Measurement
Which marketing channel actually made the sale?
Last-click credits whoever stood nearest the till. What a position-based split does instead, and where in the funnel the money leaks.
How MarketWar handles it
Credit is split by a U-shaped position-based model computed with no wall-clock and no randomness, so the same touchpoints always produce the same split — a report you cannot reproduce is a report you cannot argue with. The funnel runs eight stages from impression to referral and NAMES the biggest drop-off rather than averaging it away, and every figure is labelled an estimate derived only from counts you supplied.
Ask most analytics tools which channel made a sale and they will tell you the last thing the buyer clicked. It is the cheapest answer to compute and it is wrong in a specific, expensive direction.
Last-click credits whoever was standing nearest the till. For almost every small business that is brand search — somebody typing your name into Google because of something else they saw. So brand search looks like your best channel, you move budget into it, and you defund the thing that made people search for you.
Two different questions, and they need different maths
Where is the money leaking? That is the funnel: eight stages from impression through attention, engagement, click, lead, purchase, repeat purchase and referral. The useful output is not the overall conversion rate — it is which single step loses the most, named rather than averaged into a percentage that hides it.
A business with great reach and a dead landing page has the same blended conversion rate as one with weak reach and a great page. They need opposite fixes.
Who deserves the credit? That is attribution, and it uses a U-shaped position-based split: the first touch that created the discovery and the last touch that closed it each take a substantial share, with the middle touches splitting an assisting share.
That is a judgement, and it is worth saying why it is defensible. The two hardest things to replace in a purchase are the moment somebody first heard of you and the moment they decided. The middle is real but more substitutable. Crediting all three equally overstates the middle; crediting only the last pretends the first never happened.
Reproducible, which is rarer than it should be
There is no wall-clock and no randomness anywhere in this maths. The same touchpoints always produce the same split.
That sounds like a technical footnote. It is the difference between a report you can act on and one you cannot: if a figure moves, you know your data moved, not the model. A number that quietly drifts between two readings of the same week cannot be used to make a decision, and it cannot be defended to anybody who asks where it came from.
What it refuses to do
It works from the counts you supply, and it labels every derived figure an estimate rather than a measurement.
A sale with no recorded touchpoint is reported as unattributed. It is not sprinkled across channels in proportion so the pie adds to 100%. That tidy pie is the most common lie in marketing reporting: it takes the sales you cannot explain and distributes them to the channels you can, which flatters whatever you already spend most on.
If a quarter of your revenue is unattributed, the honest report says so — and then the useful action is to fix the tracking, not to admire a chart.
What to do with the answer
Pair it with the money. The funnel tells you which step to fix; the campaign guardrail tells you what you can afford to spend acquiring the next customer, computed from your own contribution rather than a benchmark.
The question to ask any analytics vendor
Run it twice on the same data. Do I get the same number?
If the answer involves sampling, modelled conversions or a figure that moves when nothing moved, you are being sold a story with error bars nobody will show you.
What this does not do
It attributes revenue you actually recorded. It cannot see the customer who saw your advert on Tuesday and walked into the shop on Saturday with no record in between, and it will not invent a touchpoint to make the model look complete.
Common questions
What is wrong with last-click attribution?
It credits whoever was nearest the purchase. Brand search usually wins, which makes a business cut the channels that created the demand brand search captured.
What is a U-shaped model?
Position-based: the first touch that created the discovery and the last touch that closed it both get a substantial share, and the middle touches split an assisting share. It credits the two moments that are hardest to replace.
Will two runs give me the same answer?
Yes. There is no wall-clock and no randomness in the maths, so identical inputs give an identical split. If a figure changes, your data changed.
Does it guess at offline sales?
No. It works from counts you supplied and labels every derived figure an estimate. A sale with no recorded touchpoint is reported as unattributed rather than assigned to a channel for neatness.
Related
Should AI send your customer replies for you?
The case against auto-replying to customers, what a drafted reply does instead, and how nine channels become one queue without anything leaving unread by a human.
How do you know when to stop a campaign that is losing money?
The four numbers that decide it, why ROAS on its own is misleading, and what a holdout group tells you that attribution never will.
What does an AI marketing action actually cost?
Why AI marketing tools hide the price until the invoice, what a single action costs here, and what happens when your balance runs out mid-job.