Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Analytics › Attribution

Measuring TV, Radio, Podcasts and Print

How to measure media that leaves no click: response analysis around airings, vanity URLs and codes, search and direct lift, regional tests and mix models.
Guideadvanced12 min readUpdated

Why offline media needs different evidence

Attribution models divide credit among recorded touchpoints, and a television spot, a radio read, a podcast mention or a print page is not recorded on anyone's path. The attribution models comparison explains why no rule, including data-driven, can credit a touch it never saw. Offline media therefore needs evidence that does not depend on paths. There are two kinds.

The first is traceable response: some share of the people who respond do so through a mechanism the advertiser controls, such as an address, a code or a phone number used only in that advert. That share can be counted exactly, but it is only a share, and it leaks in both directions.

The second is aggregate change: the outcome, or a close signal of it such as branded search, moves when the media runs and does not move when it does not. Read carefully against a baseline, and best of all against a comparison group that did not receive the media, aggregate change measures the whole response, including the people who never used a code. It is also noisier and needs more planning. Addressed mail is the exception among printed media, because the business knows who was sent each piece; the guide to measuring direct mail draws the comparison group from the mailing list itself.

The post-campaign analysis guide covers the other half of offline measurement: whether the buy delivered the Reach, frequency and cost that were paid for. That analysis proves delivery, not effect. This guide is about effect. Whether audio belongs in the plan at all, and when to choose sponsorship, guest appearances or an own show, is the subject of the guide to podcasts and audio in the marketing mix.

Response mechanisms: vanity URLs, codes and numbers

A response mechanism gives the people who respond a route that only the advert told them about.

Vanity addresses. A short, memorable path or domain, said aloud or printed, that redirects to the landing page with campaign parameters attached. Google's URL builder documentation describes adding campaign parameters to a URL so that analytics can collect the campaign data, and the redirect is how a listener who types a short address arrives with those parameters. The parameters follow the same UTM convention as every other campaign, as the UTM and naming conventions guide sets out, and the redirect has to pass them through, because a redirect or shortener that strips them sends the visit to direct. A QR code in print or on screen can encode the tagged address directly. Testing those redirects, and recording the baseline before the first spot, are lines of the offline media flight launch checklist to tick in the days before a flight starts.

Codes. A discount or offer code used only in one medium, one show or one publication. Codes catch people who buy later, on another device or through another route, because they enter the code at checkout whatever path they took.

Dedicated phone numbers. A number printed or read out only in one advert, so calls to it can be counted and matched to orders or bookings.

Each mechanism leaks, and the direction of the leak matters.

MechanismWhat it countsLeaks out (undercount)Leaks in (overcount)
Vanity address or QR codeVisits that used the addressPeople who search for the name or type the main domain; redirects that drop parametersThe address shared online or reused in other material
CodeOrders that entered itBuyers who forget the code or never needed a discountCodes collected by deal sites and used by people who never saw the advert
Dedicated numberCalls to that numberCallers who look up the main numberNumbers saved and reused long after the campaign, or listed in directories

The rule that follows is to read every mechanism as a floor with some contamination, not as the campaign's total. Mechanisms are most useful for comparing like with like: two podcast shows with the same offer and code format, two print titles with the same address pattern, two flights of the same advert. The ratio between them is more trustworthy than either absolute count. A generous code also changes what is being measured, because some of the response is to the discount rather than the advert, so keep the offer constant across the placements being compared.

Asking is the fourth mechanism. A "how did you hear about us?" question on forms, at checkout or on the sales call catches people who used none of the routes above; the self-reported attribution guide covers wording, placement, coding and the biases that make offline media both over-named and under-named.

Response around airings

Direct-response television and radio often produce a visible spike in site visits, searches, calls or app installs shortly after each spot airs. Where the media runs as many spots at known times, that spike can be measured spot by spot.

The method needs two inputs at a fine time grain. The first is the airing log: the actual time each spot ran, by station or channel, from the broadcaster's or agency's confirmation rather than from the plan, since spots move. The second is the response series at the same grain: site sessions, branded searches, calls or orders, counted per minute or per few minutes. For each spot, compare the response in a short window after it with a baseline for the same window, built from the same time of day on comparable days without a spot, or from the period just before the spot. The difference, summed across spots, is the immediate response; split by station, daypart, creative or spot length, it becomes a buying signal. The guide to buying radio advertising covers how stations, dayparts and script rotation are chosen in the first place, and how the station's log is checked against the schedule.

Three conditions limit it. The window has to be set from your own data by looking at how quickly the response decays, and it should be fixed before the analysis rather than tuned to flatter the result. Spots close together blur into each other, so the method works best on schedules with gaps. And the spike is only the immediate response: people who search tomorrow, buy next week or tell a friend are outside it, so for brand campaigns and considered purchases it understates the effect, sometimes by most of it. Treat it as a measure of which spots work better than others, not of what the campaign earned.

Search and direct lift

When there is no spot-level timing, as with most print, podcasts and outdoor, the next signal is the daily or weekly movement in branded search and direct visits. People who hear a brand name and want to act on it either search for it or type it, so those two series respond to almost every offline medium.

Read them against a baseline, never against the previous week. A usable baseline accounts for the series' trend and seasonality: the same weeks last year adjusted for growth, a forecast from the weeks before the campaign, or, best of all, regions or markets where the media did not run. Annotate every promotion, launch, press mention and change in paid search activity, because each moves branded search for reasons unrelated to the offline campaign. The brand awareness and lift guide treats Branded Search as a proxy for awareness and explains its weaknesses, chief among them that it measures interest from existing customers as well as new ones.

Two cautions apply. First, a lift in branded search is attention, not sales; follow it through to orders or leads from those sessions before claiming a commercial effect. Second, paid brand search ads will capture much of the lift and claim it under last click, which is the mechanism by which offline media disappears from attribution reports in the first place. Read organic and paid brand search together, and do not let the paid search team's figures improve because the radio campaign started.

Regional tests

Many offline buys can be bought by region: local radio stations, regional television, regional press editions, outdoor by city. That makes the strongest evidence possible: run the media in some regions, hold it back in comparable ones, and compare total outcomes. The held-back regions are the Holdout Group, and the difference between the groups after adjusting for how they compared before the test is the incremental outcome, including sales in shops, by phone and through partners that no tracking would connect to the advert. The same design can check a plan's regional split, and the tutorial on weighting a media plan by region sets one up to test whether heavier weight in one region pays back against a matched comparison region.

The incrementality testing guide covers the design: matching regions on their outcome history, choosing the minimum effect worth detecting before the test, running across a full purchase cycle with a cool-down afterwards, and reading the interval rather than the midpoint. The comparison of geo holdout tests and conversion lift studies explains why a regional design is the natural one for offline media, which no platform can randomise at the level of individual people.

Some media resist geography. A national podcast or a national newspaper reaches every region at once, so the comparison has to come from time instead: flighting the media on and off in planned blocks, so that weeks with and without it can be compared against the same baseline. A planned on-off schedule is weaker than a regional test, since anything else that changed in those weeks is confounded with the media, but it is far stronger than a single burst with nothing to compare it to. The comparison of radio and podcast advertising shows which of these rungs carries most weight for each medium when a plan is choosing between them.

Illustrative example — not a real result

Thistlebeam Beds reads a radio burst and a podcast read

Thistlebeam Beds, an invented bed retailer with a website and showrooms, runs a four-week radio burst in two of its six sales regions and a national podcast read across the same weeks with the code DREAM15 and the vanity address thistlebeam.example/podcast. Branded searches in the two radio regions rise from a baseline of about 2,100 a week to an average of 3,050 during the burst, while the four other regions move from 4,300 to 4,520, in line with the podcast and seasonal drift. Direct visits show the same shape. Showroom and online orders in the radio regions are 11 per cent above their matched baseline over the four weeks and the two weeks after, against 3 per cent in the comparison regions, so the team estimates the radio effect at roughly 8 per cent of orders in the radio regions, with a wide interval because only six regions exist. The podcast's vanity address records 640 visits and the code is redeemed on 97 orders, 14 of them by customers who arrived from a deal site. The team reads the podcast figures as a floor, books a planned on-off schedule for the next two podcast flights to get a time comparison, and extends radio to one further region with the remaining three as comparisons.

Mix models for the long run

When offline media is a large and recurring part of the budget, the question shifts from "did this burst work?" to "how much should each medium get next year?". That is the question marketing mix modelling answers: a statistical model of weekly outcomes against spend in every channel, with controls for price, promotions and season, which estimates each channel's contribution, carry-over and diminishing returns without ever looking at an individual path. It is the only method on this page that sets offline and online media side by side on one scale.

A MMM needs a long, consistent history with variation in each channel's spend, which is where the earlier rungs feed it. Regional tests and planned flighting create the variation the model needs to tell channels apart, and their measured lift gives points the model's curves should pass through. A mix model built on a schedule that ran the same weight of radio every week of every year has nothing to learn from; one built after two years of deliberate regional and on-off variation has plenty. The guide to preparing data for a mix model covers how offline spend, usually held in agency invoices rather than platform reports, is turned into weekly series the model can use.

Choosing the evidence for the spend

The right evidence depends on the money at stake and on what the media can support. Small tests deserve cheap evidence; large, recurring commitments deserve the expensive kind.

SituationUseBecause
A first, small test of a podcast or print titleCodes, a vanity address and the self-reported question, compared across placementsCheap, immediate and good enough to rank placements against each other
Direct-response television or radio with many timed spotsResponse around airings, plus branded search liftSpot-level response shows which stations, dayparts and creatives to buy
A regional radio, press or outdoor campaignA regional test with matched comparison regionsMeasures the incremental effect on total outcomes, offline sales included
National media that cannot be split by regionPlanned on-off flighting, read against branded search, direct and sales baselinesTime is the only comparison available, so it has to be designed in
Offline media as a large, recurring share of the budgetA mix model, calibrated with regional testsSets every channel on one scale and supports annual allocation
Any offline spend at allThe self-reported question on forms and at checkoutCosts almost nothing and catches what every other mechanism misses

The rungs are cumulative rather than alternatives. A business that runs regional tests still uses codes to compare shows and still asks customers how they heard; the cheaper signals fill in between tests and flag where the next test should go. The attribution hub shows how these methods sit beside attribution and experiments in one measurement approach. The academy exercise on planning evidence for a radio and podcast burst for a bed linen brand practises choosing these rungs before a first broadcast spend, from response mechanisms and reads around airings to a regional test and what only a mix model could later show.

Common mistakes

  • Reading code redemptions as the campaign's total. They are a floor, contaminated by deal sites.
  • Using a redirect that strips campaign parameters. The vanity address then reports as direct.
  • Judging offline media on last-click reports. The demand it creates is credited to brand search and direct.
  • Comparing with last week instead of a baseline. Season, promotions and trend do the moving.
  • Tuning the response window after seeing the data. Fix it before the analysis.
  • Changing the offer between placements. The comparison then measures the discount, not the media.
  • Running national media at a constant weight forever. With no variation, neither a test nor a model can see what it does.
  • Treating the delivery report as proof of effect. Reach and frequency achieved are not sales caused.

Frequently asked questions

Can I put UTM parameters on a radio, podcast or print advert?

Not directly, because nobody types a long tagged address they heard or read. The usual route is a short vanity address, such as a memorable path on your own domain, that redirects to the landing page with campaign parameters attached, so the visits that do use it arrive tagged. A QR code in print or on screen can carry the tagged address itself. Both capture only the people who use them; most listeners and readers search for the name or type the main address, so the tagged visits are a floor, not the campaign's total response.

Do promo codes measure podcast advertising accurately?

They measure the people who remembered and used the code, which is a share of the response and not all of it. Codes also leak to coupon and deal sites, where people who never heard the show find and redeem them, and a generous code changes the offer, so some of the response is to the discount rather than the advert. Treat redemptions as one signal beside branded search, direct visits and what customers say when asked, and use a test when the spend is large enough to justify one.

How long after an offline campaign should I keep measuring?

Long enough for the response to fade back to the baseline, which depends on the medium, the product and the purchase cycle and should be read from your own data rather than assumed. Direct-response broadcast often shows a sharp response close to each airing; brand campaigns and considered purchases show a slower build and a longer tail. Set the measurement period before the campaign starts, including a period after it ends, so that the decision about when to stop looking is not made after seeing the numbers.

Terms used on this page

Frameworks

Where this page is used

Part of these learning paths

Sources

  1. [GA4] URL builders: Collect campaign data with custom URLs — Google Analytics Help

Related