Influencer marketing is measured, in most companies, by what can be counted: the discount code the creator reads out, the affiliate link in the bio, the reach the platform reports. The codes and links catch the people who used them. The reach number is not sales. Neither answers the question the budget meeting asks, which is whether influencer marketing drives sales that would not have happened anyway, and how much per unit spent compared with paid social or TV. Marketing mix modelling answers that question at the level of total sales, provided the spend is in the model and has varied enough to read. This guide covers what counts as influencer spend, when it earns its own channel, and how to plan the programme so it can be measured.
Put influencer spend in the model as a channel, dated by when the content was published, including fees, product costs and agency fees. Keep paid amplification of creator content in paid social, because that is what it is. When influencer spend is small and comes in a few bursts a year, group it with paid social until it is big enough to earn a curve of its own, and say so. Treat codes and links as a floor on the effect. And plan the programme in concentrated bursts rather than a steady drip, because a channel that never varies is a channel the model cannot see.
| Method | What it captures | What it misses |
|---|---|---|
| Discount codes | Buyers who used the code | Everyone who saw the post and bought through search, the app or a store without it |
| Affiliate links | Clicks from the bio or the post, and the purchases that followed within the window | Purchases on another device, later, or in a store |
| Platform reach and engagement | Views, likes, saves | Sales, and whether the viewers would have bought anyway |
| Brand lift surveys | Recall and consideration among exposed people | Sales, and the comparison with other channels |
| Marketing mix modelling | Incremental sales per unit of influencer spend, next to every other channel, with a confidence range | Which creator worked, unless the spend per creator is large enough to be its own channel |
Codes and links are worth keeping. They are free evidence, and they set a lower bound: the model's estimate for influencer should never be below what the codes alone brought in. The gap between the code count and the model's estimate is the part of the effect that was invisible before. The full guide to marketing mix modelling explains the method; this guide is about influencer specifically.
The model relates sales to money spent, so the first job is to define the money. Three lines belong in the influencer channel: creator fees, the cost of product and gifting, and the agency or platform fee for running the programme. All of it dated by when the content went live, not when the contract was signed or the invoice paid. A campaign with ten creators posting over three weeks is three weeks of spend.
One line does not belong there. When a creator's post is boosted with media money, through whitelisting, partnership ads or the platform's creator ad formats, that money buys paid social reach and should sit in paid social. Put it in the influencer channel and the model reads paid social's effect as influencer's. Keep the two apart and the model can tell you whether the organic reach of the creator added anything on top of the paid reach.
Odins takes influencer spend in through a pipeline from collaboration agreements and platform exports, so it lands as weekly spend in the same structure as the digital channels. The data guide covers the rest.
A channel under about two percent of media spend rarely earns a response curve of its own, and influencer programmes often start below that, in four or five bursts a year. Model it alone at that size and the estimate is mostly the prior you gave it. The working rule is to group influencer with the channel it behaves most like, usually paid social, note the grouping in the model, and split it out the day the volume justifies it. Splitting a channel out means a retrain, so it is a decision to make at the monthly update rather than mid-month.
Two things speed the day it gets its own curve: more spend, obviously, and more variation. A programme at a steady level every month, however large, is hard to read, because a channel that never changes level looks like baseline. Bursts give the model something to learn from.
The always-on programme that cannot be seen
A micro-influencer programme that posts every week at the same cost is the hardest case in the channel. The model sees a constant and credits it to the baseline. If that programme matters to the business, the way to measure it is to change it on purpose: pause it or double it for a planned window of four to eight weeks, and let the model read the difference. A structured test of this kind is cheaper than a year of wondering.
Most of what makes influencer measurable is decided before a single post goes live.
The model gives influencer a response curve with a confidence range, an average return per unit spent and a marginal return at today's level. Compare it with paid social on the marginal return, because that is what decides where the next unit of budget goes. Influencer also tends to carry a brand share, sales that arrive over months rather than days, so compare it on the horizon the business plans on. Brand effects in the model explains how that is credited.
What the model will not tell you is which creator worked. That is a question for the codes, the links and the creative team. The model sets the level of the channel and its place in the mix; the platforms and the creators decide what runs inside it.
Odins builds and runs the model, collects offline and digital spend every week, and delivers a monthly recommendation with a confidence range per channel. Companies like CDON, Nettbil, Hyre and Megaflis run their marketing budgets on it.
As part of a bigger channel, usually paid social, until the spend is large and regular enough to carry its own curve. A concentrated burst designed as a test is the fastest route to an influencer-specific number.
Any serious MMM can include influencer spend as a channel if the spend data is collected and dated properly. The difference between vendors is who does that collection every week and whether the model will say when the channel is too small to read. Odins collects it through a pipeline and reports the confidence range alongside the estimate.
Yes, at cost. It left the marketing budget and it bought exposure. Leave it out and the channel looks cheaper and more effective than it is.
Yes, as a floor and as creator-level feedback. The model's estimate should never be below what the codes alone brought in, and the codes tell you which creators converted, which the model cannot.
The direct effect lands in the days after the post; the model estimates how long it carries over rather than assuming it. A brand share plays out over months.
Start with the complete guide to marketing mix modelling, read how recommendations and structured tests work, or book a demo and bring your influencer plan.