account. There is no universal number of days. Document the expected window before the change, then avoid interrupting it without a strong reason.

Does a lower CPA always mean better performance?

No. CPA can fall while lead quality, order value, margin or volume worsens. Use the outcome closest to business value that you can measure reliably, such as cost per qualified lead, customer acquisition cost, contribution profit or profit on ad spend.

Should I automatically apply Google Ads recommendations?

Strategic recommendations should be reviewed manually against business goals, economics, current tests and risk. Check auto-apply settings and change history. A recommendation can be useful without being universally appropriate.

Can AI optimize a Google Ads account?

AI can support analysis and execution, but it lacks important business context unless that context and reliable data are explicitly supplied. Keep a human accountable for conversion goals, economics, targeting, claims, creative quality and final changes.

The bottom line

The best Google Ads managers are not the people who make the most changes. They are the people who can explain why this change matters now.

Define the goal. Verify the data. Map the system. Find the constraint. Choose one intervention. Let it produce evidence. Then diagnose again.

At River Stone, that is how we turn optimisation from a collection of platform tasks into controlled business growth: conversion tracking tied to real outcomes, campaigns built around intent and economics, and budgets increased only after efficiency is proven.

If your account is producing activity but you cannot tell what is actually limiting profitable growth, request a River Stone Google Ads audit. We will show you where the leakage is, what we would fix first, and why—before asking you to spend more.

About The Author