Keeping marketing pipeline attribution and incrementality apart

Pipeline attribution assigns credit for observed outcomes such as leads, opportunities or revenue to a marketing interaction. Incrementality asks which additional part would not have arisen without that activity. Both perspectives are useful, but they answer different questions. A CRM records a Google Ads interaction or a source assignment; that does not establish that the campaign causally influenced that individual opportunity. Only a suitable comparison estimates aggregate incremental outcomes, not which particular sale was caused.

What is marketing sourced pipeline?

Sourced pipeline refers to sales opportunities whose defined original source is marketing. The definition needs an immutable source snapshot, a point in time and a rule for direct, organic, paid and unknown origin. Without versioning, later CRM clean-up can overwrite historical sources.

A simple model can use first source. It is understandable, but understates later contributions. Another model can use the source at opportunity creation. What matters is not that a model is perfect, but that it is stable, transparent and appropriate for the decision.

What is marketing influenced pipeline?

Influenced pipeline covers opportunities with a defined marketing interaction before or during the sales process. The figure quickly becomes large, because several campaigns can touch the same opportunity. Never add influenced pipeline across channels when the same opportunity appears more than once.

Define a permitted contact window and qualifying interactions. A page view or an impression can be visible without having influenced the sale. Show unique opportunities, touchpoint count and model separately.

How is the pipeline connected technically?

Use stable IDs from lead through contact, account and opportunity. Store the original source at entry and later touchpoints as separate events. Status changes record time, actor and reason. Revenue and currency come from the authoritative business system.

Joining records by email address, company name or daily aggregates is error-prone. Merges and duplicates need a documented process. Keep raw events, derived attribution and manual corrections separate.

Which attribution models make sense?

Models can use first touch, last touch, opportunity source or multiple touchpoints. Google Ads assigns conversion credit among ad interactions according to the selected supported model. Those CRM examples are not necessarily supported Google Ads models. A CRM pipeline model must not adopt that credit unchecked as sourced pipeline, because units, windows and events can differ.

Choose the model according to the question. First source can be appropriate for the acquisition source. An influence model can help to understand contributions from supporting campaign interactions. Neither is enough on its own for budget impact. Publish the model, window, unit and exclusions alongside every figure.

What does incrementality mean?

Incrementality is the additional change relative to a credible counterfactual. The counterfactual can be approximated through a randomised control group, a geographic holdout, a time-series model or other designs. Every design has assumptions.

Google’s Conversion Lift uses treatment and control to estimate additional conversions. The documentation separates iROAS (incremental return on ad spend) from standard ROAS: the iROAS numerator values the incremental conversions, whereas standard ROAS uses attributed conversion value. Not every account or campaign meets the requirements. According to the Google documentation reviewed, modelled delayed conversions are available only in Demand Gen-only user-based studies. Interpret such delayed estimates, where provided, separately from observed results and with uncertainty.

Which experiments answer which question?

• Campaign experiment: effect of a campaign change under platform conditions.
• Conversion Lift: additional conversions compared with a control group.
• Geo experiment: market regions are compared; spillover and regional differences are risks.
• Landing page A/B test: effect of a page experience after the click.
• Process pilot: evaluates a new sales or marketing process; without control it alone does not establish causal impact, which requires a suitable comparison.

These designs must not be mixed. A landing page test does not prove the incrementality of the entire channel. A campaign experiment does not automatically measure contribution margin.

How is an experiment planned?

Formulate the hypothesis, unit, target metric, expected direction, minimum duration, exclusion rules and stopping criteria before the start. Document the assignment and possible contamination. Check whether parallel campaigns, seasonality, sales activity or a price change affect the groups differently.

Do not use a winning metric selected after the fact. Show the primary metric, guardrails and uncertainty. An inconclusive result is a valid finding, not an obligation to tell a positive story.

How are pipeline and lift read together?

Create three levels:

1. observed CRM pipeline under a stable source rule;
2. attributed platform value under a documented model;
3. incremental estimate from a suitable design.

The levels can point in the same direction or not. If pipeline rises but lift is unclear, there may be more observed activity, different attribution or a genuine effect. If lift is positive but CRM pipeline is missing, this may indicate a connection gap, but it can also reflect a different outcome definition, lead quality, delayed maturation or unmatched data. It is not proof of a broken integration; investigate these possibilities. Attributed platform value is not automatically revenue or profit; that depends on how the conversion value is defined. No level is quietly renamed into another.

Which cohort and maturity apply?

Long sales cycles need matured cohorts. Define the entry point: click, lead, MQL (marketing-qualified lead) or opportunity. Define the time to expected maturity and the end of observation. Show open opportunities separately; do not count their amount as won revenue.

Historical cohorts can accumulate outcomes later. Store the report run and the data cut-off. Compare cohorts under the same definition. A change of CRM process, conversion window or attribution model marks a break.

Which decision may the analysis support?

Pipeline attribution can set priorities for data quality, the sales handover and hypotheses. It can support a budget decision when its limits are visible. An incrementality study can answer a narrower causal question. No figure automatically authorises a live change.

A minimal data model

The model needs lead ID, account ID (the CRM customer/account ID), opportunity ID, event time, event type, campaign or source ID, value, currency and status. A separate immutable copy of the original source holds the first known origin. Touchpoints are captured append-only; derived attribution sits in its own version. Manual corrections retain the actor and the reason.

For experiments, add unit, group assignment, start, end, exclusion and analysis version. The assignment is not overwritten according to the result. A report only totals amounts in the same currency, or converts them explicitly with documented exchange rate and rate date; it does not add different currencies merely because both are valid. Open opportunities stay separate from won revenue. If an ID is missing, the row stays unlinked instead of moving to an opportunity through an unreliable name similarity.

This minimal model is deliberately technology-independent. It can be implemented in a CRM, a warehouse or a controlled analysis layer. The architecture decision remains with the company; the article promises no maitiq CRM implementation.

For an initial implementation, a small, verifiable event list is enough. Start with a valid lead, opportunity creation, win and loss. Add further touchpoints only if they answer a concrete question. An extensive event catalogue without stable identity exacerbates the uncertainty. Maintain one data owner, one quality rule and one permitted purpose per field.

Before every management report, a reconciliation step is carried out: unique leads, opportunities and customers in the CRM are checked against the analysis layer. Differences receive categories of discrepancy such as delayed, duplicate, unlinked or outside cohort. Only then are sourced, influenced and experimental results calculated.

Frequently asked questions

How maitiq turns measurement into better Google Ads decisions

maitiq evaluates the account data and evidence actually available within the agreed scope, and keeps data update time, coverage and corrections visible alongside the decision.

For maitiq, tracking is not a separate technical project. A signal is only useful once its definition, source, data update time and coverage are known and it can support a defensible budget or performance decision. The starting point is the agreed scope: CRM, website and revenue analysis need separately supplied or agreed access and suitable fields. That is why the human decision-maker sees measurement gaps, skipped checks and exclusions alongside the proposal.

What the read-only audit reveals about the measurement chain

The first step shows what the available account data and evidence can establish, which campaigns only appear to perform well and which measurement gap must be closed before a budget shift. You do not get an artificially lengthened list of findings; you get the strongest supported priorities, remaining evidence gaps and the next decision to make. A Google Ads audit stays read-only: it produces proposals and a report, names limits and skipped checks, but makes no change in the account.

Decision rule: Use attribution for operational distribution and separate experiments or robust comparisons for genuine incrementality questions.

Sources and how to read them

The two linked Google references on Conversion Lift and attribution models explain platform functions and measurement limits. Information about how maitiq works is available at maitiq.com.

Google Ads access alone does not evidence the entire measurement chain and is not a general CRM integration. Website tracking, CRM quality or revenue attribution need separately supplied or agreed access to the relevant systems and suitable fields.

Have maitiq review your specific case.