Out-of-home is the channel with the least to count. No click, no matched household, no reliable impression, and flights booked months ahead, usually in the same weeks as the TV campaign they support. The question "how do we measure the effect of out-of-home on sales" is really two questions: how to get a number at all, and how to stop that number being TV's number in disguise. Marketing mix modelling answers the first by relating weekly out-of-home spend to weekly sales alongside every other channel. The second takes a plan. This guide covers both, and the data you need to start.
Model out-of-home as its own channel, with spend dated by the weeks the flights were up, from the booking confirmations and flight reports. Then make sure the model can tell it from TV: if the two always run in the same weeks, the model sees one combined effect and cannot split it. One planned week of out-of-home without TV, or a different flight pattern, is worth more than another year of the usual data. Where out-of-home is bought city by city, geography is the strongest tool you have, because a campaign in some markets and not others is a clean test.
Out-of-home is measured three ways, and they answer different questions.
| Method | What it measures | What it cannot tell you |
|---|---|---|
| Audience measurement (reach and frequency from traffic and mobility data) | How many people passed the sites, how often | Whether any of them bought anything |
| Brand lift surveys | Recall and consideration among people exposed, by survey | Sales, and the comparison with other channels |
| Marketing mix modelling | Incremental sales per unit of out-of-home spend, next to every other channel, with a confidence range | Which site or format worked, unless each is big enough to be its own channel |
The first two are useful inputs. Reach data helps set the shape of the out-of-home curve before the model trains, and a lift survey is a sanity check on the level. Neither replaces the model, because neither connects to sales. The full guide to marketing mix modelling explains the method; this guide is about out-of-home specifically.
Out-of-home spend lives in booking confirmations and the flight reports the media owner or agency sends after a campaign. The model needs it as weekly spend by flight week, going back as far as the sales history, with gross versus net, agency fee and production made explicit. Three decisions shape the setup:
Odins takes these files through a pipeline from the agency, so out-of-home lands as weekly spend in the same structure as the digital channels. The data guide sets out the full agency request.
This is where most out-of-home measurement fails, and it fails before the model runs. Out-of-home is planned as support for TV, so it goes up the week the TV flight starts and comes down the week it ends. The model then sees two channels that move as one. Whatever it reports for out-of-home is a share of a combined effect, decided by the priors rather than by the data.
The fix is in the plan, and it is cheap:
The prior decides, unless you plan
When two channels always run together, a Bayesian model does not refuse to answer. It returns an answer shaped by its starting assumptions, and the range will be wide. A good vendor shows you that range and calls the channel a test candidate rather than quoting the midpoint. If the range on out-of-home rivals the estimate, the next step is a test, not a bigger budget.
Out-of-home rarely gets a per-flight reading. The model reads the channel over the year: a response curve with a confidence range, the average return per unit spent, and the marginal return at today's level. Two things about that result are specific to out-of-home.
First, the horizon. A share of what out-of-home does is brand, and those sales arrive over months rather than weeks. Compare out-of-home with search on a three-week horizon and search wins every time. Odins splits each channel's effect into sales within weeks and a share that builds brand and pays out later, so the comparison can be made on the horizon the business plans on. See brand effects in the model.
Second, the headroom. Out-of-home is often bought at the same level for years, which means the model has seen it at one point on its curve. The estimate of what more would do is a range, and the way to narrow it is to vary the spend on purpose. A structured test designed by the model tells you how much to vary it by and for how long.
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.
Yes. The model runs on weekly spend, which the booking confirmations give you. Reach estimates from the media owner help shape the curve but are not required. The real constraint is variation: out-of-home that always runs with TV, at the same level, cannot be separated from TV without a planned change.
As separate channels, once each is big enough to carry a curve. Below that, one out-of-home channel is the honest setup. Programmatic digital out-of-home has one advantage: it can vary week to week, which is exactly what the model learns from.
For out-of-home it is often possible, because the buy is by city. Run the campaign in some markets and not others and read the difference. Then let that result calibrate the model, so one model answers the budget question across every channel rather than a test result sitting beside it.
The direct effect lands during and shortly after the flight. The brand part plays out over months. A model that counts only the first weeks undercounts out-of-home against channels close to the purchase.
Tools that collect the offline spend for you, model each offline channel separately, show the confidence range and can design the test when the range is wide. Odins does this for retailers and e-commerce companies.
Read the complete guide to marketing mix modelling, see how offline media data gets into the model, or book a demo and bring your out-of-home and TV plans.