Google Ads leads your sales team can work with

Which event counts as a valid lead?

Start with the event that reliably proves a valid enquiry was stored. A button click, form start or thank-you page view is a diagnostic event, not yet a business outcome. Define honeypot, spam, duplicate and validation rules. Decide which system stores the enquiry authoritatively and which systems only receive copies.

In Google Ads, primary conversion actions affect reporting and, where applicable, bidding. Secondary actions stay visible for diagnosis. Avoid counting the same enquiry more than once as primary through a direct tag, GA4 and a later import. For each conversion action, define what it counts, how it is counted and who is responsible. Check the conversions selected in the campaign goal: a custom goal can also use secondary actions for bidding, so the secondary label alone does not rule out its use for bidding.

How should you track forms and phone calls?

Form and phone leads are two capture paths, but they are not automatically equivalent. For forms you need a server-side record of completion, the source, the campaign link and the later qualification status. For calls, distinguish the ad or call-button impression or click, the actual call, the connected conversation with its duration and only then the confirmed quality. A click is not a connected or qualified call.

Google call reporting can use Google forwarding numbers and duration-based conversions. A duration threshold is only an operational filter. A long call can be irrelevant; a short call can concern an existing customer. Feedback from telephony or the CRM remains necessary for quality and revenue. Availability and capture depend on the country and the setup.

Which funnel stages should be reported back?

A robust B2B funnel separates at least:

1. valid enquiry;
2. marketing-qualified lead against documented criteria;
3. lead accepted by sales;
4. sales opportunity with a defined stage;
5. won customer;
6. revenue or, preferably, contribution margin where available.

Not every company needs every stage in Google Ads. Choose the action that is frequent, stable and timely enough and that also represents business value. Later stages can stay in reporting without becoming a bidding signal straight away. Check the conversions selected in the campaign goal: a custom goal can use secondary actions for bidding.

Enhanced conversions for leads can connect hashed, first-party data and Google identifiers with later offline outcomes. Enhanced conversions still require a clear data-processing purpose and a properly configured, monitored import. Legal basis, consent, data minimisation, hashing before transfer, retention, duplicates, upload latency and error handling must all be clarified. Current Google documentation points to Data Manager/API paths; an older guide must not be adopted blindly.

How does lead tracking become a view of the sales pipeline?

Store the original source as an immutable snapshot on the lead. Record later status changes with time, actor and reason. Link sales opportunity and revenue through stable IDs, not names or daily aggregates. A weekly report shows new enquiries and qualification errors. Assess revenue and close rate only once enough time has passed since the enquiry for a typical sales cycle.

Pipeline attribution (assigning sales opportunities to a defined source) shows which campaigns were involved in the opportunities. It does not prove that those opportunities would not have existed without advertising. Three measures are common and they overlap: First Source names the source of the first touch; Sourced Pipeline counts opportunities that came from a contact the campaign originally won; Influenced Pipeline covers every opportunity with at least one touchpoint from the campaign. Keep them apart instead of adding them into one number.

Which metric answers which question?

• CPL (cost per lead) shows media cost per valid enquiry.
• Cost per MQL/SQL shows cost per defined quality level.
• Cost per sales opportunity connects marketing with the sales process.
• CAC (customer acquisition cost) includes all relevant acquisition costs per won customer.
• ROAS divides attributed conversion value by advertising cost.
• ROMI/marketing ROI uses a clearly defined outcome measure and the full marketing cost.
• Incremental ROAS asks for additional value against a control condition.

The terms are not interchangeable. For every metric publish the formula, data source, currency, period, time to qualification or close and exclusions. If the required numerator, denominator or evidence is missing, that particular metric stays unavailable rather than zero; other metrics whose data exists remain calculable. Valid zero and negative values stay visible; a zero denominator is shown as N/A. For ROMI/marketing ROI name the chosen outcome measure and the full cost scope; the definition is not standardised across the industry, see the ROI and ROMI guide.

Data quality before model complexity

Check completeness, uniqueness, freshness, consistency and that totals reconcile. Do costs agree at account, campaign and overall level? Are currencies and time zones consistent? Are cancelled purchases or spam leads handled? Has a tag or consent change produced a visible break? Are campaign names stable identities or merely descriptive text?

A data-quality log records the data source, extraction time, row count, totals, missing values, duplicates, filters and known deviations. It is frozen before modelling. If the data basis changes, the analysis gets a new version. That way a reviewer can repeat the calculation and explain differences. Keep the original source and a documented correction history where that is lawful; apply the applicable retention and deletion rules to every stored version.

AI can help find anomalies, but should not perform silent repairs. A detected gap is flagged. An imputation needs a method defined in advance and remains visible as imputed. The team decides whether the data suffices for the question or whether the analysis stops.

Which data checks are needed before optimisation?

Check before any optimisation:

• Is every valid enquiry captured exactly once?
• Are time zone, currency and conversion window correct?
• Are primary and secondary actions set correctly?
• Are calls and forms reported separately?
• Are click or user identifiers complete and permitted?
• Do CRM updates arrive on time and without duplicates?
• Are missing values kept separate from observed zero?
• Has enough time passed since the enquiry for a revenue or pipeline statement?

If one of these conditions is not met, no performance judgement that depends on it is derived. Missing CRM or revenue data does not block observations supported by the available data, such as impressions, clicks or cost; it only blocks conclusions that rest on the missing evidence. The gap is named openly and assigned to a responsible person. Do not derive a bid increase from unverified lead quality.

Attribution as navigation aid, not causal proof

Attribution models are useful when their perspective is known. A last-click model shows a different distribution of attributed contribution than data-driven attribution. Changes to model, lookback window or channel definition can change reports without any change in customer behaviour. Methodology and change date are therefore shown next to the result.

Do not compare attribution across platforms uncritically. Each platform sees different touchpoints, identities and modelled gaps. Totals can overlap. A cross-channel dashboard needs its own reconciliation and must not simply add platform credit as unique conversions.

Use attribution to prioritise questions and tests: which paths look relevant? Where do results change strongly with the model? Where are touchpoints missing? The answer to “what had additional impact?” requires stronger evidence, such as a suitable experiment or a carefully validated quasi-experimental design.

Using experiments and marketing mix models for the right question

An experiment defines the hypothesis, population, test and control group, primary metric, runtime and evaluation rule before it starts. Control metrics make trade-offs visible, for example more leads of lower quality or a lower cost per sale with lost qualified volume; predeclared review and stopping rules determine whether and how to respond. Interim results are not used to stop at will. With small volumes the uncertainty can be so large that no reliable decision is possible. Inconclusive is a valid result.

Marketing mix modelling typically works with aggregated time series to model contributions from media and other factors. Google provides an open MMM method with Meridian and describes experiment calibration and budget analysis. That demonstrates an available methodology, not its suitability for every data set. Data length, variation, confounders, specification, validation and expertise must be checked.

Experiments, MMM and attribution answer different questions. They can inform each other but do not replace each other arbitrarily. A measurement plan assigns every business question to suitable evidence and shows which uncertainty remains.