Every attribution guide you will find was written for a company with ten thousand conversions a month. Multi-touch models, data-driven weights, incrementality tests with control groups. Good tools, for a size you are not.
At forty conversions a month, a data-driven model is a random number generator with a nice interface. The honest version of attribution at your size is smaller, plainer, and more useful than anything you can buy.
What breaks at small numbers
Percentages break first. Then models. Then confidence.
A channel that produced two of your forty customers “converts at 5%”. Next month it produces zero and “converts at 0%”. Nothing changed except the coin. Any model that weights channels by these rates is weighting noise.
The second thing that breaks is the model itself. Multi-touch attribution splits credit across touches by rules that were tuned on volumes you do not have. It will produce a confident-looking allocation. It will be wrong, and it will not tell you it is wrong.
The third thing that breaks is you. A tidy chart of wrong numbers is more dangerous than no chart, because you will act on it.
A tidy chart of wrong numbers is more dangerous than no chart.
What works instead: people, not percentages
At forty customers, you can know each one. That is the advantage, not the problem.
Trade rates for lists. For each customer this quarter, write down four things.
- The first thing they touched that you can find, with the date.
- Every touch between that and the booking, in order.
- How long the whole thing took.
- What they had in common with the others who took a similar path.
Forty rows of that is a real dataset. Patterns show up that no percentage could show: a community thread that appears in the first touch of nine customers, a webinar that appears in the middle of twelve, an ad channel that appears nowhere at all.
This is correlational. Nine customers who touched a thread first did not necessarily convert because of the thread. But nine is a pattern worth funding a test, and a test is how a small company earns the right to a bigger claim.
Say the sample size every time
When the sample is two, say two.
The one rule that makes small-sample attribution honest: every number carries its count. “Video first-touch: 9 of 40.” Not “23%”. The count tells the reader how much to trust the claim, and it stops a coin flip from becoming a strategy.
Cells with fewer than three people do not get a number at all. Hatch them. It feels like a downgrade. It is what makes the other cells worth anything.
Attribution for a small company is not a smaller version of enterprise attribution. It is a different, more honest thing: forty people you can actually know.
Questions people ask next
Does multi-touch attribution work for startups?
Rarely. Multi-touch models need volumes most startups do not have; at tens of conversions a month they allocate noise. Walk individual customer journeys backwards instead.
How many conversions do I need before percentages mean anything?
There is no magic number, but a cell with fewer than three people has no rate yet, and a channel rate should always be shown with its count next to it.
What is the honest alternative to an attribution model?
A list: for each customer, the first touch, every touch in order, the time it took, and what they had in common. Forty rows shows patterns; a test confirms them.
Throughline reads across every tool you already pay for and narrates the part that moved. Request a run to see your own line.