Choose a stage that the business trusts
A B2B team may have enough weekly leads for a model but only a few closed deals. Modelling lead volume can therefore be more stable than modelling revenue, yet it can reward cheap enquiries that sales will never accept. Define qualification consistently and retain the stage history, rather than using the latest CRM status for every historical week.
- Define a qualified stage
- Measure cohort maturity
- Separate carryover and delay
- Validate against revenue
The useful question is narrower than which channel wins: which spend change is likely to create additional qualified opportunities, and how uncertain is the bridge from those opportunities to revenue? Start with the measurement framework to separate delivery metrics, modelling outcomes and commercial validation.
A useful model starts with a consistent outcome and a decision it can inform. The marketing mix modelling learning path connects this idea to the assumptions, practical checks and next steps.
To work through the connected decisions, read SaaS MMM: connect trial acquisition to retained value and MMM data leakage: audit the path from raw data to holdout.
Distinguish delayed conversion from adstock
Advertising can influence someone after exposure; an existing prospect can also take months to sign a contract. Those are different mechanisms. An adstock transformation models persistence in media effect. A sales-cycle delay describes when the measured outcome is recorded. Letting one flexible lag absorb both can make the model hard to interpret.
For a fictional campaign producing 100 qualified leads, imagine 20 opportunities appear in the first month and another 10 in the next two months. If the latest campaign has only one month of follow-up, comparing its 20 opportunities with a mature campaign's 30 unfairly penalises the newer cohort. Report a common maturity window before comparing channels.
Build a defensible weekly dataset
Keep acquisition date, qualification date and close date separately. Reconcile stage definitions across sales teams and record any CRM migration. Track major pricing, territory, staffing or qualification-policy changes that can move outcomes independently of media.
A large deal can dominate weekly revenue. Do not delete it just because it produces a large residual. Investigate whether it belongs to the same decision process and run a documented sensitivity analysis. If a single contract reverses the recommended allocation, the recommendation should reflect that fragility.
Use the model to choose a test
Hold out future periods and wait for the chosen outcome to mature. Compare predictions by cohort as well as by calendar week. Check whether apparent media effects survive reasonable changes in lag assumptions and whether correlated channels can actually be distinguished.
Suppose two plausible models disagree about whether webinars or paid search created pipeline. A joint campaign calendar cannot settle that question by itself. Consider a limited holdout or staggered change that creates new variation, and define the opportunity-quality outcome before launch. The adstock guide gives the carryover mechanics; this article's added task is to keep those mechanics separate from the CRM clock.
The deliverable should include a stage dictionary, maturity curve, sensitivity table and a bounded next decision. If the outcome is too sparse for useful separation, report that limit and use other evidence rather than increasing model complexity.
Sources and further practice
This article’s worked example is illustrative, not a measured client result. Use the linked methodology to check the assumptions before applying it.