The Same Click, Different Funnels

Two companies buy the same ad, on the same platform, from the same auction. The click event that fires is byte-for-byte identical.

At the first company, the person who clicked reaches a checkout and buys twelve minutes later. At the second, they download a whitepaper, are one of six people at their employer who will touch the vendor over the next eight months, and the purchase is signed by somebody who never saw the ad.

The platform reports a conversion for the first and a lead for the second, and both companies put a cost-per-something in the same slide template. Only one of those numbers is about a customer.

The click is not the thing that varies

This is worth stating plainly because most attribution arguments start one step too late, at which rule should divide credit, and skip the question of what the credit is even being divided between.

What varies between funnels is three things, and all three are downstream of the click rather than in it.

The same click, arriving in four different funnels

Funnel	Unit	Outcome	Lag
Self-serve checkout	The person	A purchase	Minutes to days
Subscription trial	The person	A retained subscriber	Weeks to months
Product-led signup	The account, eventually	Usage that predicts expansion	Weeks
Sales-qualified pipeline	The account	A closed deal	Quarters

The unit is the one that breaks things quietly

In a self-serve funnel the clicker and the buyer are the same person, so a person-level attribution system has a coherent job. In a sales-qualified funnel they are usually not, and often the buyer never clicked anything.

That is not a tracking failure to be solved with better identity resolution. Six people at one company reading three pages each is one company considering a purchase. Counting it as eighteen touches and one conversion, then dividing credit across the eighteen, is arithmetic performed on the wrong object.

The window follows the lag, and most windows are inherited

Platform attribution windows are set by the platform, default to something short, and are the same for everyone buying that inventory. A seven-day click window is a reasonable default for a checkout and close to meaningless for a funnel whose outcome lands in two quarters.

The consequence is systematic rather than random. Long-cycle funnels see their conversions fall outside the window, so the channels that touch people early look worse than they are and the channels that touch them last look better. Anything that changes the length of the cycle, a pricing change, a new procurement step, moves reported channel performance without any channel changing.

What to do about it

Three moves, in increasing order of effort and value.

Say the grain on every number. A metric named without its unit is the source of most of these arguments. Cost per lead, cost per account, and cost per customer are three different numbers and they belong on the slide with those words attached.

Fix the reporting window to the funnel rather than to the platform. Pull the conversion from your own data at a horizon that matches the cycle, and treat the platform figure as a bidding signal rather than as a result.

Aggregate to the unit the outcome lives at before you attribute anything. For account-based funnels that means resolving touches to accounts first, which is the job the account graph exists to do, and accepting that the attribution question is then about accounts and always was.

Why this is a different problem from multi-touch credit Technical

Multi-touch attribution takes a set of touches and a conversion and divides the conversion across them. Everything in this lesson happens before that: which touches belong together, what counts as the conversion, and which entity is being converted. A credit rule applied to a badly specified unit produces a precise division of the wrong thing, and no choice of rule repairs it.

That ordering matters practically because the two problems have different fixes. Credit rules are a modeling choice and can be changed in an afternoon. The unit is a property of the data model, and changing it means resolving identities, choosing an aggregation rule, and usually rebuilding a year of history. Teams reliably do the afternoon of work first and then argue about the results for a quarter.

Context

Best suited for

Any organization buying media into more than one kind of funnel, and any team importing a reporting setup from a business whose funnel was shaped differently.

Data requirements

• Touches joined to whichever entity the outcome lives at, which for account funnels means an account resolution step first
• Outcome timestamps, not just outcome flags, so a window can be chosen rather than inherited
• The conversion rate between each funnel step and the eventual outcome, measured rather than assumed

What changes by context

B2C ecommerce

The easiest case and the one that generates the misleading intuitions everyone else inherits. Clicker and buyer are the same person and the lag is short, so platform attribution looks accurate and gets exported to funnels where it is not.

B2B SaaS

The buying group means the person who clicked is frequently not the person who signs, so person-level credit is being computed on the wrong entity before any rule is applied.

Product-led

The signup is an individual and the eventual account is a team, so the unit changes partway through the funnel. Deciding when a signup becomes an account is a definition question that changes every downstream number.

Sales-led

Reps create the records that mark funnel progress, so the measurable steps move when rep behavior moves. That is part of the effect and it means the step is not a clean proxy for demand.

When it breaks down

• Single-funnel businesses where every click enters the same path, in which case the grain is fixed and the argument is about credit rules.
• Programs where the outcome cannot be joined to the touch at any grain, where the route is aggregate measurement rather than a better join.

What this does not establish

• That platform-reported conversions are wrong. They are computed correctly under a stated window and rule, on the traffic that platform can see, and they answer that platform's question rather than yours.
• That account-level attribution is causal. Resolving touches to the right unit fixes the object being divided; it does not add a counterfactual, and the division is still a division.
• That a longer window is always better. It includes more genuine conversions and more that would have happened anyway, and which effect dominates depends on the channel.
• Anything about the size of the gap between cost per lead and cost per customer in a given business. That ratio is specific to a funnel and has to be measured rather than assumed from another company.