Google Ads device-performance audits for consultants: explain the pattern before choosing the change

Define the client's device question

Agree how the answer will be used. If the owner wants a board report, a concise comparison and a documented definition may be enough. If they want a live exclusion, the engagement needs a justified target and current capability checks. If they want to diagnose the booking experience, an advertising table is only one input. Asking about the intended decision at the start prevents a useful reporting task from expanding into unrequested implementation. It also helps the consultant avoid gathering every available metric when only a few selected values would answer the actual question.

A client saying “mobile is bad” has not yet defined a useful audit question. They might mean that mobile visitors produce fewer purchases, that a booking page is difficult to use on a phone, that calls are not recorded properly or that the mobile share of spend has increased. Each concern requires different evidence. The consultant should establish the intended business outcome and the scope before recommending a live change.

Ask which account or campaign, which period and which outcome the owner wants evaluated. If the request already supplies those details, use them rather than reopening settled context. A clear question such as “Compare primary booking CPA by device in this campaign over the last thirty days” gives the report a defined population and measure.

Keep the decision separate from the first read. A device report can identify a pattern worth investigating without establishing its cause or the right edit. The engagement can begin with diagnosis, then move to a supported change if evidence and business context justify it. This preserves the client's intention while preventing a broad device exclusion from becoming the automatic answer to every weak ratio.

Understand the device dimension

The Ads device dimension describes the device category associated with the reported advertising activity. The current audit labels mobile, computers, tablet, connected TV and other returned categories in readable language. A report can retain an additional returned category instead of silently discarding it because it was absent from a familiar short list.

This dimension is different from app inventory, campaign channel and the client's product category. A mobile row is not evidence that all its activity happened inside applications. A computer row is not a separate Search campaign. Keep these surfaces distinct when interpreting an apparent difference so the consultant does not select the wrong control for the actual question.

A device category is also not a user identity or a complete customer journey. The report does not independently tell the consultant which person later used another device or why a particular visitor abandoned a form. Use the observed performance row for what it measures. Additional journey claims require their relevant evidence and should not be inferred solely from the device name.

Read the account-wide performance table

The ppctor account reader requests device-segmented metrics from the customer resource over the selected date range. It selects impressions, clicks, cost, conversions and conversion value, then aggregates by the returned device key. The account report uses these base metrics rather than inventing a second device measurement source.

The shared table includes an All row and the device rows. The business-type presentation selects conversion value and ROAS for ecommerce, or conversion rate and CPA for lead generation. This presentation helps the owner focus on a commercial measure, but the selected metric definition still needs to match the actual question.

The account view can establish a broad pattern. It cannot independently identify which campaign produces that pattern. If computers show a different overall ratio from mobile, the next useful question can be whether the same difference appears within the relevant campaigns. Do not treat the account average as a verdict that applies equally to every campaign, ad group and customer intention.

Locate the campaign behind the pattern

The campaign breakdown reader queries campaign metrics segmented by device and groups the results by campaign and device. This provides the campaign-level context that the account view lacks. The implementation reads the selected period in an account-scoped query rather than making one device query for every selected campaign.

The resulting campaign report uses the same table calculations as the account report. That consistency matters when a consultant moves from a broad summary into one campaign. The ratio should not change merely because a different renderer averages device rows differently. The selected population can change legitimately; the arithmetic definition should remain stable.

Preserve current and historical distinctions when reading the campaign. The breakdown query supplies performance over its date range, rather than limiting its metrics to only currently enabled campaigns. A campaign that is paused now may have historical activity. If the owner asks specifically about enabled campaigns, apply that intention through the appropriate state evidence instead of pretending the ordinary historical breakdown already imposes it.

Separate an informational report from an issue

The device performance card is informational. Its status is hidden and it does not contribute a scored issue merely because device ratios differ. This is an important consulting distinction: the table provides evidence for review, rather than declaring that one device must be fixed.

A report can show that one device generated most conversion value or that another has a different CPA. Those statements describe the returned figures. They do not prove why the difference exists, that the weaker row has no business purpose or that a bid change is supported for the campaign.

Explain the observation in language proportionate to the evidence. “Mobile has the higher observed booking CPA in this period” is a concrete comparison. “Mobile traffic is broken” claims a diagnosis that needs more evidence. The consultant should identify the next decision and required investigation instead of inflating a useful table into a universal account fault.

Keep the reporting period exact

Last thirty days is a rolling interval, while last month is a calendar period. The assistant should preserve the user's requested dates rather than swapping them for whichever audit default is convenient. The ordinary audit context uses the last thirty days, but an explicit conversational request can require a different supported window.

State the actual dates and account time context when they affect interpretation. A daily comparison across sources can be misleading if the boundaries differ. A current partial day also differs from a completed day. Resolve the intended treatment and retain it throughout follow-up tables, charts and any decision about the weakest device.

Current configuration is not a complete record of settings across the historical window. If a device-related setting changed midway through the period, the whole row cannot automatically be attributed to today's setting. Obtain the relevant history when necessary, preserving its available coverage. The change-history guide for consultants explains why observed activity, current state and complete history are separate questions.

Read exposure alongside efficiency

Separate commercial materiality from ranking. The worst ratio in a list might belong to a small row, while a less extreme row carries most of the cost and opportunity. Both facts can be presented without inventing a universal priority rule. Ask whether the owner is seeking the lowest observed return, the largest amount of low-return activity or a particular customer-experience concern. Preserve the chosen criterion when identifying a target. A consultant should not name one device as worst under CPA and then justify excluding a different device under cost share without explaining the changed decision rule.

A ratio alone can hide the scale of the observation. One conversion from a small amount of cost and many conversions from a larger amount can produce attractive-looking ratios with different decision value. Include cost and selected outcome counts beside the efficiency measure so the owner can see the underlying evidence.

Impressions and clicks provide exposure context but are not interchangeable with purchases or bookings. A high click share can coexist with a lower outcome share for many reasons. The consultant should identify the selected commercial measure before describing the device as successful or unsuccessful.

Do not invent a universal minimum sample threshold as though the provider report guarantees it. A consultant can agree a sensible decision rule for the engagement and label it as that rule. If evidence is too limited for the intended decision, state the limitation and choose a further observation period. Waiting for useful evidence can be a justified action, rather than proof that the audit produced no value.

Select the client's commercial measure

The standard audit uses Ads conversions and conversion value. Those measures must not be described as purchase-only automatically. Eligible primary actions can include different business outcomes, and values assigned to leads or calls can contribute to standard conversion value. A client asking for actual purchase performance needs the corresponding outcome selection.

Primary and secondary actions should remain distinct. The standard primary outcome intention should not silently include secondary observations. If the owner explicitly wants a named action or a broader measure, preserve that selection and label it. The report can answer multiple valid questions, but it should not switch between them without telling the owner.

For a lead business, ask which booking or contact outcome matters when the context does not settle it. A raw total of forms, calls and imported sales can have a different meaning from completed appointments. The conversion-counting guide helps separate the configuration choice from the resulting report interpretation.

Calculate ratios from their base totals

Keep precision for arithmetic and simplify only the displayed value. Two ratios that round to the same visible number can still differ in their underlying inputs, so a threshold selection should use the defined numeric evidence rather than comparing formatted strings. Conversely, additional decimal places do not make a small observation more reliable. Show a readable value with its base counts and cost, then apply the requested exact condition in the report framework. This gives the owner a clean presentation while preserving the calculations required to reproduce the selected result.

ROAS is the selected conversion value divided by cost. CPA is cost divided by the selected conversion count. Conversion rate uses the relevant outcome count and clicks in this report. Keep those denominators visible when comparing devices so a large percentage is not detached from its actual population.

The All row should use sums of the base metrics, followed by ratio calculation. It should not average the displayed device CPAs or ROAS values. Device rows with different amounts of activity have different weights, and rounding their ratios before averaging introduces another distortion.

The same rule applies to weekly chart aggregation. Sum the relevant cost and outcomes for the week, then calculate its ratio. Do not average seven daily ratios to create an apparently precise weekly result. The shared report framework should preserve those base measures so a requested layout changes presentation without changing the underlying metric definition.

Distinguish zero, undefined and unavailable

If a row has known cost but an unavailable selected outcome, describe exactly that state. Do not rank its undefined CPA as the lowest cost or as infinite demonstrated inefficiency. The owner can decide whether the missing evidence warrants further investigation, but the report should not make that decision by substituting a convenient number. Correct missing-value treatment protects both read-only comparisons and the target selection for any later approved action.

A selected conversion count of zero can be meaningful when the relevant data was actually returned. An unavailable outcome is different. A denominator of zero can also make a ratio undefined even when its base metrics are known. Preserve these states instead of presenting every missing ratio as zero return or zero cost per acquisition.

The shared audit table displays a dash when the relevant denominator does not permit the ratio. Provider read failures propagate to the report-loading boundary rather than being converted into a successful empty set of device metrics. A consultant should retain that evidence distinction in the explanation, especially when the owner asks for the worst-performing row.

Empty returned rows can mean there is no relevant activity in the selected population. They do not certify that every device had zero performance forever. State the period and scope, and investigate the appropriate connection or data question if the owner expects activity. Repeating the same report as a confident all-zero answer does not resolve a failed or mis-scoped read.

Review campaign mix before drawing a conclusion

A commercial distinction can explain why separate campaign populations should remain separate in the review. An urgent service-booking offer and a longer-consideration consultation can have different conversion paths, even if both use the same account and device labels. Do not collapse them into a universal verdict simply to produce one ranking. Show the account summary when it answers the broad question, then identify the relevant offer context for the actual decision. If the owner deliberately wants a combined total, retain that combined scope and explain that it includes both commercial purposes.

Account device totals combine the campaigns contributing activity in the period. A device can carry a different mixture of offers, locations or campaign types from another device. The account ratio can therefore differ even when a selected campaign's own device comparison is less pronounced.

Use campaign breakdowns to identify the relevant contributor before proposing an edit. A consultancy supporting several service lines should not exclude a device throughout the account because one line dominates the summary. Preserve the actual campaign and offer that the client wants evaluated.

Keep this investigation focused. If the owner identified one campaign, request the necessary device metrics and current prerequisites for that campaign. Do not automatically expand every question into a full account inventory, website review and conversion redesign. Broader work should follow a demonstrated need, not merely the existence of additional tools.

Keep device performance and page behavior distinct

Write a page-investigation brief that somebody can actually execute. Identify the intended public destination, the relevant device context, the customer action and the behavior that needs observation. A instruction to “fix mobile” is too broad to demonstrate completion. The website colleague should be able to show whether the form can be reached, completed and confirmed in the selected route. Keep that observation separate from the advertising outcome count. A functioning route is useful readiness evidence, while a recorded conversion still depends on the measurement definition and relevant observed activity.

A lower mobile conversion rate can justify a landing-page investigation, but it does not prove that the page is technically faulty. Advertising intent, offer mix and measurement can affect the same observed ratio. Obtain actual page evidence before claiming that a button, form or checkout is broken.

A device report also does not prove the final browser journey. Configured destination fields, redirects and application behavior are different evidence surfaces. A consultant should assign a website task with a clear observation requirement if the hypothesis concerns the mobile booking experience.

That task can be reviewed separately from an Ads mutation. Changing ad creative or targeting does not repair the site's form, and fixing a form does not necessarily change campaign settings. The landing-page tracking-tag guide explains why configured URLs, actual journeys and measured outcomes need distinct verification.

Account for call and offline outcomes

Calls can be a genuine primary outcome and can carry assigned conversion value. A mobile comparison based only on online purchases may omit the business question the owner cares about. Resolve the requested outcome scope explicitly rather than assuming that every call is either irrelevant or equivalent to a completed sale.

Call clicks, recorded call conversions and completed commercial conversations are different observations. The device report does not independently certify the actual conversation quality. If the client asks about that quality, identify the required operational evidence and owner.

Offline conversion actions can also be visible through their relevant Ads definitions and reported outcomes. They should not be discarded merely because the event happened outside a website. Preserve the named action and primary selection when reporting it. Conversely, do not invent a GA4 counterpart for an offline action just to force a cross-source chart; pairing is a separate requested comparison.

Compare Analytics with an explicit scope

A follow-up requesting a combined chart should preserve the same dates and clearly label each source. The owner may want to see differences rather than assert that the two measures are identical. The assistant should choose from actual available metrics and explain a genuine scope distinction concisely. It should not block a useful comparison merely because conversion value and purchase revenue have different definitions. Nor should it force them into one unlabeled series. The framework's job is to retain both meanings and produce the requested comparison from actual returned data.

A GA4 device view can help answer a site-side question, but it is not automatically the same population as Ads device activity. Select the relevant traffic scope and business measure. All-site sessions and attributed advertising outcomes can both be useful without representing the same denominator.

For a lead comparison, ask the owner to choose the corresponding GA4 event. The assistant can inspect actual event names and suggest a likely candidate, but the business correspondence must be established. A generic form event can include more than the selected Ads contact action.

Google Ads conversion value and GA4 purchase revenue can be displayed together with their definitions clear when that comparison is requested. They need not be equal to be useful comparative series. If the owner wants all-traffic revenue divided by advertising cost, call it blended ROAS or MER. Preserve the defined scope rather than relabelling it as a device-specific attributed Ads return.

Preserve thresholds and logical grouping

A request can combine cost, outcome and efficiency conditions. “Above three” differs from “at least three,” and “cost above the threshold and no purchases” requires both conditions. The framework should preserve the exact strict or inclusive comparison and the user's AND or OR grouping.

Apply conditions at the requested grain. A campaign meeting an account-wide ratio condition does not automatically make each of its devices meet that condition. Select the relevant row population and outcome fields before filtering. Keep any missing values explicit instead of turning them into the zero that makes a condition match.

If the output shows a limited preview, state its relationship to the full selected matches. The five displayed device or campaign rows should not silently become a complete total when a broader population was available. Exact logic and honest coverage are both necessary for a consultant to use a report as the basis of an approved change.

Check bidding context before proposing adjustments

A poor observed device ratio does not establish that an ordinary bid adjustment is the appropriate control. Google's bid-adjustment guidance distinguishes eligibility by campaign and strategy. It describes Smart Bidding restrictions and the special treatment of device adjustments for Target CPA. Review the applicable current context rather than applying one percentage rule to every campaign.

A consultant should identify the campaign type, bidding strategy and actual intended effect before preparing a write. A request to reduce a bid, alter a target and exclude a device can represent different business effects. They should not be described as interchangeable because each mentions the same device.

Keep observed performance and provider eligibility separate. The report can establish the comparison while the setting discovery establishes the available operation. Both are needed for a justified proposal. If the exact requested effect is unsupported in the current context, explain that boundary and resolve a valid alternative with the owner rather than hiding a different action inside the confirmation.

Discover the supported setting precisely

Review the complete prepared effect, including any field replacement implied by the selected operation. A value update, list replacement and new resource creation can have different preservation requirements. The consultant should identify what remains unchanged because the operation preserves it, rather than assuming every omitted field survives. Read the necessary existing state when preservation depends on resupplying it. This allows local schema discovery to remain efficient while keeping the eventual proposal faithful to the owner's request. A tool identifier alone is not a sufficient explanation of the change being authorised.

The editing framework has a local schema-derived tool catalogue and bounded discovery. It includes resource operations such as ad-group bid modifiers and campaign criteria, with their defined fields and dependencies. This avoids injecting every possible API setting into every conversation while allowing the assistant to find the relevant resource surface.

Schema availability is not a universal guarantee that a field is appropriate for every campaign or that Google will accept the proposed value. Current ownership, field behavior, strategy prerequisites and provider validation still matter. The consultant should describe only the concrete operation actually prepared for the selected target.

A generic framework should be adaptable as definitions evolve. It should use shared fields, metrics and capabilities to compose the response, rather than a dedicated “bad mobile” shortcut that ignores the client's requested scope. Discovery supplies the candidate operation; the business intention and actual state establish which candidate, if any, belongs in the proposal.

Resolve the actual target and parent

Resolve the recognizable campaign and the applicable group or criterion before preparing a live edit. Similar names can exist in different parents. The assistant should retain canonical references internally while presenting names and correct relationships without raw IDs.

A device performance segment is not automatically an editable resource reference. The framework must identify the corresponding supported current configuration object or creation prerequisite. It should not place a report row ID into a resource name and hope the provider accepts it.

Local dependencies can establish that a group-level modifier belongs to a group without calling Google merely to learn the structural relationship. Focused current reads establish the actual selected parent and state. Do not invent new campaigns, treat an ad group as the parent of another ad group or rediscover every account object when the necessary context is already established.

Follow an illustrative consultant investigation

Consider an invented consultant, Isla Penn, reviewing the fictional Linden Booking Workshop account. The owner says mobile bookings look expensive. Isla establishes the selected primary booking action and a completed thirty-day period. The account device table shows a higher observed mobile CPA, with its cost and booking counts alongside the ratio.

She then reviews the campaign behind most of the mobile activity. Its offer differs from another campaign that dominates computer bookings, so the account summary alone cannot justify a blanket device decision. The focused campaign comparison still identifies a pattern worth reviewing, but the current evidence does not demonstrate a broken booking page.

The owner asks for clearer mobile-focused advertisement ideas. Isla can brainstorm from the known offer without another provider call. A website colleague receives a separate task to observe the booking route on the relevant device. If the owner later requests an Ads setting change, Isla checks the campaign's strategy and supported capability before presenting the exact effect for confirmation.

The engagement records the performance observation, the creative discussion and the unresolved page investigation separately. A verified later setting edit would close only its configuration task. This example illustrates a careful decision process; it is not a customer result, a claim that mobile is generally inefficient or proof that one prescribed bid change will improve every account.

Use creative discussion and focused reads

Ordinary conversation can contribute useful hypotheses, explanations and creative alternatives. The assistant can suggest a clearer headline or a more understandable booking promise from supplied context without calling a measurement API. It should label suggestions as creative ideas, rather than presenting them as provider-tested findings.

Current performance questions need actual data. Use the selected device fields and outcome scope, then reuse established evidence when the same follow-up does not require a new read. A request to redraw the returned rows should normally reuse those rows instead of fetching unrelated campaign inventory again.

When new current state is required for a live edit, select the narrow prerequisites. Shared capabilities and dependency definitions should guide that work. The assistant should retain the campaign in context when the owner says “that campaign,” while resolving genuine ambiguity if several campaigns were discussed. This is a general conversational reporting flow, not a collection of special cases for one phrasing.

Choose readable charts for the requested data

A device bar chart can compare the selected categories for one metric. A daily trend requires actual date-segmented data, not repetition of the period total across each day. Resolve whether the owner wants device categories, a trend by date or both before choosing the layout.

Try one native chart for up to three compatible metrics or segments. More than three needs a readability-limit explanation and a selection. The ordinary native point limit is twenty per series. Longer periods can use a clearly stated weekly or monthly breakdown unless daily detail is explicit; longer explicit daily requests need split native charts with the limit explained.

Use readable dates, omitting the year for current-year daily labels and showing actual week ranges. Keep displayed metrics to at most two decimal places. Preserve source labels, outcome definitions and aggregation arithmetic. The user mention belongs before the feedback text and controls, which remain last. A polished chart should make the evidence easier to assess without hiding unavailable values or population differences.

Present exact steps before execution

A proposal should name the target, its applicable parents and the exact change. Explain whether the prepared operation changes a bid-related field, a target or another supported setting. Do not use a vague “optimise mobile” statement in place of the actual effect.

One confirmation can cover a supported sequence, with each action presented in a clear step section. If the complete instruction cannot be prepared within the safeguards, explain the need to divide or rephrase it before execution. Do not execute an arbitrary subset, and do not hide an additional activation or exclusion inside a creative replacement paragraph.

State “No changes have been made yet. This proposal expires in 15 minutes.” Confirm and Execute applies to the displayed effects. An informational device report, agreement with a hypothesis or a previous unrelated approval is not permission to mutate the account. Revised requirements need a revised concrete proposal before execution.

Verify the setting and preserve uncertainty

Immediately show a prominent change-in-progress state and inactive controls after confirmation, without mentioning the user in that interim state. This makes the click's acceptance visible while current prerequisite checks and validation run. The owner should not be encouraged to repeat a click because the message appears unchanged.

Validation checks the proposed request; verification checks the actual resulting field or resource. An accepted request is not the same as a complete verified result. If the edit may have happened but readback is incomplete, reconcile the affected configuration before retrying.

When checks pass, show executed and verified prominently and describe the established effect. That confirmation does not establish future serving, website behavior or commercial improvement. A consultant should retain those separate evaluation tasks. Clear successful configuration output and honest measurement boundaries can coexist without making the client read an internal technical payload.

Evaluate later without inventing causation

Preserve the difference between a decision rule and a promised outcome. An owner can agree to review a setting when observed CPA remains above a selected threshold over the next period. That agreement makes the follow-up actionable; it does not guarantee the threshold will be reached. If the later evidence remains incomplete, keep the review open rather than inventing a pass or fail. The consultant can still document the completed configuration milestone and the actual data available. Accountable evaluation depends on making those states distinct enough that another reviewer can follow the same logic.

Choose the observation window and stable measure before reviewing a change. The period should reflect actual exposure to the new setting rather than merely the date a proposal was discussed. If the campaign stayed paused after a change, those days do not establish its delivery effect.

Document other material changes during the comparison. New offers, creative, budgets and website behavior can influence the same device ratios. A before-and-after chart is useful evidence for a decision, but it should not automatically be described as a controlled test proving one cause.

Keep the next decision concrete. The owner might retain the setting, review a remaining page issue or request another supported adjustment after observing the agreed period. A consultant should not promise a fixed improvement merely because the previous device row looked weak. The purpose is to make the subsequent choice accountable to actual results and the business objective.

Leave evidence and accountable next steps

Before closing the report, check that its visible targets match the approved intention. The campaign should not become an account-wide target in the result, and a modifier should not be described as having paused the campaign. Use names and correct parents without raw asset IDs. If a multi-step request included another action, make its verified effect or unresolved status visible too. A polished final banner cannot compensate for an omitted step. The client should be able to compare the original request, proposal and achieved effects without deciphering technical payloads.

Leave the next observation date and responsible person in the engagement record. A website investigation, a measurement selection and a campaign setting review can each have a different owner. Naming those owners prevents the device report from circulating indefinitely without a decision. Keep the evidence itself focused on the selected period and measures. A successor can request genuinely changed current state when necessary, while reusing the established explanation of the commercial purpose and resource relationships instead of starting every review from scratch.

The handover should record the original question, selected scope, actual dates and commercial measure. Include the device observation, the campaign context and any supported change that was approved and verified. A colleague should be able to understand the reasoning without reconstructing every Slack reply.

Assign remaining website, creative and measurement tasks to their owners. Give each task an observable completion condition. An ad setting's readback cannot close a form-usability investigation, and a creative suggestion cannot be reported as an uploaded advertisement.

Preserve deliberate non-action where appropriate. If the observed device difference does not justify a change, record the reason and the evidence that would prompt another review. This prevents a future colleague from applying a broad exclusion solely to make the device table look more uniform.

The consultant audit guide provides broader context. Useful device consulting connects a precise report to a real business decision, uses the correct supported capability for any approved write and keeps configuration success separate from a demonstrated customer or financial outcome.