Reduce Google Ads wasted spend for agencies: investigate before you exclude

Define waste for this client

Wasted spend is a conclusion about a business objective, not a synonym for every row with no recorded conversion. An agency needs to establish what the client sells, whom it can serve and which outcome the report measures. Otherwise the same table can support several conflicting recommendations.

A confirmed service mismatch is different from weak measured performance. A retailer that does not sell replacement parts may decide that replacement-parts searches are irrelevant. A relevant product keyword with recent clicks and no purchases needs a performance investigation. Both can deserve attention, but their evidence and proposed actions differ.

Define the review question in language the client can confirm. Which offer, customer need, geographic scope and measured result matter? Record known exceptions, such as a seasonal service or an exploratory campaign with an agreed test period. Do not infer these commercial facts from the campaign name.

Use cautious labels for unfinished analysis. A candidate for review, a confirmed intent mismatch and an approved exclusion are distinct states. This vocabulary makes it easier for another account manager to understand why the agency has identified a problem without treating every observation as an instruction to pause.

The goal is a better allocation decision with an assessable explanation. It is not a large headline about recoverable money obtained by adding every low-return row. Historical cost can identify where attention is needed; it cannot establish what would have happened under a different targeting choice.

Handle the agency's review constraint

The agency-specific difficulty is applying consistent analytical discipline across clients whose commercial definitions differ. One client uses purchases, another counts primary lead actions, and another asks about a particular contact-form conversion. A shared review format helps only if it preserves these differences.

Imagine a fictional agency, Cedar Lane Media, reviewing a specialist retailer and a regional installer. The retailer's irrelevant intent might concern a product it does not stock. The installer's mismatch might concern a service it does not offer. A generic negative list copied between them could block useful demand for one of the clients.

Assign a reviewer who can interpret the evidence and an owner who can confirm business constraints. These can be different people. The data specialist can establish which selected rows spent money; the client contact may need to confirm whether a seemingly irrelevant request is actually part of the offer.

Keep each client's review in its connected account context. Similar campaign names are not proof of a shared account, and a reusable report template is not permission to combine client populations. The target of any proposed action must be resolved within the right account and parent structure.

A useful agency process therefore standardises questions, evidence labels and handoffs rather than recommendations. It makes an unfinished investigation visible and lets the team distinguish verified findings from assumptions awaiting a client's answer.

Choose the measured outcome

Select the outcome before ranking candidates. In ppctor's conversational reporting, aggregate primary conversions, primary purchase conversions and primary lead conversions are different measures. An account's primary conversion value can also contain value from primary actions outside the purchase category.

If the client is asking about sales, specify whether the analysis uses purchase count, purchase conversion value or another agreed action. If the question concerns all primary outcomes, keep that broader definition visible. A campaign can look weak under one selected measure and contribute to another.

For a lead business, do not assign a quality stage based solely on conversion-category wording. The reporting framework groups primary lead categories as leads. A comparison with Analytics needs the corresponding event selected or confirmed; a plausible event name does not prove the implementation records the intended business result.

Named conversion actions are useful when the client wants a specific activity reviewed. Ask for or resolve the action by its name, including ambiguity where several names are similar. Preserve that choice in follow-ups rather than silently reverting to the account-wide conversion total.

Offline and call conversions can be part of a primary outcome definition. Their presence does not imply access to the client's complete CRM or call-quality record. State what the selected provider metric measures and which additional business interpretation still comes from the client.

Select a useful evidence period

Use actual dates and a consistent reporting basis. A rolling thirty-day window and the previous calendar month answer different questions. The period should match the decision being considered rather than whichever date range happened to appear in an earlier dashboard view.

For daily conversion outcomes, the current conversational reporting default uses conversion date. An explicitly requested interaction-date analysis has a different interpretation. Keep the chosen basis stable while comparing candidates and do not describe a timing difference as proof that a campaign's results vanished.

Google's data-freshness guidance describes reporting delays and later updates. Recent low outcome counts should be read with that context. A complete calendar day is not necessarily the final state of every conversion-related metric.

Longer evidence is not automatically better. A long window can combine different offers, budgets or targeting settings. If a major change occurred, consider the period before and after it separately, keeping enough context to explain why the comparison is useful.

Record the included dates and any known change in tracking or offer. The agency then has a basis for a follow-up review instead of choosing a new period that happens to produce a more dramatic result. A repeatable question is easier to defend than a conveniently selected interval.

Keep the candidate population explicit

Define whether the review concerns campaigns, ad groups, keywords, reported search terms or a supported product-performance population. These are not interchangeable rows. A campaign total can contain many lower-level assets, and adding both levels would count the same spending again.

State the status filter. Enabled assets describe the current active structure; paused assets may explain historical spending. A review of currently enabled keywords should not quietly include paused candidates just because they appear in the selected dates.

Distinguish a parent status from a child's own status. An enabled keyword inside a paused ad group does not have the same present serving context as an enabled keyword in an enabled group and campaign. The parent path matters for both interpretation and an eventual action.

Use a complete scoped count when answering how many candidates match. A displayed preview is for readability, not the population denominator. A small list of examples cannot establish the total unless the returned result explicitly covers the complete selected population.

If the available dataset cannot answer the proposed grouping, change the plan transparently. A valid supported read with an explicit population is more useful than a confident claim assembled from unrelated totals. The same scope should remain attached to a chart, table and later change proposal.

Separate zero from missing evidence

A retrieved zero is different from an unavailable value. If the selected metric is absent, unreadable or incompatible with the report, the tool has not established that the asset produced no outcomes. Missing data should remain unknown in the review.

Similarly, an empty candidate result means no rows matched the successful scoped query. It does not prove every possible spending source was examined. Name the population and criteria so the agency can understand exactly what the absence of matches means.

For ratio-based filters, check the required inputs. A cost-per-acquisition calculation with no selected acquisitions is not a normal numeric value that can be compared like a measured CPA. A missing denominator should not become a convenient very large number in a client recommendation.

When the client reports that outcomes exist elsewhere, first reconcile source and definition. Ads primary conversions, a specific Analytics event and a business spreadsheet can have different scopes. Do not treat one source's silence as evidence that all other records are wrong.

Keep these distinctions in the handoff. The next reviewer should see whether a row had a confirmed zero, whether a read was unavailable, and whether an investigation remains open. This prevents a temporary reporting problem from becoming a permanent targeting exclusion.

Test the business-intent mismatch

An intent review connects a reported search or targeted keyword to the client's actual offer. Look for a specific mismatch that the business can confirm: the wrong product, an unsupported service, a different use case or an audience whose request the client cannot fulfil.

Do not rely on a word in isolation. A term that looks informational can still be relevant to a considered purchase, while a commercial-sounding term can refer to a product the client does not carry. Read the full phrase and its account context before proposing an exclusion.

Ask the client about uncertain cases. A short clarification about whether the business sells an item can be more valuable than another broad performance read. The assistant can help structure that question without an API call when it is interpreting supplied commercial context.

Use creative assistance to explore possible meanings, not to invent evidence. The AI can suggest hypotheses about ambiguous intent or draft clearer ad wording. Those suggestions are not retrieved search volumes, verified customer behaviour or proof of an unserved market.

Record the reason for a confirmed mismatch in plain language. For example, an invented installer might confirm that it does not provide equipment rentals. That is a business constraint supporting a review of rental-related targeting; it is not a universal rule for every installer account.

Read search terms within their coverage

Google's search terms report guidance distinguishes actual searches from targeted keywords and notes that some low-activity queries are omitted for privacy. A reported terms list should therefore not be described as every search that generated account spending.

This matters when presenting a total. Spending on the retrieved terms is spending represented in that selected result. It is not automatically the full keyword or campaign cost. Keep the total's coverage visible instead of forcing the rows to reconcile by inventing missing searches.

Review terms in relation to their targeting and parent context. A search observed in one campaign may be useful there but unsuitable elsewhere. The desired scope of an exclusion depends on the client's structure and confirmed offer, not only the text of the query.

A reported search with no selected outcomes can be a candidate for investigation. The next question is whether its intent is relevant and whether the period is sufficient for the decision. It does not become an exact exclusion automatically because it appears near the top of a cost-sorted list.

Keep reported terms and brainstorming separate. The agency may want possible exclusions drafted from the offer, but those drafts should be labelled as suggestions. If actual search evidence is requested, use a supported returned dataset rather than presenting generated phrases as account observations.

Choose negative-keyword scope carefully

Google's negative-keyword documentation explains that negative matching differs from positive matching. An exact negative concerns the exact term sequence without additional words; it is not a general exclusion of every related search or close variant.

The current conversational negative actions support exact negatives at a campaign, ad group or attached shared set. Read the proposal's match type and target instead of assuming the tool has selected broad exclusion coverage. A request for a different unsupported operation needs clarification or another appropriate workflow.

Choose the narrowest scope that matches the confirmed business rule. An irrelevant term in one ad group may be useful in another. A shared set can affect several campaigns, so the review must consider its existing attachments rather than treating the list name as sufficient context.

Separate adding a negative from changing an existing exclusion. They are distinct actions with different target evidence. A proposal should identify exactly what will be added or changed, and the human reviewer should not have to infer the resulting match behaviour from a vague cleanup label.

Do not estimate guaranteed savings by multiplying an exclusion's historical cost into the future. Traffic and allocation can change. The supported conclusion is that the specified exclusion was prepared and, after approval, executed and checked—not that every previously observed cost will disappear from the next report.

Treat audit findings as investigation leads

An audit can direct attention to settings or structure that deserve review. ppctor's negative-keyword check examines the presence of negatives in relevant enabled campaign contexts. Presence does not establish that the exclusions are appropriate, and absence does not quantify wasted spending.

A low Quality Score finding is another diagnostic lead. Google's Quality Score guidance treats the displayed score as a diagnostic rather than a business-performance KPI. Read it alongside the offer and measured outcomes instead of pausing a profitable asset solely because the number is low.

An empty ad group finding concerns missing enabled ad structure in its selected context. An asset with no serving ads cannot prove that it consumed avoidable cost during the chosen period. Historical performance and current structure are related questions that need separate evidence.

Keep the audit's unavailable sections visible. A partially completed read should not become a clean bill for unread areas. The agency can investigate a supported finding while acknowledging that another part of the assessment remains unresolved.

Connect each diagnostic to a next question. Does the negative list cover the confirmed mismatch? Does the landing page fulfil the advertised offer? Does the low-score keyword produce the client's selected outcome? This produces an actionable review without turning a checklist into an automatic profitability verdict.

Review geographic intent and serviceability

A client can have a location rule that is commercially clear: serve a particular area, accept customers elsewhere or attract visitors planning a trip. Start with that rule. A geographic setting cannot be judged properly without understanding whom the business intends to reach.

The audit's location-setting recommendation is a heuristic that needs this context. A setting involving location interest can be intentional for some businesses. Its presence alone does not establish the amount spent outside a service area or prove that all such traffic was irrelevant.

Ask for a supported geographic performance view when the decision needs observed data. Keep the geographic dimension's meaning explicit rather than substituting target settings for user-location performance. What the account targets and what the selected report measures are different facts.

If the client cannot fulfil an identified demand, document that constraint before proposing a setting change. A national retailer and a local service provider can require different decisions even when their reports contain the same geographic label.

Review the exact campaign and operation. A targeting proposal should not quietly rewrite every campaign because one area is unsuitable for one offer. Keep the client's confirmed serviceability rule attached to the selected target and check the resulting setting after execution.

Understand the asset and its parents

A keyword belongs to an ad group within a campaign. A product-listing path may belong to a Performance Max asset group or a Standard Shopping ad group. Read the complete supported parent context before turning a candidate into an edit target.

Names can repeat. A keyword phrase or ad group name alone may be ambiguous across the account. The client-facing proposal should use understandable asset and parent names so the human can assess the intended target without a list of technical identifiers.

Do not mistake a parent action for a child action. Pausing a keyword leaves other keywords in the ad group available; pausing the group has wider consequences. The operation should match the reviewed evidence and be stated clearly in the confirmation.

For an enabled-asset review, check the relevant parent statuses as part of scope. Current structure can explain why an asset is not presently eligible to serve, while the historical period can still contain cost from earlier activity. Do not merge these facts into a single unexplained status label.

Keep target resolution separate from creative planning. The assistant can help discuss how a structure should be organised, but a live change requires the supported existing target or an explicit creation plan. A conversational reference is useful only when it resolves unambiguously in the current account.

Review product exclusions in context

A shopping review needs to distinguish performance from current listing-group configuration. An individual product can appear in more than one group, with different inclusion or exclusion decisions. A product identifier alone does not establish its complete serving context across the account.

For supported listing-tree questions, the reporting framework retains the group and product occurrence. In Performance Max, the relevant parent is an asset group; in Standard Shopping, it is an ad group. Reviews restricted to enabled groups should keep that restriction explicit.

A count of excluded product occurrences is therefore not necessarily a count of unique inventory products. The same item excluded in two enabled groups contributes two group-specific occurrences. Broad category exclusions are also different from individually identified excluded products.

Use performance filters only where the selected performance evidence supports them. An excluded item with unavailable revenue or cost should not be assigned a zero ROAS. A historical result is also not proof that the currently excluded item will resume the same performance if enabled.

If a supported inclusion or exclusion change is requested, resolve the precise listing path and parent. Do not turn an account-wide exclusion count into a blanket enable action. The business decision may differ across groups that use the same inventory for different campaigns or offers.

Compare candidates without distorting ratios

Compare like populations using the same outcome definition and dates. A campaign-level CPA and a keyword-level CPA can be useful in separate views, but combining their rows does not create an independent set of acquisitions or cost.

For an aggregate ratio, add the relevant inputs before division. Total selected value divided by total selected cost produces the combined ROAS for that population. Averaging row-level ROAS equally can overemphasise a small row and misrepresent where the client spent most of its money.

Keep counts and values distinct. A low number of purchases with high selected purchase value is different from many purchases with low value. The client can review both facts without the assistant choosing an unannounced winner based on whichever measure looks most dramatic.

Use inclusive conditions when the request calls for them. Cost greater than or equal to a chosen amount and outcomes less than or equal to a chosen count should preserve their stated boundaries. Multiple criteria should also preserve whether the user meant both conditions together or either condition.

Write the criteria beside the result. A candidate list based on a minimum cost and zero selected outcomes is a defined analytical selection. It is not a universal waste detector. The agency still needs intent, timing and commercial context before recommending a live change.

Work through an invented agency review

The following figures and names are invented to illustrate a review; they are not a customer result. Cedar Lane Media has a client selling new garden equipment. The client confirms it does not rent equipment and wants the agency to review recent purchase performance.

Suppose the selected report uses primary purchase conversions over a defined complete-day window. Three hypothetical candidates are shown below. The labels describe the reviewed intent, not a claim that every query in the account was visible or that the entire campaign has the same intent.

The first row has both a recorded zero and a client-confirmed offer mismatch. The second has purchases and needs its value context. The third has no recorded purchases but still needs an intent decision. Adding the first and third costs would not establish 360.00 of recoverable savings.

The agency can prepare a precise supported exclusion for the confirmed rental mismatch after resolving its target. It should leave the advice row as an open review rather than claim the same certainty. This makes the recommended action smaller, clearer and tied to the evidence actually available.

Prioritise a bounded investigation

Start with candidates that combine material selected cost and a specific unresolved question. Cost sorting can help locate them, but the order is an investigation priority rather than an automatic action sequence. The highest cost row may also be the client's most valuable source of purchases.

Separate straightforward business mismatches from measurement disputes. A confirmed unsupported offer can support a targeting review. A sudden absence of every measured outcome across previously active campaigns may require checking the selected definition and tracking context before individual keywords are blamed.

Limit each review to a coherent decision. Mixing keyword exclusions, budget changes, creative replacements and product inclusion into one vague cleanup task makes approval difficult. The human should be able to understand each proposed step and its dependencies.

Record an unresolved question rather than expanding the read indefinitely. The next useful input may be a client's confirmation, a named outcome choice or a landing-page URL. Another account-wide query does not answer a commercial fact that only the business knows.

Use a queue of assessed candidates with an owner and next step. This is an agency working method, not a claim that ppctor creates an external task-management system. It helps the team turn a bounded report into accountable follow-up without inventing integrations or automatic repairs.

Ask a scoped question in Slack

Begin with the selected outcome, dates and population. A useful request might ask for enabled keywords with at least a stated amount of cost and zero primary purchase conversions during the last thirty days. It gives the assistant a defined read rather than an unexplained instruction to find waste.

The generic reporting tools describe supported fields and construct a validated plan. They can retrieve a compatible dataset, apply stated conditions and retain the result for follow-up presentation. The assistant should not need a special-purpose tool for every wording of a low-performance question.

Ask a clarification when the commercial objective is unclear. If the account has different primary actions, the user may mean all primary conversions or a particular purchase action. Clarifying that choice prevents a seemingly precise ranking from answering the wrong business question.

Follow-up requests should preserve established dates, account and outcome unless the user changes them. Asking for a table of the same candidates does not require a new interpretation of the period. A new measure or grouping can require another compatible read.

Keep creative discussion conversational. Asking for possible explanations or better wording does not automatically request Keyword Planner evidence or a live edit. The assistant can brainstorm from supplied context, while actual account facts and changes remain grounded in their appropriate tools.

Present evidence without a wall of rows

Lead with the scoped result and what it means. A concise response can state the selected period, outcome, candidate count and highest-priority examples. The client should not have to scan technical field names before understanding which question was answered.

Use readable names and parent context. For a keyword candidate, show the keyword with its ad group and campaign where needed. Avoid exposing asset identifiers in the client-facing narrative simply because they were required internally to resolve the target.

A table can compare selected cost, selected outcomes and a clearly labelled review reason. Keep the reason separate from the retrieved metric. The provider can return performance data; the assessment of a service mismatch may come from the client's own confirmation.

When a chart is requested, use compatible returned data and a readable grain. Native conversational charts support bounded series and points, so a long unqualified trend can use a coarser breakdown. If daily detail is explicitly requested, preserve it within the supported layout and explain any necessary split.

Keep displayed metrics to at most two decimals. Missing values should have understandable unavailable labels rather than invented precision or zeros. Readability is part of review safety: the human needs to recognise both the measured facts and the proposed interpretation.

Prepare an exact change proposal

A spending review and a change request are different steps. Once the client confirms the intended action, prepare the supported operation for the exact target. Do not interpret a question about poor performance as authorisation to change the account.

For a negative-keyword proposal, show the text, match type and target scope. For a status change, name the asset and relevant parents. For a campaign budget change, show the current and proposed amounts where the supported operation permits it.

Separate each logical action in the confirmation. A keyword addition and an ad-group status change should not disappear inside a long sentence. The human must be able to see what each step does and which existing or newly created asset it concerns.

Respect supported operations. The ordinary conversational campaign budget action covers non-shared daily budgets, not every budget type. An audit recommendation about another setting does not prove that a universal conversational edit exists for it.

Keep the proposal's untouched state and expiry clear. The live account has not been changed by preparing the request, and the approval has a defined validity period. The reviewer can reject or revise the proposal if the evidence, target or action is not what the client intended.

Approve and verify execution

The human uses Confirm and Execute to approve the prepared edit. Preparation alone does not send the live mutation. The execution path checks relevant authorisation and settings again, applies the supported request and reads back what can be verified.

Current settings can differ from the original proposal because another person changed the account. A recheck protects against applying an operation based on a stale assumption. A previously discussed amount or status should not override the current target evidence silently.

Read the execution result as an account-operation result. A verified pause means the selected asset's status was checked after the request. It does not prove that the client's future cost, revenue or profitability will improve.

If execution is uncertain or not fully verified, investigate the existing account state before retrying. This matters especially for creations, where a request may have succeeded even if the follow-up read was incomplete. Blind repetition can create duplicates or apply an unintended second change.

Separate provider review from configuration verification. An ad's requested content can exist while policy review remains pending. Likewise, a saved targeting change does not establish its later delivery or commercial effect. Each claim should match the verification that actually occurred.

Measure what happened afterward

Choose a follow-up question before executing. For an exclusion, the agency may review the remaining reported terms and selected outcomes under the same account context. For a budget change, it may check observed cost against the agreed period amount.

Use a comparable outcome definition. Switching from purchases before the edit to all primary conversions afterward can make the apparent result misleading. If the business changes its definition, explain that the later measurement answers a different question.

Record other relevant changes and the evidence dates. Offers, creative, tracking and budgets can change during the same interval. An observed improvement after an edit is not by itself proof that the edit caused every part of the improvement.

Do not claim saved money merely because a candidate no longer appears in a preview. The absence may reflect selected coverage, a changed period or insufficient query activity. Check the supported configuration and the scoped performance evidence separately.

A useful follow-up reports what changed, what was measured and what remains unresolved. It can confirm that the selected operation exists and describe subsequent outcomes without promising a counterfactual saving that the report cannot measure.

Write the client and colleague handoff

Write a short decision record that another colleague can use without reading the whole Slack thread. Include the client-confirmed objective, selected evidence period, metric, target and reason for the proposed operation. Keep unsupported assumptions out of the recorded finding.

Distinguish supplied context from retrieved evidence. The client may have confirmed that a service is not offered; the tool may have returned cost for a selected population. Both support the decision, but neither should be misrepresented as a fact verified by the other source.

State whether the action was proposed, approved, executed or independently checked. These stages should not collapse into a generic done label. An unresolved read or execution outcome needs a clear next owner and investigation step.

For client-facing wording, explain the practical change without implementation jargon. Name the asset and relevant parent, then describe the action. The client usually needs to understand what demand will be excluded or what status changes, not the internal API resource representation.

Close the handoff with the planned follow-up period and question. This prevents the next account manager from treating a historical recommendation as a permanent unreviewable rule. Business constraints can change, and the agency needs a record that can be reassessed.

Make the review repeatable

A recurring report can surface defined spending and outcome evidence for the team. It should retain its account, period, population and measurement. The appearance of a low-outcome row is an investigation prompt, not an automatic targeting instruction.

Use thresholds for specific signals when appropriate. An absolute alert, a historical anomaly and a budget projection serve different purposes. None independently proves commercial waste. Choose the signal that answers the agency's monitoring question instead of treating every notification as a profitability diagnosis.

Keep monitoring definitions aligned with the review. A scheduled provider conversion total may differ from the named business measure used in a conversational investigation. Explain that difference before comparing the two outputs as though they share a definition.

Assign a response owner and a fallback colleague. Review incoming evidence, check whether the condition still applies and decide the next supported step. A notification delivered to Slack does not imply that somebody has assessed or approved a change.

Use related guides for the distinct workflows: agency audits for settings triage, agency budget tracking for agreed spending, and Confirm and Execute for reviewed live operations. Keep their responsibilities separate in the agency process.

Agency wasted-spend questions

Does zero conversions mean the keyword should be paused?

No. First confirm the selected outcome, dates, reporting coverage and business relevance. A retrieved zero is a useful observation, but it does not independently establish that the intent is unsuitable or that a pause is the right operation.

Can the agency report guaranteed savings from exclusions?

The historical cost of selected candidates is not a guaranteed future saving. Report the defined cost, the client-confirmed reason and the verified operation. Review later performance separately rather than presenting a hypothetical counterfactual as a measured result.

Can the same exclusion list be used for every client?

A shared format can help the review, but targeting choices depend on each client's offer and structure. Check scope, match type and attachments. A phrase irrelevant to one business can be useful for another.

Does the AI need an API call for every explanation?

No. It can discuss supplied context, suggest hypotheses and draft wording conversationally. Account metrics need returned tool evidence, and live edits need the supported proposal and human confirmation path.

What should the agency retain after an edit?

Retain the objective, evidence definition, named target, exact action, execution state and next review. That record makes the decision understandable across colleagues without exposing unnecessary technical identifiers or claiming an unmeasured business effect.