Operations
Part of Marketing funnels: ten steps to define customer states, handoffs, economics and data
Putting a real process behind marketing funnels
Marketing funnels work when journey states are observable, counts reproduce, and owners can act on lead signals before budget spend changes.
What to take away
- Define each journey state by its entry event, maximum time, and owner before you build anything.
- Reconcile funnel counts against orders, tickets or CRM records; if they do not match, the baseline is not ready.
- Write the event dictionary before tagging. GA4 lets the same action arrive through more than one collection path, and duplicated events inflate every rate downstream.
- Match the measurement method to the questiona single transaction, an end-to-end journey, or a strategic issue.
- Hold the launch when counts will not reproduce, no one owns the handoff, or delivery cannot absorb the result.
A funnel is a set of states a customer can be observed entering and leaving. Learning how to build marketing funnels starts there, not with a diagram of stages someone drew in a workshop.
States a stranger could verify
Write each state so two analysts reading the definition reach the same count.
A state needs several things. Here is what it needs:
State definition checklist
- Eligibility
- Entry event
- Preceding sequence
- Maximum time inside
- Identity rule
- Exclusions
- Owner
- Next states
"Consideration" fails that test. "Returned to the pricing page within 14 days of a trial start, no support ticket open" passes it.
Keep person, account, session, opportunity, order and cohort definitions separate. Most funnel arguments are two people counting different units and neither saying which.
The event dictionary
Event dictionary fields
- Event name
- Trigger
- Parameters
- Sending system
- Identity key
- Consent handling
- Metric it feeds
Google's GA4 event reference lists required and optional parameters for recommended events and warns that some events can duplicate when they are sent through more than one collection method. Check which collection path the business actually runs before trusting a rate.
Then run known journeys by hand: a purchase, a cancellation, a return, a phone order, a handoff to a rep. Reconcile the totals against the operational system that owns the money.
Match the method to the question
The UK Government's guide to measuring service success recommends matching the method to a transaction, an end-to-end journey, or a strategic issue, and pairing performance measures with user research where that fits.
A transaction measure answers whether the checkout worked. A journey measure answers whether people who needed the thing got it. A strategic measure answers whether the funnel is worth running at all. Pick one before you pick a dashboard.
Build sequence
| Stage | Output | Release gate |
|---|---|---|
| Scope | Customer, outcome, owner, boundary | One decision is explicit |
| Model | States and handoffs | Normal and failure paths both work |
| Measure | Events, identity, metrics, reconciliation | Counts reproduce |
| Operate | Baseline, test plan, review cadence | Owners can act on findings |
Before you launch
Delay the launch when material counts cannot be reproduced, when no one owns the handoff between two teams, when customer harm is plausible, or when delivery cannot absorb the volume the funnel produces. Those four conditions are the ones that turn a measurement project into a repair project.
The GOV.UK technology selection guidance recommends adaptable choices, data control, security review, and ownership-cost analysis. Those questions apply to a funnel build as much as to a public service.
The CISA software acquisition fact sheet covers development practice, supply-chain exposure, deployment and vulnerability management. Add those questions to the vendor review; they are not local approval.
Once counts hold, the cost of each state belongs in marketing budgets, and the ranges other teams report sit in marketing planning benchmarks. The same discipline applies one level up when building marketing frameworks.
Common questions
What should be built first?
The narrowest funnel that supports one important decision, with states a second analyst can count the same way. If it cannot answer that decision, it is a reporting exercise.
How much baseline data is enough?
Enough complete cycles to see normal variation and cohort maturity. Document the sample size, the mix, the known limitations and every definition change, so a later reader knows what moved.
When should a funnel launch be delayed?
When counts will not reproduce, when a handoff has no owner, when customer harm is plausible, or when the team receiving the output cannot act on it.
