Attribution for a Business With Four Channels

Why the models do not apply

Attribution models were built for businesses with tens of thousands of interactions, where a percentage shift is statistically meaningful.

A business with forty enquiries a month across four channels has ten per channel, which cannot support any weighting.

Applying a model to that produces confident percentages calculated from single-digit differences.

Which is worse than having no attribution, because the numbers look authoritative.

The four channels most small businesses have

• Search, branded and unbranded.
• The local listing.
• Referral and word of mouth.
• Repeat customers.

Two of those four are invisible to any tracking. A referral who searches your name is recorded as branded search, and a repeat customer who types your address is recorded as direct, so the tracked picture systematically overstates search.

Ask, which is the whole method

Since the alternative does not work at this scale.

One question on the form or in your first reply, in plain words, asking how they came across you.

Write the answer beside the enquiry in the same record where you count them.

After a hundred enquiries you have a distribution that no tracking arrangement could have produced, covering the two invisible channels.

It is imperfect, since people misremember and some give the last step rather than the first.

It is considerably more accurate than any model applied to these numbers, which is the standard it has to beat.

A worked example

A business had analytics showing most enquiries coming from search and concluded that was where the work came from.

Asking directly for six months produced a different picture: about a third referral, a quarter repeat, a quarter search, and the rest from their listing.

Most of the referrals had searched the business name before making contact, which is why the tracking had counted them as search.

They had been about to increase spending on search and instead put the effort into referrals.

The tracked figure had not been wrong about what happened last; it had been wrong about what caused it.

Six months of a single question changed a spending decision.

The useful question is which to stop

Rather than how to divide credit.

Nobody with four channels needs to know that search deserves thirty-one per cent.

What they need to know is whether any channel is producing so little that the money and time should go elsewhere.

That is a much cruder question and it is answerable: count the enquiries from each over six months and look at the smallest.

Where a channel produced two enquiries in six months and costs a monthly fee, the decision makes itself.

Attribution done for that purpose is proportionate and done for weighting is not.

Assisted paths are real and unmeasurable

Worth accepting rather than solving.

Somebody sees a van, searches later, reads three pages, forgets, sees a recommendation in a local group, and calls a month afterwards.

No arrangement captures that, and asking produces whichever step they remember.

So a channel producing few direct enquiries may still be doing substantial work, particularly anything building recognition.

That is an argument for not cutting a channel purely on its direct count, and for a longer window before deciding.

Six months rather than one, and a conversation rather than a calculation.

Where a channel is genuinely borderline, pausing it for a quarter and watching the total is more informative than any attribution.

Record the first touch you can establish

Which is more useful than the last.

The last interaction before an enquiry is nearly always a search for your name, which tells you nothing.

How they first heard of you is the question worth the answer, and it is what a direct question naturally produces.

So phrase it as how did you first come across us rather than how did you find us today.

That single word changes what people report and it changes what you learn.

Keep it in one column

Since the recording has to be trivial to survive.

One column beside the enquiry count, with about six possible answers written the same way each time.

Not a system, not a field in a tool, not a dropdown somebody has to maintain.

Consistency in wording matters more than granularity, since twelve variations of referral cannot be counted.

Review the distribution once every six months and not more often, because the numbers are too small to read monthly.

Reading it monthly produces exactly the reactive changes that the whole approach exists to avoid.

Ask a second question when it becomes a job

Which produces the figure that actually matters and costs nothing extra.

Enquiries by channel is useful, and enquiries that became jobs by channel is what decides where the effort goes.

A channel producing many enquiries that rarely convert is worth less than one producing half as many that usually do, and the first figure hides that entirely.

Add a second column recording whether each enquiry became work, which you know anyway and are probably recording somewhere else.

After a hundred enquiries that pair tells you which channel sends the customers you actually want, which is a different question from which sends the most.

The counter-case

Some businesses need more.

Anybody spending substantially on advertising needs to know what that spending produced, and a direct question is too coarse for that.

Businesses with hundreds of monthly enquiries have enough volume for a model to say something.

And asking has its own bias, since people report what they remember rather than what happened.

The defence is volume: a hundred imperfect answers describe the pattern well enough to act on, where ten do not.

Ask how they first came across you, record it in one column, review every six months, and use it to decide what to stop.

The method

1. Add one question to your form.
2. Ask how they first came across you.
3. Record it in one column.
4. Use consistent wording.
5. Wait for a hundred enquiries.
6. Look at the smallest channel.
7. Decide what to stop.

Step two is the wording that matters, since asking how they found you today produces a search answer from somebody who was recommended you last year.

The single-channel version is covered in attribution for a business with one salesperson.

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.