Attribution for small firms: good enough beats perfect

A 5–50-staff B2B service firm does not need attribution software; it needs a mandatory source field on every deal, one "how did you hear about us?" question asked and recorded, and the discipline never to leave either blank. That covers the great majority of the decisions attribution exists to support — which channels to fund, which to fix, which to stop. Perfect multi-touch attribution is a research project; good-enough attribution is a CRM field.

Why is perfect attribution a trap at this size?

Because the cost of precision rises steeply while the value of precision stays flat. Enterprise attribution tools exist to divide credit across dozens of touchpoints and thousands of monthly leads, where a 2% reallocation of a seven-figure budget pays for the tooling. A firm winning three to eight clients a month faces different maths: the sample sizes are too small for statistical elegance, and the decisions are coarse — keep the channel, fix it, or kill it. You do not need to know that a webinar deserves 23% of the credit; you need to know whether outbound, referrals or the website is producing clients.

There is a second trap: sophistication becomes an excuse. Firms postpone recording anything because they are evaluating tools, and eighteen months later still cannot say where last quarter's clients came from. Meanwhile the numbers that should sit on the management dashboard — the small honest set defined in The MD Dashboard Blueprint — have a blank where "revenue by source" should be.

What does good-enough attribution actually require?

Three components, all cheap:

1. A source field with a controlled list. Outbound, referral, website, event, partner, existing-client expansion. Six to eight options, no free text, mandatory at deal creation. Free text produces "LinkedIn", "linked in" and "Li" as three different channels.
2. The question, asked out loud. "How did you hear about us?" on the first call, recorded in the CRM verbatim where the answer is interesting. Self-reported attribution is unfashionable and remarkably useful — people usually remember the thing that actually moved them.
3. Consistent recording at the moment of creation. Attribution captured later is fiction; nobody reconstructs a lead's origin accurately in week six.

Note what is absent: pixels, data warehouses, model selection. Those can come later, if ever.

How do you set it up?

The mechanism takes an afternoon and runs indefinitely:

1. Agree the source list. When a channel is one you actively invest in or could stop investing in, then it earns its own value on the list; everything else goes to "other". If "other" grows past roughly a tenth of deals, split it.
2. Make the field mandatory. When a deal is created without a source, then the CRM rejects it — not flags it, rejects it. Optional fields decay into empty fields.
3. Capture the self-report. When the first conversation happens, then the "how did you hear about us" answer is logged; when it contradicts the recorded source, then the human answer wins and the field is corrected.
4. Report revenue by source monthly. Not leads by source — revenue. Leads flatter noisy channels; revenue tells you what feeds the firm. Pair it with cost, which is the subject of cost per lead is vanity; cost per client is sanity.
5. Decide quarterly. When a channel has produced no revenue for two consecutive quarters despite real spend, then it is fixed or stopped. When one channel exceeds roughly half of revenue, then that concentration itself becomes the risk to manage — the arithmetic of over-reliance is laid out in the referral-only pipeline maths.

What about deals with more than one touch?

They exist, and at this scale the pragmatic answer is to record the originating source and note the rest. A client who found you by referral, read four articles and replied to a cold email is genuinely multi-touch — but the decision the data must support is still "keep funding referrals, content and outbound?", and a first-touch source plus an honest note answers it. When two channels routinely appear together, treat that pairing as a finding, not an accounting problem. The precision lost is real; it is also smaller than the error introduced by leaving fields blank while waiting for a perfect model.

The deeper risk to guard against is not crude attribution but dishonest reporting built on top of it — pipeline and revenue numbers distorted before attribution is even considered, which is covered in where sales reports lie.

Which decisions does this actually support?

Budget, effort and pricing. You learn which channel produces clients rather than enquiries, which produces the clients who stay, and roughly what a client costs to acquire from each. That is enough to reallocate spend, to justify or retire a channel, and to notice concentration risk early. And because the data lives in the CRM as structured fields, the monthly source report can compile itself — the same principle behind the weekly report that writes itself. Good enough, recorded always, beats perfect, recorded never.

---

Next step: the Growth System Audit — £450, seven days, credited against any build — maps where your growth system leaks, including whether you can actually see where clients come from, and what to build first.