Google Ads device-performance audits for agencies: compare results and investigate the right campaign

Start with the client's business question

A device-performance review should answer a real account-management question. The client may want to understand which device produces most recorded value, why mobile acquisition costs increased or whether a planned campaign should keep its existing reach. Those questions require different conclusions. Begin with the intended decision rather than treating every device table as a request to disable its weakest row.

Agree on the business metric before ranking devices. An ecommerce account may care about conversion value relative to cost. A lead-generation account may care about acquisition cost while separately investigating lead quality. The device with the largest spend is not automatically the problem, and the one with the most conversions is not automatically the most efficient.

The agency should record the selected period, account population and conversion definition alongside the question. These details make the result explainable when the client forwards it to a colleague. A clear brief also reduces unnecessary follow-up reads because the specialist knows which additional evidence could change the decision.

Define the useful outcome in advance. It may be a campaign investigation, a website task, retention of an intentional device strategy or preparation of a supported setting change. The review does not need to end in a mutation. Understanding a result and assigning the right next task are valid account-management outcomes.

Keep the client meeting focused on that outcome. A long table can invite several unrelated explanations, but the first decision should follow the original brief. If the question is why acquisition cost changed, identify the observations needed to investigate that change. If the question is simply where value comes from, report the distribution clearly. The agency should not expand a descriptive request into a broad account restructuring project.

Understand the current account device report

The current ppctor account reader requests device-segmented metrics from the Google Ads customer reporting resource. It selects impressions, clicks, cost, conversions and conversion value for the requested date range. The helper folds the returned rows into a metric total for each returned device key.

The audit card is informational. It shows a performance table and a summary rather than applying a device-quality score or supplying a direct automatic fix. A visible review status in the internal finding should not be interpreted as a scored account defect; the card hides that status in its informational presentation.

This account report is a starting population, not a campaign-specific diagnosis. It does not identify which creative, offer, landing page or bidding strategy caused a difference between device rows. A client asking about a named campaign needs its relevant breakdown rather than a relabelled account total.

The practical agency benefit is one consistent initial view. It establishes where the recorded results appear across devices and highlights questions worth investigating. Keep that usefulness attached to the declared source and selected period so a compact summary does not become an unsupported causal recommendation.

Set a comparable reporting period

The reader applies the requested date segment using the configured account time zone. The agency should state the actual selected dates when interpreting results. A rolling completed-day period and a calendar month are different populations, even when a Slack conversation uses a short informal phrase.

Use the same interval for the device rows being compared. A current partial day beside completed historical days can answer a live monitoring question, but it is different from a completed-period efficiency comparison. Do not change the reporting window silently to obtain a tidier conclusion.

A second period needs the same metric and scope if it is intended as a comparison. Account changes, promotions, stock availability and website releases can make two periods operationally different. Keep those facts in the review rather than assuming a device setting caused every difference in the later table.

Record configuration observation time separately from historical metrics. A campaign's current bidding strategy or status does not prove that the same configuration applied throughout the selected reporting interval. If historical configuration is important to the decision, obtain relevant evidence instead of inferring it from a current resource read.

Preserve the account population

The current account device query does not impose an enabled-campaign filter. It requests the selected customer-level reporting population for the interval. The agency should not describe the result as only today's enabled campaigns unless it deliberately obtains a report with that requested scope.

Paused campaigns can still matter to a historical interval. Excluding them because they are paused now can change the meaning of a comparison with the account total. Current eligibility and recorded historical activity are separate facts, and the client should know which one the analysis uses.

Account aggregation can combine several campaign types and objectives. One device row may contain a different mix of acquisition strategies from another. The device name remains a valid segment label, but it does not make the underlying campaign mix identical.

When the client narrows the question, preserve that narrowed population throughout the follow-up. A named campaign, selected channel or specified group should receive its own relevant evidence. Do not compare one narrow row against an account-wide counterpart and present the difference as though both had the same scope.

Interpret the returned device categories

The presentation recognises mobile, computers, tablet, connected TV and other device keys. It gives these known categories a readable order and preserves additional returned keys afterward. The visible categories therefore reflect the provider rows available for this request rather than a compulsory fixed checklist with invented values.

A missing category is different from a returned category whose metrics are zero. The table does not manufacture every possible device row. If the client asks why a category is absent, establish the returned population and relevant reporting evidence rather than claiming it had no activity from absence alone.

Device categories describe the reported segment, not the physical identity of a unique person. A user can interact through more than one device, and this account table is not a deduplicated customer journey. Avoid describing its clicks or conversions as distinct people merely because the rows have device names.

Keep unusual categories visible when they are returned and relevant. Hiding a category to make the report look simpler can change the apparent account mix. A readable summary can emphasise major rows while retaining the actual total and explaining the requested scope.

Calculate totals from their underlying metrics

The account table builds its All row by summing the returned base metrics. It then calculates the displayed efficiency measures from those totals. This is different from averaging the devices' displayed ROAS or CPA values, which can give small and large device populations the same weight.

For example, an invented pair of rows has 100 of cost and 400 of conversion value on one device, and 900 of cost and 1,800 of value on another. The combined ratio is 2,200 divided by 1,000, or 220%. It is not the simple average of 400% and 200%. These illustrative values are arithmetic examples, not client results.

Apply the same discipline to acquisition cost and conversion rate. Sum the appropriate numerator and denominator before computing the total. A row with a small number of conversions should not determine the account CPA simply because its displayed CPA is unusually high.

Use the total as a summary of the declared returned population. It does not certify coverage of a different account, period or conversion definition. Reconcile scope before treating a difference between another report and this All row as a calculation defect.

Display rounding can also make a hand-added total differ slightly from the underlying result. The table calculates with its retained metrics before formatting the cells. A specialist should reconcile the base values when precision matters rather than repeatedly adding rounded displayed ratios. The ordinary client output should remain readable, with the exact calculation preserved internally for any genuine numerical investigation.

Read commerce value and ROAS correctly

For accounts classified as ecommerce, the current table shows conversion value and ROAS beside clicks, impressions, cost and conversions. ROAS uses conversion value divided by cost. The product displays that ratio as a percentage, so a ratio of three is shown as 300%.

Conversion value is the selected Google Ads conversion-value metric. It should not automatically be renamed purchase-only revenue, accounting revenue or profit. The actual contributing conversion actions and assigned values matter when the client asks what that figure includes.

A device with positive cost and no recorded value can show zero achieved return for that selected evidence. A device without cost has no defined return on spend, and the formatter displays a dash rather than inventing zero. These are different business conditions and should remain distinct in the explanation.

Keep currency context explicit when discussing money. The current compact formatting uses a dollar symbol, but that presentation alone is not proof of the account's currency. Confirm the real context before translating a value into a financial recommendation, especially when an agency manages accounts across several markets.

Read lead-generation CPA and conversion rate

For other account classifications, the current table shows conversion rate and CPA. Conversion rate uses conversions divided by clicks, while CPA uses cost divided by conversions. These calculations describe the selected recorded metrics; they do not establish that every conversion is a distinct qualified lead.

Fractional conversions can appear in the source and remain part of the totals. The presentation limits their displayed precision rather than rounding each underlying contribution into a unique person. A client seeing a fractional total needs the measurement context, not an explanation that invents a fraction of a physical customer.

When there are no conversions, CPA is undefined rather than zero. The dash communicates that the denominator does not support the ratio. A row with cost and no conversions deserves review, but the report should not fabricate a finite acquisition cost merely to make ranking easier.

Recorded conversion rate and sales quality can differ. If the business wants to know whether mobile leads close less often, establish the corresponding downstream evidence. Do not infer close rate, spam rate or customer value solely from the device's Google Ads conversion count.

Explain what the denominator represents when a client compares reports. This table's conversion rate is based on its selected conversions and clicks. A website session conversion rate can use a different population and denominator. Those figures can both be valid while answering different questions. Preserve the requested definition instead of renaming one as the other to make a conversation comparison appear simpler.

Establish the conversion definition

The device reader selects the standard conversions and conversion-value metrics. It does not separately request every conversion-action name or category in this compact table. A question about purchases, calls, leads or a named action therefore needs the corresponding conversion scope to be established.

Do not infer that all recorded conversions are purchases because the account sells products. An account can track several primary actions with different values. The agency should preserve the user's intended metric rather than silently change a general conversion-value report into a purchase-only claim.

If the user asks to compare a named Ads conversion with a particular Analytics event, identify both requested definitions and obtain compatible evidence. The account device table by itself does not establish the event pairing. A generic total cannot be relabelled as that named event merely because its count looks plausible.

Keep goal selection separate from data-quality repair. The user may deliberately want all relevant primary conversions and accept the account's existing definitions. The review should answer that selected metric without inventing a deduplication project or removing legitimate actions just to make two systems agree.

Separate volume and efficiency rankings

The compact summary identifies the device that drove the most value or conversions by absolute volume. That is a scale statement. It is not a ranking by highest ROAS, lowest CPA, best conversion rate or strongest lead quality. The agency should preserve that distinction when explaining the sentence.

A large device row can produce the most total value while having a lower return ratio than a smaller row. A small efficient row may also have limited scale. Neither observation independently determines where another unit of budget will produce the best marginal result.

Ask which ranking the client wants when the word best is genuinely ambiguous. Often the account type and established conversation already provide the intended metric. Use that context, then state the selected metric clearly instead of making the owner repeat the whole brief.

Rank only defined values within the requested population. A zero-denominator row is not automatically the worst finite CPA, and a missing result is not a measured zero. The conclusion should identify the metric and relevant evidence rather than turn every exceptional cell into an arbitrary ordered score.

Distinguish missing data and zero outcomes

The performance report has a loading state while its separate report phase is in progress. Loading is not a final account result. The completed report phase records failures by individual report so a device read failure does not silently erase successful channel or location results.

The unavailable mapping associates the device finding with its report dependency. A failed provider request should be described as unavailable evidence, not a confident empty device population. This protects the client's decision when another account report succeeded but the device read did not.

A successful empty device map causes the informational card to be hidden. It does not generate a comprehensive zero-valued table for every device. The agency should not use a hidden card as proof that all possible device categories recorded no activity.

Undefined ratios are another separate condition. The underlying cost or conversion metric can be available while the ratio denominator is zero. State that condition plainly. A retried provider read cannot make a mathematically undefined ratio meaningful unless the actual requested evidence changes.

Assess the evidence for a decision

Look at the volume behind a ratio before proposing a major reach restriction. A small number of clicks or conversions can produce a striking efficiency difference without supporting a durable account decision. The current informational card does not implement a statistical certainty test for the agency.

Choose an investigation window appropriate to the client's trading pattern. A brief promotion and a longer steady period can answer different questions. Do not impose an arbitrary universal threshold and describe it as a provider requirement when the actual business context calls for a different review.

Check whether the concern persists within relevant campaign populations. An account-level difference can result from campaign allocation rather than an inherent device problem. Further segmentation is useful when it changes the decision, but indiscriminate slicing can obscure the original question and generate unnecessary work.

Use careful conclusion language grounded in the observed result. The agency can state that mobile had higher recorded CPA in the selected interval and recommend a focused investigation. It should not claim that every mobile visitor is unprofitable or that excluding the device guarantees a better account outcome.

A decision record can identify what would change the recommendation. For example, the specialist may need the named campaign breakdown or confirmed booking-event behaviour. Naming that missing observation is more useful than a vague request for more data. It also prevents the team from collecting every available dimension while the actual question remains unanswered. The next read should have a purpose tied to the client's decision.

Investigate campaign mix before editing

An account device row can combine brand search, prospecting, remarketing and other strategies. Those populations can differ in intent and conversion value. A comparison across account devices therefore does not isolate the effect of a device setting within one campaign.

Identify which campaigns contribute to the concern when that evidence is relevant. The next useful read may be a named campaign's device breakdown rather than every available account dimension. Preserve the original period and metric so the additional evidence answers the same question.

Current campaign status and strategy help establish a proposed setting's applicability. They do not automatically explain historical results. Keep the observed performance, current configuration and proposed change as separate facts in the client recommendation.

A single campaign can warrant action while the rest of the account remains appropriate. Conversely, an account policy can be broader than one campaign. Choose the actual scope deliberately and name it in the proposal. Do not convert an account summary into a silent blanket edit of every campaign.

Keep the comparison understandable when campaign mix explains part of the result. The client does not need a technical query description; it needs to know that different campaigns contributed differently to the device totals. Show the relevant contribution or narrowed comparison where available, then explain which decision it supports. Do not conceal the scope change behind a repeated account-level headline.

Review the visitor journey separately

A device difference can motivate a website investigation, but the reporting table does not test the landing page. It does not inspect navigation, checkout, forms, consent or page rendering. A low recorded mobile conversion rate is an observation, not a verified diagnosis of a broken mobile website.

Assign a focused visitor-path task when the evidence supports investigation. The website owner can examine the relevant destination and interaction rather than a random page. Keep that task separate from a provider configuration change so the client knows which system each operator is working on.

Measurement behaviour also needs its own evidence. A form event can be missing or overcounted without the device's campaign targeting being wrong. Establish the event implementation and business definition before using the device table to prescribe a measurement repair.

If the user only wants a device comparison, provide it without automatically launching a broad website audit. If the user asks why the difference exists, collect the additional relevant observations. The agency should preserve the boundary between a descriptive report and an established cause.

Check the actual bidding strategy

Google's bid-adjustment guidance distinguishes manual device adjustments from Smart Bidding behaviour. It explains that eligible Target CPA device adjustments affect the CPA target, while several other automated strategies permit only a device opt-out adjustment. Do not assume a manual bid change has the same meaning under every strategy.

That guidance also has campaign-type eligibility boundaries. A reported device category does not establish that the campaign supports the desired control. Current campaign configuration and provider validation matter before the assistant can prepare an actual mutation.

The agency should explain the requested effect in business language. Changing a target is different from changing an auction bid directly, and excluding a device is different from a modest preference adjustment. The owner needs the correct effect before confirming, not an ambiguous instruction to optimise mobile.

A report can support an investigation without supporting a particular setting. If the intended adjustment is unavailable for the actual strategy, identify the concrete limitation and choose the next deliberate workflow. Do not substitute a strategy change merely to make the originally requested device control possible.

Retain the client's target and budget brief during that discussion. An eligibility limitation does not grant permission to alter either one. The agency can explain the available choices and request the missing decision where necessary, but the eventual prepared effect must reflect the actual instruction. A convenient tool should serve the chosen strategy rather than determine it by accident.

Resolve the setting scope and hierarchy

Device reporting scope and setting scope are different. The account table contains a device population; an eligible control can belong to a campaign or group. The proposed action therefore needs an actual resource target and relevant parents rather than the device row alone.

The linked Google guidance describes group and campaign device-adjustment interaction, including a campaign-level opt-out exception. Preserve that hierarchy when investigating a control. Reading only one child setting can miss a relevant parent effect.

The current device audit brick has no direct live fix, and the convenience catalogue does not provide a dedicated universal device editor. A schema-mediated edit must establish actual supported fields, exact resource identity, current eligibility and result checks before it can be offered as a proposal.

Use readable names and real relationships in approval output. A group belongs to its campaign; a device segment is not its parent. Internal IDs may resolve the provider target, but they should not become the client-facing headline or appear as undefined relationship text.

Follow an invented agency device review

This case is invented. Reed Harbour Media manages Maple Court Workshops. The client asks why mobile acquisition appears expensive in a completed reporting period. The agency opens the account device comparison with CPA as the stated metric and confirms the selected conversion definition.

The table shows an account-level difference, but the specialist does not immediately recommend disabling mobile. A focused campaign breakdown establishes that the account rows contain different campaign mixes. The agency preserves that observation rather than attributing the entire difference to the device itself.

Maple Court Workshops then asks about its Evening Classes campaign. The specialist uses the relevant campaign evidence and reads its current strategy before discussing any possible setting. A website owner also receives a separate task to inspect the supplied booking path; the device table has not tested that path.

The team determines that the immediate deliverable is a measurement and visitor-path investigation. No device mutation is prepared merely to make the review look actionable. The Slack reply states the observed CPA difference, the scope distinction and the next assigned evidence task.

If the client later requests an eligible supported edit, it receives a new exact proposal with the real campaign and any required group relationship. The technical result will establish that setting, while a later performance review will assess the commercial outcome.

The handoff records the period, metric, conversion definition, account mix, named campaign evidence and remaining task. Another operator can continue without assuming that mobile was excluded or that a website defect was already verified. The invented case illustrates a decision process rather than a promised client improvement.

Reed Harbour Media also retains the distinction between the account view and Evening Classes. If the client later asks whether the same pattern affects another campaign, the specialist obtains that campaign's relevant evidence. The earlier explanation remains valid because its scope was explicit. There is no need to repeat the entire account investigation merely to restate the original comparison, but the new campaign claim still requires its own appropriate observation before the agency recommends a setting change.

Choose a supported response to the finding

A device review can lead to several legitimate responses: retain the current setup, investigate one campaign, review measurement, assign a website task or prepare an eligible supported control. Choose according to established evidence and the client's objective rather than a compulsory device-ranking rule.

If the requested live action cannot be prepared through the current tool, state the actual supported next workflow. Do not fabricate a direct device fix because the underlying API has related resource fields. Capability requires the current preparation path and the account's actual eligible configuration.

Creative planning can still help. The assistant can suggest clearer mobile landing-page messaging or a test brief from supplied product context without pretending those ideas are current account assets. A creative suggestion remains conversation until a separate supported asset edit is deliberately requested and approved.

Keep broad account changes proportional to the evidence. Switching a strategy, pausing a campaign or changing a budget has its own effect. The fact that such a tool exists does not make it an equivalent substitute for a device adjustment the user requested.

Present the exact approved effect

A prepared live change should name the target, its actual parents and the requested resulting setting. Explain whether the operation changes a preference, a target or eligibility. The owner should not need to infer that effect from an internal field name or a vague optimise-device label.

Show each logical action on its own visible step within the supported confirmation. When the complete strategy needs smaller requests for safe review, explain that requirement before writing. Do not silently execute only the first supported part and describe the entire device strategy as changed. The client approves the actual listed effects.

The pending message states that no changes have been made yet and that the proposal expires in fifteen minutes. The connected owner confirms before a live provider edit. A descriptive report, follow-up explanation or proposed investigation is not itself permission to mutate account settings.

Current validation checks applicability before execution, and result verification checks the declared resulting state afterward. These are separate safeguards. The approval description should remain faithful to the actual prepared effect even if a broad performance discussion preceded it.

Verify configuration without promising uplift

A successful readback establishes the requested configuration fields and relevant relationship checks. It does not establish that the campaign immediately achieved a better CPA or ROAS. Keep the executed-and-verified receipt prominent while preserving the separate performance milestone.

Changes in progress should be visibly acknowledged after confirmation so the owner does not assume the button had no effect. The final receipt follows the actual execution result. A delayed notification should not invite another blind click or duplicate mutation.

If an edit was sent but complete readback is unavailable, reconcile the current resource and durable execution record before another attempt. The setting may already have changed. A missing final message is different from a provider rejection, and neither state should be guessed from notification timing alone.

Review later results with a defined interval, scope and conversion metric. Historical activity can remain in a report after a valid setting change. The agency should explain the dates and outcome rather than treat every older device row as proof that the verified configuration never applied.

Choose readable tables and native charts

A device table is useful for comparing several metrics within one selected period. A time-series chart answers a different question about changes over time. Choose the presentation requested by the client and obtain the relevant dimensions instead of turning a period total into invented daily points.

The Slack chart flow can present up to three selected metrics or segments together where the requested composition is supported. If the request exceeds that readable limit, explain it and ask for a smaller selection. Do not create several unrelated charts when one supported combined chart answers the actual brief.

Native charts have a twenty-point-per-series limit. Longer periods should use a stated coarser breakdown when daily detail was not required. When the user explicitly requires daily points beyond that limit, preserve the request with the necessary split and explain why. Do not silently discard dates or invent an image replacement.

Keep dates and numbers readable. Current-year date labels do not need the year, and a weekly range can communicate actual dates rather than an opaque year-week code. Use at most two displayed decimal places for ordinary metrics and preserve undefined values instead of padding them into numeric-looking results.

Continue the device conversation in Slack

A clear opening asks for device performance over a stated period and metric. The assistant should use the account type and established context when the metric is reasonably implied, then make its selected definition clear. Genuine ambiguity needs a focused question rather than a full restart of the conversation.

A follow-up asking about the device with the lowest return should preserve the prior period and relevant scope unless the user changes them. A later request naming one campaign narrows the population. The assistant needs the appropriate evidence for that narrowed claim rather than reusing an account total as a campaign fact.

Explanations and brainstorming can remain conversational where the facts are already available. Current setting claims and measured results need relevant provider evidence. Avoid repeated calls merely to restate a returned number or define CPA, while retaining the reads needed for an actual changed scope.

The client-facing message should state the observed result, useful caveat and concrete next decision. Keep user mentions before the final feedback text and controls. Progress messages should be clearly visible without a user mention, and raw IDs or technical payloads should remain behind the scenes.

Leave a clear agency handoff

Record the business question, actual selected period, metric definition and report population. Include the device observation and the campaign context that changed the recommendation. Another account manager should be able to distinguish a volume finding from an efficiency finding without repeating every provider read.

Assign website, measurement and account-setting tasks separately. State which observation will complete each task. A verified provider edit should not close an unresolved booking-path investigation, and an unfinished website task should not obscure an established configuration result.

If the client retains the current device strategy, record why. Retention can reflect the evidence and business objective rather than inaction. Future reviewers should not repeat an unsupported device-exclusion proposal simply because the informational table still contains an uneven distribution.

Close the review with the established observation and next accountable step. If no live change was requested, say that the review produced an investigation or retention decision. If a setting changed, retain its exact verified result. This helps another operator continue the client's work without assuming that a descriptive device comparison silently became a completed account edit.

Use the agency audit workflow for account context and the agency approval workflow for exact edits. The related app-category review concerns inventory categories, which remain different from device reach.

Answer common device-review questions

Does this informational card change the audit score?

No. It is a performance view with hidden status and no direct fix.

Does the account query include only enabled campaigns?

No. The current customer-level device query has no enabled-campaign predicate.

Does most value mean highest ROAS?

No. The summary's leading device uses absolute value or conversion volume.

Is total ROAS the average of device ROAS values?

No. It uses total conversion value divided by total cost.

Does a missing category prove zero activity?

No. The current table presents returned categories rather than manufacturing every possible row.

Can every campaign use the same device adjustment?

No. Actual campaign and bidding-strategy eligibility must be established.

Does the table verify the mobile website?

No. Visitor-path and measurement checks require separate evidence.

Does verified execution guarantee better CPA?

No. It verifies the requested setting; later business outcomes remain a separate review.