Attribution for a Business With One Salesperson

Where the models came from

Attribution modelling was developed for businesses running several paid channels simultaneously, at volumes where a percentage point of efficiency is worth real money.

In that setting, working out whether the display advert or the search advert deserves the credit changes where a substantial budget goes.

The techniques then travelled downward into businesses with one salesperson, three channels and forty jobs a year, where they answer a question nobody was going to act on.

The models are not wrong. They are built for a scale most businesses are nowhere near.

The volume problem

The mathematics that makes attribution meaningful requires enough events for patterns to separate from noise.

At forty jobs a year, a difference of three jobs between two channels is well inside the range of ordinary variation.

You would need several years of stable trading to distinguish a genuinely better channel from a lucky one, and by then everything else has changed.

So a model that assigns forty percent credit to one touch and sixty to another is producing a number with far more precision than the underlying data can support.

What you can actually establish

• Which broad sources produce work at all.
• Roughly what proportion each accounts for.
• Which produce the larger jobs.
• Which produce work you want rather than work you tolerate.
• Which produce nothing, consistently, over a long period.

Those five are answerable at small volumes and each can change a decision. Anything finer is describing noise.

Proportionate is not the same as lazy

Worth separating, because the argument for simplicity is often used to justify measuring nothing.

Proportionate measurement is deliberate: you decide what you need to know, collect exactly that, consistently, over a period long enough to mean something.

Measuring nothing is what most small businesses actually do, and it leaves them guessing.

The difference is a sheet with six columns kept for a year, against either an elaborate model nobody maintains or a shrug.

Last touch, understood honestly

Most small businesses effectively use last-touch attribution, since they record how somebody arrived at the moment they arrived.

It is the simplest model and it has a known bias: it over-credits whatever is closest to the sale, usually search and direct visits, and under-credits everything that created awareness earlier.

Knowing that is enough. You do not need to correct for it mathematically; you need to remember it when reading the numbers.

In practice that means not cancelling the things that build recognition on the grounds that they show few direct conversions, which is the specific mistake last-touch reporting invites.

A worked example

A specialist contractor doing around fifty projects a year was persuaded to install multi-touch tracking across their site and advertising.

After eight months the reporting was extensive and nobody could explain what to do differently.

The numbers moved substantially month to month, on volumes of three or four projects, so every trend reversed the following month.

They abandoned it and kept a sheet instead: date, source in the caller's words, quoted, won, value.

After a year the picture was plain. Referrals and repeat clients were roughly half the work and nearly two thirds of the value. Search produced most of the rest. Two paid channels had produced four enquiries and one small job between them.

The decision followed immediately: stop the two paid channels, spend the money on staying in touch with past clients.

The elaborate system had never produced a conclusion that clear, because it was distributing fractional credit across a sample too small to distribute anything.

Use a longer window

The single most useful adjustment at low volume.

Monthly reporting on four jobs a month is noise with a chart attached, and reading it monthly guarantees reacting to randomness.

Quarterly is better. Annual, compared against the previous year, is better still and is what most small businesses should actually be looking at.

The instinct to review frequently comes from wanting to be responsive, and at these volumes it produces the opposite: decisions made on variation, reversed two months later.

The one number that matters most

If you keep only one figure, keep the proportion of work coming from people who already knew you.

Repeat customers and referrals, counted together, as a share of the total.

For most established small businesses it is between a third and two thirds, and it is almost always higher than the owner assumes.

It matters because it reframes where effort should go. A business getting half its work from people who already know it should be investing in that relationship, and most of them are instead spending on reaching strangers.

Deciding to stop something

The decision low-volume measurement is usually being asked to support, and the one it is worst at.

A channel that has produced nothing in three months has produced nothing in three months. At your volume that is perhaps twelve enquiries total across everything, so the absence of one or two is not evidence of much.

Two rules keep this sensible.

Give anything a full year before cutting it on performance, unless the cost is significant enough that the money itself is the argument. Seasonality alone can hide a channel's entire contribution inside one quarter.

And separate the cost question from the performance question. A directory costing very little that produces two jobs a year is fine and does not need justifying further. The same directory at ten times the price is a different decision, and it can be made on cost without waiting for data.

Where you do stop something, stop only one thing at a time and note the date, so that any change afterwards has one candidate explanation rather than three.

The counter-case

Proper attribution earns its cost in some small businesses, and the threshold is about volume rather than size.

An online shop doing four hundred orders a month has plenty of events, even with three staff, and modelling will tell it something real.

Anybody spending seriously on advertising needs conversion data flowing back to the platform, which is a requirement of the advertising working rather than a reporting preference.

And a business genuinely testing between two channels needs enough measurement to tell them apart, though it also needs to accept how long that will take at its volume.

The test is whether a finer answer would change a decision you are actually able to make.

What to do instead

1. Keep one sheet with source, quoted, won and value.
2. Use broad categories, five or six at most.
3. Review quarterly, and properly once a year.
4. Sort by value won, never by enquiry count.
5. Track repeat and referral as a single proportion.
6. Remember last touch flatters search when you read it.

Six columns, twelve months, and it will answer more than a model would have.

Choosing which figures to watch is covered in the first five numbers worth looking at.

Frequently asked questions

West Coast Media Solutions Inc. provides web design, web development, hosting, digital marketing, and business consulting to organisations across Canada, drawing on more than twenty-five years in the field.