Measuring offline media honestly
Billboards, radio, TV, print and events do not produce a click trail, and any report that assigns them an exact number of sales should be read with suspicion. They can still be measured. Geo experiments compare areas that saw the media with areas that did not. QR codes, vanity URLs and promo codes count direct responses but miss most of the effect. Brand search lift shows interest, not sales. Marketing mix models estimate contribution over time but need months of varied spend. Use the strongest method the plan can support, report it as a range with its method, and decide in advance what result would cut or keep the channel.
Executive findings
- 1
Google researchers describe geo experiments, where non-overlapping regions are randomly assigned to treatment or control, as a conceptually simple way to measure ad effectiveness with results that are easy to interpret.
Documented: official source. Scope: Google Research publication on geo experiments, as of . [1]
- 2
Google’s Meridian and Meta’s Robyn are open-source marketing mix modelling tools; Meridian supports geo-level data, which lets a model learn from spend that differs between regions.
Documented: official source. Scope: Official project documentation for Meridian and Robyn, as of . [2]
- 3
IAB guidance on modernising MMM and on incremental measurement in commerce media recommends combining models with experiments rather than relying on one method.
Secondary source. Scope: IAB guidelines, US-led and used internationally, as of . [4]
- 4
In our illustrative example, QR scans account for under a tenth of the lift a matched-market test reads for the same outdoor flight, which is why codes should be reported as a floor.
Illustrative model, not a benchmark. Scope: FOCUSADS worked model with labelled assumptions, not a benchmark, as of
Method and limitations
- Question
- How should an advertiser measure offline media so that decisions are defensible without claiming precision the data cannot support?
- Written for
- CFOs, heads of growth and marketing managers adding outdoor, broadcast, print or events to a digital plan.
- Decision at stake
- Which measurement method to set up before booking offline media, and how to report the result to the board.
- Sources
- Google Research on geo experiments, documentation for Google Meridian and Meta Robyn, IAB guidance on MMM and incrementality, and Google Ads documentation on offline conversion imports.
- Period
- Documentation as published on 17 October 2026.
- Markets
- All markets; geo methods need enough comparable regions in the country.
- Definitions
- Geo experiment: regions assigned to see or not see the media, with outcomes compared afterwards.
- Matched markets: regions chosen in pairs with similar past sales, one exposed and one held out.
- Brand search lift: the rise in searches for the brand name in exposed regions against held-out regions.
- Marketing mix model (MMM): a statistical model estimating each channel’s contribution from spend and outcome data over time.
- Limitations
- Single-location businesses cannot run geo experiments; they are left with pre/post comparisons and direct-response codes, which carry larger uncertainty.
- National TV or radio cannot be withheld by region in many markets, which rules out simple geo tests.
- We publish no typical lift figures; they depend on category, creative and weight.
Five methods and what each can claim
FOCUSADS view
The methods differ in what they can prove, what they cost and when they must be set up. The table below orders them from the weakest claim to the strongest.
| Method | What it shows | What it cannot show | Set up before booking? |
|---|---|---|---|
| QR codes, vanity URLs, promo codes | People who responded directly | Anyone who saw the ad and bought later another way | Yes, codes go into the creative |
| Brand search lift | Rise in interest in exposed areas | Whether interest became sales | Held-out areas help |
| Pre/post comparison | Change over the flight | What would have happened anyway | No, but needs a clean baseline |
| Geo experiment or matched markets | Incremental outcomes in exposed areas | Effects outside the test period | Yes, areas must be chosen first |
| Marketing mix model | Estimated contribution across channels over time | Precise effect of a single flight | Needs months of varied spend |
Worked model: why codes are a floor
Illustrative model, not a benchmark
An outdoor flight runs in four matched areas, with four areas held out. The creative carries a QR code.
| Line | Value | How it is calculated |
|---|---|---|
| Orders in exposed areas | 2,760 | Assumption |
| Orders expected without the flight | 2,400 | From held-out areas’ trend |
| Incremental orders read by the geo test | 360 | 2,760 minus 2,400 |
| Orders through the QR code | 31 | Assumption |
| Share of the measured lift seen by the code | 8.6% | 31 ÷ 360 |
Assumptions
- All volumes are placeholders. The point is the ratio: most people who act on outdoor media do not scan a code.
- The geo read itself has a confidence interval; report it, for example 360 incremental orders with a range of 210 to 510.
- Effects after the flight are excluded.
Reporting offline results honestly
FOCUSADS view
Keep offline results in their own section of the report. Name the method, the period, the comparison areas and the range. Do not add modelled offline sales to platform-tracked conversions in one total, because the two have different certainty. If the board needs one number, use the low end of the range.
Where sales happen offline after an online click, import them back into the ad platform through offline conversion imports. That is a different job from measuring offline media, and the two should not be confused in reporting.
What we would test, in order
| Step | Measured by | Counts as success when |
|---|---|---|
| Pick the method before booking: list regions, check that comparable held-out areas exist, and add codes to the creative. | Week-to-week variation of the outcome across candidate regions. | The detectable lift at the planned budget is below the lift needed to break even. |
| Run the flight in exposed regions only and freeze other spend changes in all test regions. | Outcomes, brand search and code responses by region. | Exposed regions beat held-out regions by at least the break-even lift, with the range reported. |
| Feed the result into a mix model once there are several flights with different weights. | Model estimate against the experiment reads. | Model and experiments agree within their ranges, or the model is recalibrated. |
Which method fits your situation
Use the strongest method the plan can support.
| Situation | Strongest practical method |
|---|---|
| Several comparable regions, regional media | Geo experiment or matched markets |
| One location, local media | Pre/post with codes and brand search, reported with caveats |
| National broadcast only | Mix model over several flights, plus brand search |
| Trade show or event | Named-account follow-up in the CRM, with a pre-agreed qualification rule |
How this connects to our method: Launch with measurement in place
References
- Measuring Ad Effectiveness Using Geo Experiments, Google Research, accessed
- Meridian: open-source marketing mix model, Google for Developers, accessed
- Robyn: open-source marketing mix modelling package, Meta Open Source, accessed
- Modernizing MMM: Best Practices for Marketers, IAB, accessed
- Guidelines for Incremental Measurement in Commerce Media, IAB, accessed
- About offline conversion imports, Google Ads Help, accessed