Influencer attribution that survives a finance review

The standard influencer report shows revenue attributed to discount codes, divides by spend, and calls the result ROAS. Finance teams are right to be sceptical of it, and the reason is worth understanding properly.

The problem with code-based attribution

A creator posts. Their code gets used 200 times. The report claims 200 sales.

Some of those sales would have happened anyway. The customer was already going to buy, searched "brand name discount code" at checkout, and found the creator's code on a coupon aggregator. You paid a creator fee and gave a discount for a sale you already had.

This isn't a rounding error. For brands with strong existing demand, the share of code redemptions that are non-incremental routinely runs high enough to flip a campaign from profitable to negative.

Three layers, in order of rigour

Layer 1: Tracked links and unique codes

Still worth doing. It's the cheapest signal available and it tells you relative performance between creators, which is genuinely useful for deciding who to re-book.

What it cannot tell you is whether the programme as a whole made money.

Practical notes:

• One code per creator, never shared
• Codes that are hard to guess. PRIYA15 gets scraped and aggregated within days, whereas a less obvious string survives a lot longer
• UTM parameters on every link, with a consistent naming convention
• A holdout on code visibility: don't let codes appear in your own site's checkout suggestions

Layer 2: Post-purchase survey

One question at checkout: "Where did you hear about us?", either free text or a short list that includes creator names.

It's self-reported and therefore imperfect, but it catches the entire category of influence that no tracking picks up: the person who saw a video, didn't click, and searched for you directly three weeks later. That path is invisible to Layer 1 and it's frequently the majority of real creator impact.

Compare survey attribution against code attribution. Large gaps in either direction are informative:

• Survey ≫ codes: creators are driving genuine discovery your tracking is missing. You're under-crediting the channel.
• Codes ≫ survey: your codes are leaking to coupon sites. You're over-crediting it.

Layer 3: Geo holdout testing

The only method that actually measures incrementality.

Split your market into matched geographic regions. Run creator activity in some, hold the others dark. Compare total revenue between them over the test window, and note that this means total revenue, not attributed revenue.

The gap is your incremental lift. It is the number a finance team will accept.

Requirements, honestly stated:

• Enough volume for the regions to be statistically comparable. Below roughly a few hundred orders per region per week, the noise swamps the signal
• A test window long enough to cover your purchase consideration cycle
• Discipline to leave the holdout dark even when someone senior wants to "just try one creator there"

That last point is where most geo tests die.

What to actually report

Give three numbers, clearly labelled, and never collapse them into one:

1. Tracked revenue, from codes and links. Precise, and an undercount.
2. Survey-attributed revenue, broader and self-reported.
3. Incremental revenue, from the last geo test, with the date it ran.

A report that presents one blended number is either overstating or understating, and the person reading it has no way to tell which.

The uncomfortable finding

Run a proper geo test and there is a fair chance you will find that your best-performing creator by code volume is one of your least incremental ones, because their audience overlaps almost entirely with people who already buy from you.

That's not a reason to avoid testing. It's the entire reason to test. The alternative is scaling spend against a number that flatters itself.