To improve Google Ads lead quality, identify why each enquiry fails before changing the campaign. Separate spam, repeat requests and existing customers from genuine new demand. Then distinguish people you cannot serve from people your team has not yet reached. Connect those outcomes to the campaign, advertised service, landing page and intake process. The resulting pattern tells you which change is worth testing.
A form submission does not prove that someone wants the service you sell. A missed call does not prove the opposite. When both appear in a report as “bad leads”, marketing and sales can spend weeks fixing the wrong problem.
This guide provides a qualification rubric, a feedback taxonomy, a copyable sales-feedback CSV and a hypothetical campaign review. It is designed for service-business owners, paid-media managers and intake teams. The task here is diagnosis: establish what happened between the click and the conversation, then choose a controlled test.
The decisions that matter
- Define a qualified enquiry using observable service requirements, not whether the salesperson liked the conversation.
- Keep lead fit, contact handling and sales outcome in separate fields.
- Review actual enquiries alongside campaign evidence; a conversion total cannot explain suitability.
- Keep unknown fit unresolved, even after follow-up ends without contact.
- Test one suspected cause at a time and watch qualified volume as well as percentages.
Agree on a qualification rubric before reviewing campaigns
A qualified enquiry is a genuine request from someone whose need the business can reasonably serve and whose request meets its agreed commercial criteria. It is evidence of relevant demand, not a guarantee of a booking or sale.
For a residential painting company, the service could be interior painting, the service area could be specified municipalities, and the minimum scope could be a defined type of project. For an accounting firm, suitability might depend on the service requested and the type of business. The criteria should describe what you deliver, without guessing at a person’s circumstances from their name, accent or neighbourhood.
Have the owner, campaign manager and intake lead complete this worksheet together. Replace every example with the business’s actual offer. Do not let an undocumented preference become a rejection rule after enquiries arrive.
| Requirement | Evidence to collect | Decision rule |
|---|---|---|
| Genuine request | A coherent request for help, supported by the form or conversation. | Reject as spam only when there is evidence of a non-genuine submission. Poor spelling is insufficient. |
| Service match | The requested service, in the prospect’s own terms or a confirmed category. | Pass if the business supplies it. Fail if it does not. Clarify ambiguous requests. |
| Serviceable location | Where the work or service must be delivered. | Use the delivery location, not the caller’s area code or current location. |
| Scope and commercial fit | Project type, minimum scope or another published, relevant requirement. | Fail only against an agreed rule. An unanswered budget question remains unknown. |
| Timing and next step | Required timing and the appropriate consultation, estimate or follow-up. | Distinguish a future opportunity from an impossible deadline. Immediate purchase is not always required. |
We build and run Google Ads campaigns measured on calls, forms and booked jobs, not clicks.
Use gates rather than an unexplained score out of 100. If the requested service is unavailable, enthusiasm cannot make the lead qualified. Conversely, a prospect should not lose points because they prefer email or need time to compare estimates.
Write one operational qualification rule
For example: “Qualified means a genuine new request for an offered service, within our delivery area, meeting the stated project requirements, with enough confirmed information to select a next step.” Record the rubric version and its effective date.
Keep capacity separate from fit. A suitable customer who cannot book because your team is full may represent qualified demand lost to capacity. If your campaign promises same-day service that you cannot provide, record that expectation problem too. Do not quietly redefine those customers as unsuitable.
Calibrate the rubric with a small, deliberately varied set of redacted enquiries. Ask two reviewers to classify them independently, compare disagreements and tighten the definitions. This exercise is useful even when the business has too few leads for a reliable campaign experiment.
Resolve borderline cases with evidence
Consider four fictional requests before approving the rubric:
- A tenant enquires about work requiring the owner’s approval. Do not automatically reject the tenant. Confirm whether approval is needed at this stage and whether it can be obtained. The request may be pending clarification rather than low fit.
- A prospect cannot name the technical service. Record the problem they want solved and let a qualified team member route it. Customers do not need to use your campaign terminology to represent suitable demand.
- A caller asks for a price range before booking. Price comparison alone does not fail the rubric. Establish the service and scope, explain the estimate process, and record any actual mismatch.
- A suitable prospect needs a later start. Use a future follow-up stage when the business accepts planned work. Reserve a timing failure for a requirement the business truly cannot meet.
These decisions matter because an overly strict rubric can make marketing appear inefficient while discarding reasonable buyers. An overly loose rubric can make every polite conversation look qualified. Review exceptions together, record the rule you settle on and apply it prospectively. If a rule must change, retain the earlier version so comparisons remain interpretable.
Use a feedback taxonomy that explains the failure
Every contact needs a current lead state and a reason. It may also need handling flags. The state answers whether the enquiry belongs in the new-demand review and whether its fit is known. Flags describe what happened during follow-up. This prevents a missed call from being counted again as a separate unsuitable lead.
| Category | What it means | First diagnostic check |
|---|---|---|
| Spam or non-enquiry | Automated nonsense, malicious submissions or unsolicited pitches unrelated to buying the advertised service. | Form route, repeated patterns, delivery logs and technical safeguards. |
| Wrong geography | The required service location is outside the agreed coverage. | Location settings, actual service address category and geographic promises. |
| Wrong service | The person requests something the business does not provide. | Search intent, ad wording, destination page and service selection. |
| Low fit | A confirmed scope, timing or commercial requirement fails, with no confirmed service or geography failure taking precedence. | The failed rule, the evidence and whether the requirement was communicated. |
| Duplicate | A second contact about the same active request. | Original request ID, repeated form events and cross-channel contact history. |
| Existing customer | A known customer seeking support, a change or further work. | Customer relationship and whether the campaign is meant to acquire new customers. |
| Uncontactable | The approved follow-up process has ended without establishing contact, and fit remains unknown. | Delivery failures, permitted contact channels and completed attempts. Preserve any already-known fit decision. |
| Missed call | An incoming call was not answered by the intended team. | Routing, opening hours, queues and callback completion. This is a handling flag. |
| Sales response issue | A handoff, response or follow-up failed the agreed process. | Assignment, response timestamps and next-step ownership. This is also a flag. |
| Qualified demand | The enquiry meets the documented rubric. | Preserve the evidence and track the later sales outcome separately. |
Classify in a consistent order
First identify spam, duplicates and existing customers. Retain those records for diagnosis, but exclude them from the denominator for unique new prospective requests. Keep a duplicate linked to its original request; never erase the fact that it generated another call or form.
Next assess the remaining requests. Any confirmed failure of a required gate makes the request unqualified, even if another gate is unknown. When several failures are confirmed, choose the main reason in this order: wrong geography, wrong service, then low fit. Record other observed mismatches in a restricted note when useful. This precedence keeps totals tidy; it does not identify which factor caused the click.
Use qualified when all applicable required gates pass. With no confirmed failure but missing required evidence, use unresolved: pending review while checks or follow-up remain, or uncontactable after the agreed contact attempts end. If a form already proves the service is unavailable, use unqualified with wrong_service even if later calls go unanswered. A known, suitable request remains qualified if the person subsequently stops replying.
Handle exceptions without moving the goalposts
A caller asking for a quotation and later choosing a competitor remains a qualified enquiry if the rubric was met. “Too expensive” after a detailed estimate is a sales outcome unless a genuine minimum requirement was clearly established and failed during qualification.
A repeat customer buying another service may be valuable. Report that demand separately from new-customer acquisition. Similarly, two people from one company might represent one project or two independent requests. Define duplication around the buying request, not merely a shared domain or telephone number.
Reconcile the units before judging quality
Write down what each system counts. Google Ads may report conversion actions. Your telephone system records calls. Your forms record submissions. Your CRM might hold contacts, requests and opportunities. None of those units automatically equals a unique qualified enquiry.
- Preserve raw contact events. Keep the incoming call or submission reference, timestamp, channel and source evidence.
- Group events into requests. Link repeated contacts concerning the same need, using information held in the operational system.
- Identify the review population. Separate spam and existing-customer contacts from unique new prospective requests.
- Apply the rubric. Count qualified, confirmed unsuitable and unresolved requests.
- Track the next stage. Record booked, quoted, won or lost without overwriting the original fit decision.
Inspect whether a form-success event fires twice, a button click is counted as a completed enquiry, or one submission creates several CRM records. An apparent quality decline may begin with a measurement change rather than a different audience.
Google Ads offers “One” and “Every” conversion-counting options. “One” can suit a lead action, but it counts in relation to an ad interaction and conversion action; it does not deduplicate every person or request across your CRM. Review the intended unit before changing it. See Google’s conversion-counting guidance.
Check what the campaign treats as success
List each conversion action, what triggers it, whether it is primary or secondary, and which goals the campaign uses. In standard goals, primary actions are used for bidding when their goal is selected; secondary actions are generally observational. Custom goals are an important exception: included secondary actions can also be used for bidding. Google explains this in its primary and secondary conversion documentation.
A campaign optimizing towards initial enquiries has not necessarily received evidence of sales qualification. Likewise, moving a weak signal to secondary is not a complete repair if a custom goal still uses it. Document the discrepancy and plan the measurement change separately. Implementing offline conversion feedback requires its own technical work; this worksheet is not an upload specification.
Choose a cohort based on when enquiries arrived and allow a consistent time for follow-up. Comparing yesterday’s unanswered forms with last month’s fully reviewed requests makes the newer cohort look worse before it has had a fair chance.
Trace the pattern from the enquiry back to the campaign
Start with one recurring failure, such as wrong-service requests. Assemble the relevant records and compare them with qualified enquiries from the same period. Look for a shared campaign, service, page or contact route. Keep unmatched records visible; an incomplete source trail should not turn into a confident attribution.
Give source evidence a confidence level
Use “linked” when a permitted tracking reference connects the enquiry to campaign data. Use “reported” when the prospect says they came from an advertisement. Use “unknown” when neither is available. A remembered Google search may refer to organic results, Maps or an ad, so self-reporting deserves its own label.
A captured keyword parameter is not necessarily the person’s actual search query. Nor should an analyst assign a particular search term to an individual simply because the timestamps seem close. Keep person-level evidence distinct from patterns observed in aggregate reports.
Use search intent as one diagnostic clue
Compare available search-term themes with intake reasons. Employment or another-service searches may explain a repeated mismatch; a price question alone does not establish poor fit. If this is the dominant pattern, continue with the high-intent search-query guide for query review and exclusions. Keep the enquiry classifications as the evidence for that work.
The search terms report is incomplete: Google omits some low-activity queries for privacy, while search terms insights may aggregate them without showing the individual queries. The search terms report documentation explains this limit. Aggregate query patterns cannot establish what a particular unlinked prospect searched.
Diagnose geography using the service location
Compare the target area, the campaign’s location options, the locations described in the ad and page, and the location where the customer needs service. A property owner calling from another province may still need work at a property inside your coverage.
Google distinguishes “Presence or Interest” from “Presence”. The former can include people interested in the target location; the latter focuses on people likely to be in or regularly in it. For a local service with repeated out-of-area requests, Presence may be worth testing. It is not universally preferable: relocation services and businesses serving remote property owners can need interest-based reach. Google also states that location accuracy is not guaranteed. Review its advanced location options against the actual delivery model.
Do not exclude a municipality because two unsuitable requests happened to come from there. Look at the service requested, qualified enquiries from the same area and whether the business’s boundary is understandable. A vague “serving the GTA” message can create expectations beyond the routes a team actually covers.
Use segments to ask better questions
Where reporting supports it, compare campaign type, network, device, time of day and contact route. A segment is a clue, not a verdict. Mobile enquiries might appear weaker because a form hides the service limitations on a small screen. Evening calls might appear weaker because nobody returns them until much later.
Show record counts beside percentages. A segment with three classified requests cannot support the same confidence as one with a substantial, consistently reviewed history. Avoid combining Search and Performance Max into one conclusion when their inventory and available controls differ.
Check the promises between the ad and the enquiry
Read the ad, landing page and form as one sequence. At each step, write down what a reasonable prospect would believe they can obtain. Compare that belief with the service the intake team is actually prepared to offer.
Consider two hypothetical mismatches: a business stops commercial work but its page still promises “Commercial and residential repairs”; or a page offers a “free consultation” when the first appointment is paid. Both can attract enquiries based on an outdated promise. Clarify the offer before treating those enquiries as careless.
Inspect the page visitors actually reach
Check served destinations where reporting makes them available, including automated URL choices. In Performance Max, Final URL expansion can send traffic to another relevant page on the domain. Google documents URL exclusions and the option to turn expansion off. A page feed alone does not restrict destinations while expansion remains on. See Google’s Final URL expansion guidance.
Search campaigns using AI Max can also have Final URL expansion enabled. Review that configuration only if it applies to the campaign; ordinary Search campaigns should not be assumed to use it. Google’s Search URL expansion documentation describes the option.
If recruitment enquiries are arriving through a careers page, the proposed test may be a destination exclusion. If visitors land on the correct service page but misunderstand the offer, the page wording may be the better intervention. These are different hypotheses.
Ask only questions that change the next step
A useful form question routes or qualifies a request: service needed, delivery area, project category or timing. An unnecessary question adds friction without improving a decision. Prefer a clear choice with an “unsure” option over forcing a prospect to invent an answer.
Explain relevant limitations near the form. If a minimum project scope genuinely applies, state it plainly. Avoid making the prospect disclose private financial details just to establish whether a consultation is appropriate. Test the mobile form, error messages and confirmation step after any change.
Google-hosted lead forms have their own options and eligibility requirements. Where available, “More qualified” introduces more submission steps than “More volume”; it does not certify that someone meets your business rubric. Google’s lead-form guidance describes the trade-off. Supported qualifying-response features can also label submissions from selected answers; these are currently documented for Search campaigns. Keep that response-based qualification separate from sales-confirmed fit.
Treat spam as a specific technical problem
For repeated fake submissions, compare the form route, timing patterns and validation results. Consider proportionate server-side validation, rate controls or an accessible bot challenge, with a legitimate-user check after implementation. Do not treat an unfamiliar email provider as proof of fraud.
Google explains that its invalid-traffic protections do not by themselves prevent fake website leads. A poor enquiry is therefore not automatically evidence of a billable invalid click or entitlement to a credit. Its invalid-lead guidance separates advertising protections from the advertiser’s form safeguards.
Separate call quality from the team’s ability to respond
For each call, determine whether it reached the correct destination, whether someone answered, whether a conversation occurred and whether the request could be assessed. A telephone system accepting a call is not the same as an intake specialist speaking with the caller.
Google call reporting can provide details such as start time, duration and connected status when the feature and forwarding numbers are supported. Those details help locate failures, but duration alone does not establish suitability. A long call may contain hold time; a short call may efficiently confirm a service mismatch. See Google’s call reporting documentation.
Google also documents AI-qualified call leads, which assess recorded conversations and can fall back to other signals when recording is unavailable. Its label need not match your business rubric. Check whether recording is already enabled: Google currently documents it as on by default, with exceptions. Recording is currently available only when both the dialling and receiving numbers are in the United States or Canada. Review the current feature conditions, actual account settings and applicable recording requirements before using it.
Check the delivery route as well as the submission. A lead can exist in a form log while its notification lands in junk mail or its CRM assignment fails. Compare received, delivered and assigned timestamps before deciding the person ignored your response. A delivery failure belongs in the handling investigation; it should not be relabelled as poor audience quality.
Create a response workflow that can be audited
- Assign each new request to a named role or queue, with a backup for absence.
- Record the first genuine response attempt separately from an automated acknowledgement.
- Use the permitted contact channels and the customer’s stated preferences.
- Set the next action and due time after every attempt.
- Close as uncontactable only after the agreed process is complete; retain unknown fit.
Choose response expectations that fit the service, staffing and promise made to customers. Do not borrow an unsupported universal “five-minute rule” or prescribe repeated calls regardless of consent and preference. Urgent repair enquiries and planned renovation consultations need different handling.
If missed calls cluster during staffed hours, inspect queues and routing before reducing ad exposure. If a response is delayed because assignments fail, repair the handoff. Restricting ad hours may be appropriate when the offer requires immediate telephone service, but it could also remove valid enquiries that would happily schedule a callback.
Use the CRM to preserve these distinctions. A practical CRM setup gives each request an owner, a next step and a reason for its current status. It should make unanswered enquiries visible rather than automatically burying them under “lost”.
Use a small, consistent sales-feedback CSV
The feedback file should help marketing understand outcomes without becoming a second customer database. Keep names, phone numbers, email addresses, precise addresses and detailed conversations in the authorized operational system. Use a pseudonymous request ID to connect the review back to that system when necessary.
The following CSV uses fictional records. Copy the header into a UTF-8 CSV file, remove the example rows and apply controlled choices in your spreadsheet or CRM. This is an internal review template, not a Google Ads conversion-import file. Set ID and timestamp columns to text when importing so the spreadsheet preserves their values.
record_id,request_id,received_at,campaign_ref,source_confidence,channel,service_category,lead_state,main_reason,service_fit,area_fit,scope_fit,handling_flags,first_response_at,contact_attempts,sales_stage,reviewed_at,rubric_version,note_code
EX001,REQ001,2026-09-07T13:00:00Z,CAMP_A,linked,form,office_cleaning,qualified,qualified,pass,pass,pass,none,2026-09-07T13:20:00Z,1,consultation_booked,2026-09-08T16:00:00Z,R1,criteria_confirmed
EX002,REQ002,2026-09-07T14:00:00Z,CAMP_B,linked,call,unknown,unresolved,pending_review,unknown,unknown,unknown,missed_call,2026-09-07T14:30:00Z,1,follow_up_due,2026-09-08T16:00:00Z,R1,callback_unanswered
EX003,REQ003,2026-09-07T15:00:00Z,CAMP_A,linked,form,office_cleaning,unqualified,wrong_geography,pass,fail,unknown,none,2026-09-07T15:40:00Z,1,closed,2026-09-08T16:00:00Z,R1,outside_service_area Keep the field rules beside the file
- record_id and request_id: keep one row per contact event. Use the first event as the original request row, update its fit as evidence arrives, and mark later same-request events duplicate. Count only the original row in fit summaries. Combine handling flags across the request without counting another lead.
- campaign_ref and source_confidence: use a stable campaign reference with linked, reported or unknown evidence. Leave the campaign unknown when attribution is uncertain. If a request has contacts linked to different campaigns, apply one documented attribution rule to the request summary; do not count it once per campaign.
- lead_state: use spam, duplicate, existing_customer, qualified, unqualified or unresolved.
- main_reason: pair spam with spam_non_enquiry; duplicate with duplicate_same_request; existing_customer with existing_customer; and qualified with qualified. For unqualified, use wrong_geography, wrong_service or low_fit. For unresolved, use pending_review or uncontactable. Keep one main reason per record.
- Fit fields: use pass, fail, unknown or not_applicable. service_fit covers the offer, area_fit covers delivery location, and scope_fit covers the agreed project, timing and commercial requirements. Use not_applicable only when the rubric excludes that gate. Never use blank to mean both unknown and passed.
- handling_flags: use none, missed_call, response_late, routing_failure, assignment_failure, delivery_failure or follow_up_missed. Combine flags with a pipe when necessary; never combine none with another flag.
- Timestamps and contact_attempts: examples use UTC, shown by Z. Use one declared time zone and retain the offset. Leave first_response_at empty until a genuine attempt occurs, and start contact_attempts at zero. Count attempts consistently; automatic acknowledgements do not count.
- sales_stage and note_code: separate the commercial next step from qualification. Use brief controlled notes rather than copied conversations.
Maintain a change history in the source system when classifications change. The CSV can hold the current review snapshot, with a review timestamp and rubric version. Do not overwrite last month’s exported snapshot and then claim the historical results were always the same.
Restrict access and retention to what the review requires. Pseudonymous IDs and exact timestamps can still be linkable to individuals. Remove unnecessary detail from shared summaries, and avoid copying health, legal, financial or other sensitive narratives into campaign notes. Import external text as text so a spreadsheet does not interpret a submitted value as a formula.
Worked example: two campaigns, different problems
The business, campaign labels and every number below are invented for illustration. This is not a Canada Create client case, a benchmark or a result from a test. Imagine a commercial cleaning company reviewing a completed enquiry cohort from two campaigns: A promotes office cleaning broadly; B promotes a more specific office-cleaning service.
For simplicity, this hypothetical cohort covers enquiries received from 1 to 7 September 2026. Each request is classified using its first seven days of follow-up, and the cohort is compiled on 14 September. It assumes reliable campaign links and a consistent rubric. Real reviews must show missing source data and differences in review age.
| Measure or classification | Campaign A | Campaign B | Total |
|---|---|---|---|
| Ad clicks | 600 | 400 | 1,000 |
| Raw contact events | 70 | 50 | 120 |
| Spam or non-enquiries | 9 | 3 | 12 |
| Duplicate contact events | 5 | 3 | 8 |
| Existing-customer contacts | 6 | 4 | 10 |
| Unique new prospective requests | 50 | 40 | 90 |
| Wrong geography | 10 | 2 | 12 |
| Wrong service | 9 | 5 | 14 |
| Low fit | 5 | 3 | 8 |
| Qualified demand | 16 | 22 | 38 |
| Pending review | 6 | 4 | 10 |
| Uncontactable; fit unknown | 4 | 4 | 8 |
The first reconciliation is 120 contact events minus 12 spam, eight duplicates and 10 existing-customer contacts, leaving 90 new prospective requests. The second is 38 qualified plus 34 confirmed unsuitable plus 18 unresolved, also totalling 90. The mutually exclusive categories prevent the same request from appearing in several rejection totals.
Campaign A produced 70 contact events per 600 clicks, a ratio of approximately 11.7%; B produced 50 per 400, or 12.5%. These are event-to-click ratios, not the percentage of people who converted: repeated contacts are included. The ratios look close but do not describe the same mix of demand.
Confirmed qualified demand represents 16 of A’s 50 new prospective requests, or 32%, compared with 22 of B’s 40, or 55%. Across both campaigns, it is 38 of 90, approximately 42.2%. Another 20% of the review population remains unresolved. These are descriptive percentages for this invented cohort, not recommended targets.
If you report only resolved fit decisions, the calculation changes to 38 divided by 72, approximately 52.8%. That denominator excludes the 18 unknowns. Label it clearly and show the unresolved count beside it; otherwise the higher percentage can hide a response problem.
Translate the review into a hypothesis
Campaign A’s geography failures justify checking its location options and service-area wording. They do not prove that a particular setting caused the failures. Some people may have requested work elsewhere despite correctly seeing a local ad.
Suppose the separate handling log also shows seven missed-call flags within the 18 unresolved requests. Those seven are already in the table; they are not seven additional bad leads. Improving callback handling could reveal either qualified or unsuitable demand. Until contact is established, the outcome remains unknown.
The next action is to investigate the strongest repeated pattern and select one test. Immediately transferring all budget from A to B would assume that B can absorb extra spend while preserving its observed mix. The example supplies no evidence for that assumption.
Run a disciplined test of one suspected cause
Write a test card before making changes. The card connects an observed problem to an intervention, a success measure and a decision rule. Google’s experimentation guidance recommends testing one variable at a time and choosing success metrics in advance.
| Field | What to write |
|---|---|
| Observed problem | Repeated requests requiring work outside the company’s service area. |
| Evidence | The reviewed request IDs, cohort dates, failed rubric gate and relevant campaign settings. |
| Hypothesis | Interest-based geographic reach contributes to out-of-area requests. |
| Single intervention | Test the applicable location option while keeping target locations and other settings stable. |
| Primary measure | Wrong-geography requests divided by unique new prospective requests, with counts shown. |
| Guardrails | Qualified-request volume, unresolved share, spend and delivery. Avoid accepting a prettier percentage caused by losing useful demand. |
| Review conditions | Defined exposure period, comparable follow-up age, stable rubric and enough reviewed requests to assess uncertainty. |
| Decision | Retain, reverse or mark inconclusive, using the rule agreed before launch. |
Follow the same sequence for each test
- Establish a usable baseline. Reconcile contact counts, source coverage and outcome completeness. Repair a broken form or failed routing before treating the period as stable.
- Select one intervention. Choose a negative-keyword change, a location option, a clearer service limitation, one qualifying question or one response-process change. Do not bundle them.
- Choose the comparison. Use a supported campaign experiment when the intended setting and campaign type are eligible. Verify that experiment-arm information can be connected to qualified outcomes.
- Set the decision rule. Agree what size of improvement would matter, what qualified-volume loss is unacceptable and how long follow-up needs to mature. These are business-specific choices.
- Log changes and interruptions. Record the start, affected scope, staffing changes, outages and overlapping edits. Necessary urgent repairs take priority, but may invalidate the comparison.
- Review and decide. Compare counts and uncertainty, then retain, reverse or continue collecting evidence under a documented extension.
If the outcome cannot be linked to experiment arms, the platform’s conversion result alone cannot establish which arm improved sales-qualified demand. If a controlled experiment is unavailable, use a dated before-and-after comparison and describe it as directional. Seasonality, auction changes and intake availability can explain part of the difference.
Do not declare a winner after the first favourable day or invent a universal minimum number of leads. Plan sample requirements around baseline frequency and the smallest change worth detecting. Low-volume businesses may need a longer observation period or may only be able to justify a cautious operational decision. “Inconclusive” is a useful result.
Stabilize feedback definitions before testing bidding or conversion-goal changes. Otherwise the measurement itself changes halfway through the comparison. Review platform learning status and conversion delay where relevant, and allow the sales review to catch up before judging the result.
Be precise about what the evidence cannot prove
Attribution connects recorded outcomes to recorded interactions under a chosen method. It does not prove the enquiry would not have occurred without the advertisement. A person may have encountered referrals, organic search or an earlier campaign before submitting a form.
Tracking gaps, consent choices, device changes, reporting delays and mismatched dates can leave incomplete connections. A campaign-level pattern cannot fill those gaps for individual people. Keep an unattributed group and report how much of the cohort it represents.
Outcome review can also be biased. Sales teams may follow up more quickly on familiar services, or reviewers may judge one campaign more harshly after seeing its cost. Where practical, classify fit before showing campaign performance. Review a sample of both qualified and rejected requests for consistency.
A rising qualified percentage is not automatically improvement. It can rise because qualified volume increased, unsuitable volume fell, unknowns were excluded or valid prospects abandoned a harder form. Report the underlying counts and handling completeness so each explanation remains visible.
Make the review small enough to maintain
Assign intake responsibility for factual outcomes, the campaign manager responsibility for source patterns, and the owner responsibility for service and commercial rules. Review unresolved requests first, then repeated rejection reasons, then the status of the current test.
Before closing the review, confirm:
- The rubric, cohort dates and follow-up window are recorded.
- Raw contacts reconcile to excluded events and unique new requests.
- Qualified, unqualified and unresolved totals reconcile; handling flags stay separate.
- Missing attribution, unreviewed records and unresolved fit remain visible.
- One action has an owner, a review date and a qualified-volume guardrail.
Preserve the evidence that would change your mind. If the main problem is an unclear offer or a missed handoff, fix that process before asking the campaign to produce a different kind of customer.
For businesses reviewing their Google Ads management, the most useful brief is a reconciled set of enquiry outcomes and a clear qualification rule. It gives everyone the same question to answer: where did a relevant click become an unsuitable request, an unresolved contact or a qualified opportunity?
Sources and review method
Factual review: 8 October 2026. Google product statements were checked against the official documentation below. Features and eligibility can change; this is documentation-based guidance, not a test of a particular advertising account. The rubric, classification rules, CSV and test card are editorial tools. All worked records and campaign numbers are hypothetical; no private client data or claimed client results are used.
All sources are Google Ads Help pages, accessed 8 October 2026:
- About conversion counting options.
- About primary and secondary conversion actions.
- About the search terms report.
- About advanced location options.
- About Final URL expansion in Performance Max.
- About Final URL expansion in Search.
- How to use lead forms in responsive search ads and campaigns.
- About qualifying responses in lead forms.
- Prevent invalid leads.
- About call reporting.
- About AI-qualified call leads.
- Test with confidence with the Experiments page.
Recheck these sources before changing account settings, particularly hosted-form qualification, automated destinations and AI-qualified call reporting. Use the worksheet to establish business definitions first; implementation of conversion imports and financial budget modelling are separate tasks.
