What brand outcomes measurement misses about TV advertising in the performance age
Opinion – The Future of TV: Global Series
Standard attribution measurement does not just undercount television; it can also overcount paid search. The same error causes the value to be moved from one channel to another. Econometric modelling for TAM Ireland helps explain how that happens.
Television’s share of UK ad spend has been falling for around two decades. Over the same period, the way brands evaluate media has shifted decisively from audience delivery to measured brand outcomes: visits, searches, downloads, conversions and sales.
Many forces have pushed the industry in that direction, but one has been hiding in plain sight. Attribution measurement has become a powerful arbiter of media value, especially in maintaining the ledger between television and paid search.
That ledger often makes two compensating errors at once.
It undercounts television because TV effects build over weeks and months, shaping consumer behaviour through pathways that click-based attribution cannot trace. It overcounts paid search because PPC (pay-per-click) often assigns last-click credit to demand that other channels helped generate earlier.
Value disappears from the television line and reappears on the paid search line. The book balances, but the entries are wrong.
A new econometric modelling project from Television Audience Measurement Ireland and Colourtext, The TV Persistence Dividend: Cut TV Once, Pay Twice, provides evidence of this pattern across 12 major brands in four market categories.
Two findings matter most
The first is persistence. Econometric modelling captures the rate at which an advertising effect fades through an adstock decay parameter, known technically as alpha.
Slow fade, longer memory
In plain English, alpha tells us how slowly the light bulb dims after the ad has played out. A high alpha means a slow fade and a longer memory effect.
Nine TV-focused models were selected from a larger modelling pool because they showed clear evidence of this persistence effect. Eight of the nine showed a long-memory effect for TV.
Across those models, the median TV half-life was around nine weeks. That means half of television’s accumulated advertising effect was still influencing brand outcomes nine weeks after exposure.
The implication is direct. Attribution windows measured in minutes, days or even a few weeks are not long enough to capture much of what television delivers.
If a TV-primed consumer visits a website six or eight weeks after exposure, the conversion is unlikely to be linked back to the scheduled TV impact. It is more likely to be credited to whatever sat closest to the click.
On many occasions, that channel is paid search.
The second finding concerns interaction. A standard additive marketing mix model assumes each channel works independently from anything else on the schedule. Television does its work, paid search does its work, and the model adds the two together.
A channel interaction model asks a sharper question: does one channel become more productive when another is active?
In two of our model cases, one for web visits for an online services brand and one for app downloads for an insurance brand, paid search became materially more productive when television was running alongside it.
In the online services model, paid search delivered around 51% more web visits when supported by television at typical TV weights. In the insurance app downloads model, the lift was closer to 69%.
Standard additive models struggle to capture that kind of channel synergy. Most attribution systems are even less likely to see it, because they assign credit to the touchpoint closest to the response.
The Google tax
This is where the Google tax appears, and its logic is simple.
Television creates demand for a brand; the consumer later arrives at a search results page already inclined to choose it, and then paid search steps in to capture the click.
The brand pays once for the TV advertising that built the intent, then again for the paid click that captured it.
Our analysis found this dynamic across three brands in the supermarket, insurance and retail categories. PPC interception rates ranged from around 3% to 12% across app downloads, organic search referrals, web visits, and branded keyword searches.
In the insurance app downloads model, around 17% of the app downloads credited to paid search were likely TV-primed misattributions.
That number is not a curiosity. It is the sort of metric that later budget decisions are built on.
None of this means paid search is ineffective, or that advertisers should stop investing in it.
PPC can be highly effective at converting intent where intent already exists. The problem starts when attribution systems treat captured demand as if it were created demand.
That distinction matters because attribution numbers do not stay inside research reports – they travel.
They feed automated bidding systems, budget allocation tools, dashboards, post campaign reviews and planning recommendations.
Channels credited with more conversions per pound look more efficient. Channels credited with fewer look weaker. Budget then moves toward the former and away from the latter.
Television’s share is pushed downward by a mechanism that looks rational because the credited conversion figure appears to justify the move. Yet if that figure has undercounted TV and overcounted PPC, the budget decision is rational only inside a flawed attribution system.
The commercial effect is slow and cumulative.
Brand memory drains away
Cut TV and the reservoir of brand memory does not disappear overnight. It drains away gradually.
Paid search may continue to look healthy for several weeks because it is still converting demand that earlier TV activity helped create. By the time that support has faded, the next planning cycle may already have run on the same biased numbers.
This is not just a television problem. The same logic applies to any channel whose effects build over a longer horizon than common attribution windows can see.
Radio, outdoor, cinema and press can all be undervalued when the credited conversion becomes the only figure that counts.
That does not mean the movement of budget towards short-term response platforms is entirely an attribution error. It does mean that some of it may have been accelerated by systems that see the end of the customer journey more clearly than the beginning.
This is one modelling project, not a definitive verdict on every brand or every media plan. But it gives measurable shape to a structural problem.
Standard attribution is not always the impartial referee it appears to be. It can misread television, misread paid search and misread how the two work together.
What you can do
Three practical changes follow.
1) Evaluation windows should match the persistence of the channels being judged. If TV effects have a nine-week half-life, a short attribution window will not do.
2) Models should test whether channels are working together, not simply assume they work independently.
3) Before budget is moved out of television and into paid search, brands should ask whether some of paid search’s apparent productivity is being lent to it by upstream TV exposure.
These are not exotic demands. They are basic safeguards against a familiar measurement error. The channel that cannot be measured properly becomes the channel that gets cut.
And even if we are not interested in attribution measurement, attribution measurement is certainly interested in us.
Jason Brownlee is the founder of Colourtext