Start with one number
Before any model, agree on a single definition of a qualified lead and count it in one place. Most attribution arguments are really definition arguments. Once the definition is settled, the arithmetic stops being contentious.
This sounds too simple to be the main point, and it is the main point. In most disputes about attribution the two sides are counting different things and have not noticed. One is counting form submissions, the other is counting opportunities that reached a sales stage, and the ratio between them is where the entire disagreement lives.
What counts as qualified
Write it down as a test somebody could apply without exercising judgement. A real contact method, a stated requirement matching something you actually sell, and a budget or timeline signal is a workable bar for most business-to-business teams. For e-commerce the equivalent is a completed transaction, net of refunds and cancellations, counted once.
The definition matters less than the fact that it is fixed and written down. A mediocre definition applied consistently produces a usable trend. An excellent definition that shifts every quarter produces nothing at all.
What you can safely ignore
You do not need multi-touch modelling to run a good marketing programme. You need clean events, deduplicated conversions, and consistent channel grouping. Everything past that is refinement, and refinement on dirty data is theatre.
Multi-touch models are not wrong. They are expensive to maintain, unusually sensitive to exactly the data quality problems most teams already have, and they answer a question most organisations are not yet ready to act on. If you cannot currently agree how many leads you got last month, a model distributing fractional credit across seven touchpoints will not settle it.
The setup, concretely
Three components, none of which requires an engineer: a single source of truth for the conversion, a consistent way of labelling traffic, and a weekly place where the number gets read aloud.
One conversion, defined once
Pick the system that owns the truth, which is almost always the customer relationship manager rather than the ad platform, and treat everything else as an estimate. Deduplicate on a single identifier. Count the conversion when it meets the written definition, not when the form fires.
Labelling traffic consistently
Agree a campaign tagging convention, write it in a shared document, and enforce it with a spreadsheet that generates the links. Most channel-grouping chaos is not a tooling problem. It is four people spelling “newsletter” three ways over two years.
Reading it weekly
A number nobody reads aloud drifts. A fifteen-minute weekly review, where one person states the qualified lead count, the spend and the cost per lead and explains any change, does more for attribution quality than any platform you could buy.
Making it defensible
Write the model down: what counts, what is excluded, and where the known gaps are. A finance team will accept an imperfect model that is documented far sooner than a sophisticated one nobody can explain.
Name the gaps explicitly, because they are coming up anyway and it is better that you raise them. Direct traffic that was really a forwarded email. Offline conversions that never make it back into the system. Cross-device journeys you cannot stitch. Writing those down converts them from an ambush into a known limitation, and finance teams deal with known limitations every day.
The conversation with finance
Finance teams are not hostile to marketing numbers. They are hostile to numbers that change without explanation, and marketing supplies those more often than it admits. A figure that moved because the definition moved, presented as though performance moved, is what destroys credibility, and it is usually unintentional.
Three habits fix most of it. State the definition every time you state the number, so a change in one is never mistaken for a change in the other. Report the same metric every month even when it is unflattering, because a metric that appears only in good quarters is read correctly as marketing. And when the definition genuinely has to change, restate the prior period on the new basis before anybody asks you to.
When to upgrade
There are two honest triggers. The first is when the written model’s known gaps grow large enough to change a decision you are about to make. The second is when you have a person whose job actually includes maintaining the thing you are about to build.
Absent either, a documented spreadsheet beats a sophisticated model, because it will still be accurate in six months. A sophisticated model running quietly on duplicated conversions is worse than no model at all, because it is believed.