Choose the next investigation from the evidence
Google Ads device-performance audits for freelancers: turn the account mix into a practical client investigation
Read device totals and efficiency correctly, establish conversion and period definitions, and investigate campaign context before recommending a destination, measurement or supported bidding change.
Define the freelance device question
A client who asks whether mobile is working may be asking several different questions. They may want to know where the account gets most of its enquiries, which device has the lowest cost per conversion, or whether their website is difficult to use on a phone. Start by identifying the decision behind the question. The device-performance audit gives a useful account-level view, but it does not answer every one of those questions automatically. A freelancer should connect the requested measure to the business purpose before proposing a change. This makes the investigation both more accurate and more efficient: the next read or test has a clear reason instead of becoming another broad report that never resolves the client's concern.
Write a small brief containing the account, period, outcome and intended comparison. For a shop, the owner may want to compare recorded conversion value relative to spend across devices. For a service business, the question may concern the cost of the conversions that represent useful enquiries. A third client may care about the experience of completing a booking on a phone. These require different evidence. The account table can describe the observed mix and derived ratios. A business-quality review needs the relevant conversion meaning, and a usability review needs a journey test. The freelancer's first contribution is to distinguish these tasks without making the owner learn the technical vocabulary of every provider report.
Agree the deliverable and the action boundary. A reporting engagement can end with a recommendation and a clearly scoped next investigation. An implementation engagement may also prepare a supported edit after the cause and target are established. Do not turn a request for a device comparison into permission to restrict traffic. The owner should be able to ask what the account shows, discuss an explanation and refine the question before any live operation is prepared. A clear freelance brief preserves that space for reasoning. It also prevents the final report from promising an improvement that the initial table cannot prove, such as guaranteed savings from excluding whichever device looks weakest in one snapshot.
Understand the account report
The current account reader queries Google Ads customer performance segmented by device over the selected date range. It requests impressions, clicks, cost, conversions and conversion value, then aggregates the returned rows by their device key. The audit card presents those results as account-wide information. It has no recommended numeric target and hides the scoring status. That presentation matters: a device breakdown is evidence for a decision, rather than a failed setting that should automatically be fixed. A freelancer should describe the table as the account's observed performance mix and preserve its scope before drawing conclusions about a specific campaign, landing page or bidding control.
The reader's account query does not filter the population to campaigns that are currently enabled. Historical activity in the selected period can therefore differ from the list of resources the owner sees as active today. Do not remove a paused campaign's contribution mentally while quoting the account total. If the client wants only a defined enabled-campaign population, request and label that population separately. This distinction is important after a restructure or pause: the current inventory and the historical reporting population are different facts. A freelancer should state which one the answer describes so the owner can understand why the table includes activity from a campaign that is no longer serving.
The card is not a website analytics report, a count of unique people or a direct measurement of cross-device journeys. It uses the returned Ads metrics and segmentation. If the client needs a comparison with Analytics, establish the source, metric and population for that comparison rather than assuming a similarly named device dimension produces an identical report. A useful account audit can still begin with the Ads breakdown. The next step is to formulate the question that the second source can answer. This keeps the investigation coherent and avoids turning ordinary differences between report definitions into an unsupported claim that a device was mismeasured or that the account lost conversions.
Resolve the reporting period
The reader defaults to the last thirty days when no range is supplied and uses the account time zone in resolving its date segment. The selected audit period remains the relevant context for a particular result. A freelancer should preserve those dates in the evidence note rather than assuming every audit snapshot covers the period mentioned later in Slack. Last month and the last thirty days are different requests. A rolling interval and a calendar interval can have different endpoints and different business conditions. Resolve the requested period before comparing devices, then keep that definition consistent across any campaign drill-down and subsequent chart.
Prefer a clearly defined comparison when the client is evaluating a change. If the period crosses a promotion, a site release or a major campaign launch, the device mix may reflect those events. Do not treat the whole interval as a uniform experiment. Record the relevant boundary and decide whether the analysis needs separate periods or a trend. An owner may initially request an overall table, then ask what happened after a website update. That follow-up changes the analytical question. The freelancer should obtain the appropriate period rather than reuse an earlier total and describe it as current evidence for the new question.
Keep incomplete periods visible. A result that includes part of today should not be compared casually with a set of completed days. Likewise, a current month-to-date result should not be called the whole month. The client-facing explanation can be concise, naming the completed date range and why it fits the question. The technical record should retain the resolved endpoints and account time zone. These details make the work reproducible for another reviewer. They also reduce repeated disagreement about totals, because the team can establish whether two reported answers use the same population and dates before investigating any difference in the underlying device metrics.
Read the returned device categories
The current table uses friendly names for Mobile, Computers, Tablet, Connected TV and Other where those keys are returned. It preserves additional provider keys instead of silently dropping them. Only present keys become rows. This is a reporting map, not a list that guarantees every device category had activity. A freelancer should therefore distinguish an absent row from a returned row with zero outcomes. If the owner asks why a device is not listed, inspect the reporting result and scope rather than assuming it is excluded or broken. A device category's absence in one table does not establish its current targeting configuration.
Keep the provider grouping separate from the client's practical website-testing plan. A Mobile row is an aggregate, not a list of every phone, browser and operating system. It can identify a reason to investigate the mobile journey, but the test must select a manageable set of relevant conditions. The same is true for Computers and Tablet. A freelancer should avoid promising that a report row diagnoses a particular screen size or browser issue. Instead, use the commercial evidence to prioritise a concrete journey review, then record the actual observations from that review separately from the provider's broad device totals.
Do not combine categories casually to simplify a conclusion. The owner may want all non-computer traffic grouped together, but that is a new reporting transformation and should be named. Combining Mobile and Tablet can be useful for a defined question, yet it should not hide a materially different outcome within either group. If the client asks for a grouped comparison, retain the underlying totals and calculate the derived metrics from them. This gives a clear answer without losing the ability to investigate the separate rows later. The freelancer should make the grouping explicit rather than present a locally created category as though it were an unchanged provider device label.
Interpret the All row correctly
The current All row sums the underlying impressions, clicks, cost, conversions and conversion value from the returned device rows. It then calculates the relevant derived metrics from those summed bases. This is the right distinction between a total ratio and an average of device ratios. A small row should not have the same influence on the account's CPA or ROAS as a much larger row merely because both are displayed once. A freelancer explaining the account result should use the total derived from the bases and preserve any filters that define the returned population. The All row describes that population, rather than every possible report the client might compare it with.
Use a simple invented illustration when the client needs the arithmetic explained. Suppose one device has a small amount of spend and a high ratio, while another has much more spend and a lower ratio. Averaging their displayed ratios equally would exaggerate the account result. Adding the underlying spend and value before dividing gives the ratio for the combined activity. The same principle applies to CPA using cost and conversions. The illustration is a way to explain the method, not a customer result. A freelancer can keep the client-facing answer short while retaining the actual bases in the working note for anyone who wants to reproduce the calculation.
Rounding belongs at presentation, not at the start of aggregation. The current helpers retain the numeric metrics for arithmetic, while display formatting makes counts and money easier to read. A conversion total can be fractional even when the table shows a compact number. Do not rebuild an account total by adding rounded cells and then treat a small difference as a provider defect. If a reconciliation matters, use the underlying returned values with the same definitions. The freelance report should explain significant differences in population or calculation rather than chase display rounding as though it changed the business outcome.
Evaluate value and ROAS
For an account classified as ecommerce, the device table adds conversion value and ROAS. The current calculation divides returned conversion value by cost. Its formatter displays the ratio as a percentage: a ratio of four is shown as four hundred percent. That presentation should be clear to the client before a target or historical comparison is discussed. Do not compare a displayed percentage with a multiple as though they were different performance measures. A freelancer can write both representations when explaining a result, then use one consistent convention in the deliverable. The purpose is to make the commercial relationship understandable without changing the underlying metric.
The table's value is the Ads metric selected by the reader. It is not automatically a purchase-only inventory reconciliation or all-site revenue. If several primary conversion actions carry value, the relevant definition must be established before the client calls the number purchase revenue. A freelancer should keep the metric label accurate and ask what actions the business includes in its commercial goal. The device comparison can still be useful with the existing value definition, but its interpretation must match that definition. Do not silently restrict or expand the conversion population to make a device appear stronger, or assume that an ecommerce account's every returned value came from the same type of transaction.
ROAS does not answer every profitability question. The audit ratio uses recorded value and advertising cost; it does not automatically subtract product costs, fulfilment expenses or refunds under the client's business accounting rules. If the owner wants a profit decision, identify the additional business measure and its source. The device table can prioritise a discussion without supplying that entire calculation. A freelancer should describe what the ratio establishes and what it leaves open. This makes the recommendation more useful: the owner can distinguish a strong reported advertising return from the broader question of whether the associated orders or customer relationships are commercially attractive.
Evaluate conversions, rate and CPA
For a non-ecommerce account, the table adds conversion rate and CPA. Conversion rate is returned conversions divided by clicks, and CPA is cost divided by conversions. These formulas describe the selected Ads metrics, not a verified count of unique qualified customers. A freelancer should establish what the conversion column represents before judging the result. The owner may use forms, calls and other actions in one primary conversion total. That can be an intentional account goal, but the business meaning needs to be clear. The device table should not be described as cost per qualified lead unless the selected conversion definition and supporting evidence actually establish that meaning.
A lower CPA can coexist with less valuable enquiries. The device report alone does not read the client's sales process or decide which submissions were useful. If the owner suspects poor quality, identify the relevant business outcome and how it can be linked to the analysis. Keep that investigation separate from the initial arithmetic. A freelancer can report that one device had a lower cost per recorded conversion while asking the business owner to evaluate the associated enquiry quality. This is a precise conclusion. Calling it the best source of customers without that evidence would overstate what the table can establish.
Calls deserve the same care. A phone-oriented device may generate useful primary call conversions, and those actions can carry value as well as counts. Do not ignore them simply because the account also uses a form. If the owner asks for a comparison with a particular Analytics event, select the corresponding event explicitly rather than treating every lead category as a universal event name. The freelancer's role is to align the measure with the client's actual setup. Once that definition is settled, the device comparison becomes a meaningful way to prioritise a campaign or journey investigation, instead of a contest between columns whose business meanings were never agreed.
Confirm the conversion definition
Record the conversion selection alongside the dates and account. The reader requests conversions and conversions value, rather than all-conversions variants. If the client asks for all primary conversions, purchase-only conversions, leads or one named action, preserve that requested meaning in the subsequent report. A broad audit snapshot should not be relabelled as a narrower named-action analysis. A freelancer can use the conversion inventory and account goal context to explain what the current number means, then request the appropriate filtered evidence when needed. This prevents repeated confusion in which two tables have the same device labels but different conversion populations.
Do not treat a conversion category as a complete description of business quality. Categories help identify the intended type of action, while the account's actual setup and business process define what it represents. A client may record several lead-related actions without wanting separate semantic distinctions in a high-level device comparison. Respect that instruction. If a more specific comparison is requested, resolve the named action and corresponding event rather than guessing. These are generic reporting principles, not a special device-only workaround. They allow the freelancer to explain results consistently across Ads and Analytics while keeping each source's measure and selected population visible.
Changes to conversion settings can alter a comparison even when traffic did not change. If the period includes a goal update, record the date and investigate its effect before attributing every movement to devices. The device table is a snapshot of returned metrics under the chosen reporting definition. A freelancer should avoid making a causal claim from a change in that definition alone. The final note can say that the interpretation is affected by the account's conversion configuration and identify the required follow-up. This is more informative than either dismissing the whole report or presenting an apparently precise device ranking built on an unexplained shift in what the account counts as success.
Separate contribution from efficiency
The current summary identifies the device with the most returned value for ecommerce or the most returned conversions for other accounts. That is a contribution ranking by absolute volume. It is not a ranking by ROAS, CPA or conversion rate. A freelancer should preserve that distinction when summarising the audit to the owner. The largest device can be commercially important while having a weaker efficiency ratio. A smaller device can have a strong ratio while contributing relatively little total value. Both observations can matter, but they answer different questions. Do not turn the summary's most-value statement into an instruction to move all activity to that device.
Ask what the owner means by best or worst. Best might mean highest total value, highest return, lowest cost per recorded conversion or the strongest qualified-enquiry outcome. Those choices can select different rows. Where context makes the intended measure clear, use that context and state the measure. Where it remains ambiguous and affects the next action, clarify before preparing a live change. The freelancer can keep this practical by presenting the relevant volume and efficiency facts together. The owner then understands why a device may deserve investigation without being told that one generic ranking decides every advertising and website question.
Do not rank undefined ratios as though they were ordinary numbers. A device with no cost has no meaningful return on spend in this table, and a device with no conversions has no finite CPA. These cases require different interpretation from a low but defined return or a high but defined CPA. Keep them visible in the comparison rather than converting them into zero or an arbitrary worst score. A freelancer who explains the distinction helps the client make a better decision. The owner can see whether the issue concerns absent activity, absent outcomes, missing evidence or a genuinely weak measured efficiency under the agreed definition.
Preserve missing and undefined metrics
The current metric helpers show a dash when the denominator needed for a ratio is absent or zero. ROAS requires positive cost; CPA requires positive conversions; conversion rate requires positive clicks. A zero outcome with a positive base is different from a missing base. Preserve that distinction in the client explanation. The freelancer should not turn every dash into zero or claim that a device had no business value merely because a particular ratio could not be calculated. If the row carries useful base metrics, use them to explain what is known and what remains undefined.
The loading and failure states also matter. When reports are still loading, the audit uses a loading finding. When a device report fails, the report loader records that failure independently so the affected section can be unavailable while other reports remain usable. A successful empty map hides the ordinary device card. These are different evidential states. A freelancer should not call an unavailable report a zero-performance result, and should not call a completed empty result proof that every possible device had zero activity. Establish the actual returned scope before deciding which interpretation belongs in the handoff.
Keep display precision proportionate. Costs and rates are formatted for readability, conversions may retain a compact fractional display, and ROAS is presented through the shared percentage formatter. The audit's compact currency symbol is not by itself evidence of the account's currency. Verify currency context when a financial comparison requires it. A freelancer can give the owner a readable table without obscuring these definitions. The working evidence should retain the values and scope needed for reconciliation, while the client-facing answer avoids long decimal strings and unexplained symbols that make an otherwise simple device comparison feel technical and unreliable.
Assess the strength of the evidence
A ranking becomes more persuasive when the underlying population is substantial and relevant to the decision. A tiny device row can change dramatically after one additional conversion. A freelancer should inspect the base metrics before recommending a broad restriction from an attractive or alarming ratio. The table gives the necessary starting quantities, but it does not supply a universal confidence threshold for every account. Choose a proportionate interpretation based on the client's volume and the consequence of the proposed action. A small exploratory observation may justify monitoring or a journey test, while a larger persistent pattern may justify a more focused campaign review.
Look for consistency across an appropriate period without manufacturing certainty. Repeating the same broad query many times does not improve the evidence if the scope and available data are unchanged. A trend or a properly comparable second period can answer a defined persistence question. Record why that extra read is needed. A freelancer should avoid both overreacting to a short fluctuation and endlessly delaying a clear business decision. The useful next step is the one that resolves the remaining uncertainty, such as checking whether the difference is concentrated in a campaign or whether a recent destination change affected a device journey.
Keep the strength of the recommendation aligned with the strength of the evidence. You can recommend inspecting a mobile form based on a device pattern without claiming the form is broken. You can recommend checking a campaign's control eligibility without promising an edit. You can recommend a supported change when the target, business reason and current settings are established. These are distinct outcomes. The freelancer's value lies in choosing the appropriate next action and explaining why it fits, rather than forcing every informational table to produce an immediate mutation or treating every weak row as a proven cause of poor account performance.
Investigate the campaign mix
An account device difference can reflect the campaigns that contribute to each device. One device may be associated with a different mix of offers, channels or bidding objectives. Before treating the aggregate gap as a device effect, inspect the relevant campaign breakdown. The current code has a separate campaign device reporting path, which supports a more scoped investigation. A freelancer should use it when the question requires campaign context, rather than repeatedly requesting the whole account and guessing which campaign caused the result. The account table identifies the pattern; the drill-down helps locate the population that deserves attention.
Choose the drill-down deliberately. If the owner asks about one campaign, preserve that campaign and period. If they ask which campaigns contribute to the device gap, request the comparison needed to answer that question. Do not fetch every available dimension simply because the assistant can. The analysis should connect each additional table to a decision. A freelancer can then explain whether the account pattern is broadly shared or concentrated in particular activity. That explanation is more useful than a blanket recommendation to lower a device across the account, especially when the available controls and business objectives differ between campaign types.
Historical status remains relevant during this drill-down. A campaign that is paused today may have contributed to the selected period, while a newly enabled campaign may have little history. Do not infer causal improvement from current status alone. Keep the period's activity and the present mutable target separate. If a proposed action concerns a current campaign, obtain its current settings before preparing the request. This is a generic read-to-edit boundary: historical metrics help identify a candidate, while current resource evidence establishes what can be changed now. The freelancer should preserve both rather than treating a reporting row as a complete edit specification.
Inspect the destination experience
A device pattern can justify a destination investigation without proving the cause. Review the actual offer, page and customer task associated with the relevant campaign. On a phone, the issue may concern navigation, form completion, a booking step or the visibility of essential information. The device table does not perform that test. A freelancer should record the journey observation separately from the Ads metrics so the team can see which evidence supports the recommendation. This avoids blaming a device category for a problem that belongs to a particular page, and avoids changing bids when the more appropriate task is a targeted website repair.
Define the test around the business action. For a form, inspect whether the intended fields and confirmation behave correctly. For a shop, inspect the relevant product and purchase path. For a booking service, inspect the route to a completed appointment request. Choose representative conditions based on the actual audience and available evidence, rather than claiming one test covers every phone or browser. The freelancer can provide a concise reproduction note with the observed issue and its impact on the task. The website owner then receives something actionable, while the advertising owner understands why the device report prompted the test without being treated as proof of the defect.
A repaired journey still needs its own verification. Confirm that the observed problem no longer occurs under the agreed test conditions, then evaluate subsequent advertising results using an appropriate period. Do not call the entire device difference solved merely because a form now submits in one test. Equally, do not wait for every account ratio to move before acknowledging that a demonstrated interface defect was repaired. These are separate claims with separate evidence. A freelancer who keeps them distinct can report meaningful progress while maintaining an honest boundary around the commercial outcome.
Review measurement as a separate hypothesis
If the device result raises a measurement concern, define that concern explicitly. A missing conversion event on a particular journey is different from a change in the account's selected conversion actions or a comparison between Ads and Analytics definitions. The device table cannot decide among those explanations on its own. A freelancer should identify the configuration or event evidence needed for the hypothesis, then perform or assign the relevant test. Avoid using the phrase tracking issue as a catch-all diagnosis. The client should know whether the investigation concerns collection, reporting scope, goal configuration or the customer's actual ability to complete the action.
Use compatible measures when comparing sources. Ads conversion value and Analytics purchase revenue can be compared as clearly labelled revenue measures, but their selection and attribution definitions must remain visible. An all-source Analytics revenue total divided by advertising cost should be identified as blended ROAS or MER, rather than silently substituted for a narrowly attributed Ads ratio. For lead comparisons, select the corresponding Analytics event that the site actually uses. These choices let the freelancer answer the owner's business question without pretending that similarly named numbers are guaranteed identical. The chart can compare them; the explanation should preserve what each line represents.
Keep the measurement test bounded and reproducible. Record the selected event or action, account or property, time and relevant journey. A successful test establishes the observed behaviour under those conditions, not a universal guarantee for every device and customer. If the required access is missing, provide a concrete handoff to the person who owns measurement. The freelancer can continue independent account analysis while that task remains open. The final recommendation should then say whether the device pattern is an observed performance result, a confirmed measurement issue or a question still awaiting evidence, with no unsupported diagnosis hidden inside a confident summary.
Establish the actual device control
Device performance reporting does not imply one universal device-edit operation. Google's guidance distinguishes campaign and ad-group eligibility, manual adjustments and automated bidding behaviour. It explains that target-CPA device adjustments concern the target, while several other automated strategies permit only the device opt-out form where supported. Campaign type also matters. Review Google's bid-adjustment eligibility guidance before recommending a control. A freelancer should establish the current campaign type and bidding strategy instead of assuming that a percentage suggested from the account table will be meaningful or accepted everywhere.
The current convenience-action catalogue does not contain a dedicated device-performance fix. Generic schema-mediated editing is a separate capability and must be established for the exact resource, field and operation before the assistant promises execution. This article therefore does not present the device card as a one-click bid optimiser. The card can support reasoning and a scoped recommendation. If an appropriate live operation is supported, it should be prepared through the normal edit framework. If it is not, give the owner an exact implementation handoff. A freelancer should never substitute a different broad operation merely because it is easier to call and produces a plausible-looking result.
Check the purpose of the proposed control as well as its availability. Reducing an eligible bid adjustment, changing a target and excluding a device have different consequences. The owner needs to understand which one is proposed and why the evidence supports it. A reporting gap may justify investigation rather than any of these actions. The freelancer should keep the recommendation proportionate and avoid guaranteeing a particular efficiency improvement. A provider-supported field establishes that an operation can exist; it does not establish that the operation is the right business response to a device pattern in this account.
Resolve the target and setting hierarchy
Before preparing a device-related edit, identify the exact campaign or ad group and the effective setting hierarchy. Google's guidance explains how campaign and ad-group device adjustments interact, including the special consequence of a campaign-level device opt-out. Keep that hierarchy in the technical review rather than presenting an ad-group adjustment as independent of its parent. A freelancer should obtain the relevant current values and establish which resource the requested change belongs to. An account-wide report is not itself a mutable account-level device target. This distinction prevents the assistant from inventing a broad setting simply because the owner began with an account total.
Use names and parent context in the owner-facing proposal. If the target is an ad group, include its campaign. If it is a campaign, name that campaign and the device action. Provider identifiers can remain behind the scenes for exact resolution. Do not display undefined parents or raw resource strings as though they were useful business context. The freelancer should review the target after any follow-up instruction, because the owner may shift from the account's weakest device to a particular campaign. The latest request should control the proposal, with current-state evidence establishing the exact object to change.
If the target remains ambiguous, resolve it before execution. Several campaigns can share similar names or destinations, and a device metric does not identify which resource the owner intended. The assistant can use current conversation context and scoped reads to propose a clear interpretation, but should not silently choose a different object to complete the task. A freelancer can keep this clarification practical by naming the relevant candidates and explaining why the target matters. Once resolved, the action should follow the local dependency requirements for that resource rather than repeat unnecessary discovery of objects unrelated to the requested setting.
Write a proportionate action plan
A device review can produce several types of plan. It may recommend a more scoped report, a destination test, a measurement investigation or a supported configuration change. Name the chosen task and the evidence that justifies it. The owner should understand why that task answers the original concern. A freelancer does not need to force every report into a live edit. An accurate recommendation to inspect the mobile booking journey can be a complete audit outcome when the existing evidence points there and the relevant test is assigned clearly.
Separate dependencies from optional investigation. Current campaign type and bidding strategy can be required for a device-control proposal. A broad account performance report is not necessarily required again once the relevant candidate and metric are already established. Website testing may be essential for a journey hypothesis but unrelated to a simple requested setting read. This distinction helps the assistant and freelancer avoid unnecessary calls while retaining the evidence that matters. The plan should identify which facts are local tool knowledge, which are current provider facts and which require a human or website observation. Each step then has a clear purpose and completion criterion.
Write the verification plan before the live operation. State which stored setting will be read back, which journey behaviour will be tested if applicable and which later period will evaluate the commercial outcome. These are separate checks. If the task creates or updates a resource, preserve the approved intent and prior state so that an ambiguous result can be reconciled. The freelancer should not promise an automatic rollback without establishing the appropriate current operation. A reviewable plan connects the observed device pattern to a bounded action and the exact evidence that will show what the team accomplished.
Follow an invented freelance review
Consider an invented freelancer, Erin Vale, reviewing the invented Oak Room Workshops account. The owner says that phones produce the most enquiries and asks whether they are the best device. Erin begins with the account table and identifies the metric behind the summary: Mobile has the most recorded conversions. She then compares CPA under the same period and conversion definition. The contribution and efficiency observations do not point to exactly the same conclusion. Erin explains both without treating either as a command to exclude a device. The owner clarifies that the immediate concern is whether the mobile booking journey is costing useful enquiries.
Erin investigates the relevant campaign and destination rather than repeatedly asking for account totals. A separate journey test identifies a reproducible issue at the booking step under the agreed test conditions. The website owner receives the reproduction note, and the advertising owner receives the scope of the finding. Erin does not claim the device table alone proved the defect. She also does not prepare a broad device adjustment as a substitute for repairing the observed journey. The plan assigns the website task and defines its verification, with a subsequent reporting period to assess the commercial pattern.
After the repair, Erin verifies the booking task under the agreed conditions and records the later device results without attributing every change to the repair. The report states the original account observation, the scoped campaign investigation, the confirmed journey issue and the checks completed. This is an invented reasoning example, not a customer result or a claim of recovered conversions. It illustrates a practical freelance method: use the device table to choose the next question, obtain evidence at the level where the suspected cause exists and report the resulting action accurately. The account becomes easier to understand without relying on a blanket device-control remedy.
Use conversation to refine the question
Slack lets the owner move naturally from a broad question to a specific investigation. They may ask which device produced the most value, then ask for a comparison in one campaign or a different period. The assistant should preserve the latest scope and requested measure. A previous account result can inform the conversation, but it is not automatically the data for every follow-up. A freelancer should review whether the assistant is using the right account, dates and conversion definition before acting on its explanation. This keeps the conversation useful without requiring the owner to restate the entire context on every reply.
Explanatory and creative tasks can remain conversational. The owner may ask for a concise client summary, a clearer recommendation or a draft message to the website developer. Those requests do not require new provider calls unless the response makes a new current-account claim. A request for a fresh metric does require the appropriate evidence. A request for a live edit requires target resolution and confirmation. These are generic task distinctions. A freelancer should expect the assistant to choose among them from the user's intent rather than use a special patched rule for every possible mention of devices, mobile traffic or efficiency.
Keep the visible answer direct. Name the metric, period and relevant device observation, then explain the next investigation or action. Avoid a raw field dump or a list of provider identifiers. If the owner asks why a device was selected as worst, show the measure used and its base rather than hiding behind an unspecified ranking. If the measure changes in a follow-up, make that change visible. The owner can then assess the recommendation without guessing whether the assistant switched from conversion volume to CPA or from account-wide data to one campaign halfway through the conversation.
Request a readable table or chart
A table is often the clearest format for a device comparison because it can show volume and efficiency together. A chart can help when the owner wants to see a trend or compare a few measures over time. The assistant should use the actual returned data and the requested breakdown, not create an unrelated illustration. Choose the format that answers the business question. A freelancer can request a concise table for the device snapshot, then a trend for the campaign and device population that needs investigation. This avoids producing several decorative charts before the analytical scope has been established.
The current native chart flow limits ordinary charts to three series and twenty points per series. For a longer period without an explicit daily request, a coarser breakdown can preserve one readable chart. If the owner explicitly requires daily detail beyond that limit, consecutive panels should retain the requested detail and explain the point limit. Do not silently split one metric into unexplained fragments or substitute an image when the agreed output is native Slack. The freelancer should also keep each series labelled by its actual measure. Cost, value and a ratio are different quantities, and a readable visual must preserve that distinction rather than imply they share an identical meaning.
Formatting should make the answer easier to evaluate. Dates should be readable, current-year labels need not carry unnecessary year detail, and metrics should avoid excessive decimals. A chart title and a short explanation can identify the account, period, population and any aggregation. The visual should support the conclusion rather than become the conclusion. A freelancer should still state which observation deserves attention and why. If the requested data cannot be obtained at the required breakdown, explain the actual limitation and the available evidence rather than inventing daily points from an account total or claiming a provider limitation that has not been established.
Approve only established supported edits
A live proposal should name the target, its parent context and the exact action. The current convenience catalogue has no dedicated device-performance fix, so any generic edit support must be confirmed for the actual operation before it is offered. A freelancer should not treat the audit's informational card as an automatic authorisation or a universal editor. Where support is established, the normal confirmation framework can prepare a reviewable request. Where it is not, the assistant should provide an exact handoff rather than select an unrelated action that appears to address the broad phrase improve mobile performance.
One confirmation can cover up to three clearly separated logical actions. Each step should state what changes and where, with no hidden budget, status or destination operation buried in technical detail. Larger requests should be reformulated into smaller requests. The confirmation should state that no changes have been made yet and that the proposal expires in fifteen minutes. The owner then uses Confirm and Execute for the live action. A freelancer should read the proposal against the latest instruction and current evidence, ensuring that the approved target and setting mean the same thing as the recommendation.
After confirmation, the flow should provide visible progress and a result that distinguishes execution from verification. The owner needs to know that the click was received without interpreting a long technical status message. Supporting provider details can remain behind the scenes. If a current-state check finds that another person changed the target, reconcile that state before proceeding. This is particularly important when the recommendation was based on historical metrics and the mutable resource has since changed. A clear approval flow protects the connection between the business decision, the exact request and the resulting configuration.
Verify settings and outcomes separately
Configuration verification asks whether the approved setting now has the expected value on the resolved resource. It should use the current resource state and relevant parent relationship where needed. That establishes the stored change. It does not establish a future CPA improvement, a repaired website journey or a complete measurement result. A freelancer should state the verified configuration separately from any other claim. If the task concerned a website issue rather than an Ads edit, the verification should test that issue directly instead of relying on a provider response that says nothing about the customer's ability to complete the page task.
Commercial follow-up uses a later defined period and the same agreed measure. Record concurrent changes and any shift in the conversion definition before attributing the result to the action. A device adjustment, a destination repair and a measurement update can all affect interpretation in different ways. The freelancer can acknowledge a completed configuration or journey fix while keeping the performance evaluation scheduled. This gives the owner meaningful progress without overstating certainty. The final report should identify which outcome was observed and which question still needs evidence, with no broad verified label covering work that was never tested.
If execution is acknowledged but readback is incomplete, inspect the current resource before retrying. The change may already have succeeded. Replaying a mutation without reconciliation can overwrite a later update or create an unintended second operation. Preserve the approved intent and available result, then compare them with current state. The same principle applies to a repair whose final message was delayed: communication uncertainty is not proof that nothing happened. A freelancer who reconciles first can give the owner a reliable next step and avoid turning an ambiguous response into repeated edits.
Leave a reproducible client handoff
The final handoff should state the account population, period, conversion definition and device observations that supported the recommendation. Separate contribution and efficiency conclusions. Name the campaign or journey investigated and record the action that followed. If no live edit was justified, say what investigation or recommendation completed the scope. The client should be able to understand the decision without reading raw provider fields. The underlying evidence should remain available for another reviewer to reproduce the calculations and see why the freelancer chose that next step.
Assign remaining work concretely. A measurement test, website check or later commercial review needs an owner and a defined question. Avoid a vague reminder to monitor mobile. Record the relevant dates and conditions so that the next review can compare like with like. If a supported configuration changed, preserve the approval and readback outcome. If a journey changed, preserve the reproduction and verification notes. These records make the account easier to maintain and reduce repeated arguments about whether an improvement came from a device control, a website repair or a change in what the account counted as a conversion.
A strong freelance device audit uses an informational account table as the start of reasoning. It understands the returned categories, summed bases, undefined ratios and account-type presentation. It resolves the business measure, investigates campaign mix and checks the actual cause before proposing an action. It confirms the supported control and target, obtains approval and verifies each claim at the level where the evidence exists. This gives the owner a practical decision they can understand and a record the next manager can use. The result is a disciplined investigation of device performance, rather than a blanket instruction to exclude whichever row looked weakest in one snapshot.