MEDIA SCHEDULING AND CARRY-OVER EFFECTS

Is Adstock a useful TV planning tool?

By Erwin Ephron and Colin McDonald

The current US Recency-based scheduling model focuses entirely on the short-term effects of advertising, arguing that in competitive markets the data show that response to advertising dissipates rapidly. This approach leads to consistent moderate weight and near continuous scheduling.

Others disagree. Adstock, a well-established concept in the UK, has been seeping into US planning through globalization and marketing-mix modeling. It has introduced carry-over effects into US TV scheduling, which encourage GRP concentration, flighting and less continuous advertising.

We believe that the way Adstock is currently being used by US agencies to modify GRP lay-down is seriously flawed, because it is based on a misunderstanding of what Adstock is. This paper will try to explain the fallacy, and pinpoint just what place, if any, Adstock can have in media scheduling.

Adstock is described as ‘The impact that advertising has over time on sales or awareness,’ (Goerlich 2001). But it is richer than that. The Adstock concept captures the idea that response to an individual ad exposure does not stand alone, but is part of a continuity of advertising `pressure’ that follows on from exposures in the past and carries forward to those in the future.[1] It is a model of how response to advertising builds and decays in consumer markets.

Each new exposure lifts awareness (if that is our goal) to a new level, but that new level will be higher if there have been exposures in the fairly recent past and lower if there have not been. The idea here is the `advertising pressure’ (as measured by response) does not end as soon as the ad has been seen, but decays over time back to its base level, unless or until this decay is reversed by a new exposure.

Adstock is the mathematical modeling of this decay process. It is usually expressed in terms of the ‘half-life’ of the ad. A ‘two-week half-life’ means that it takes two weeks for half the awareness effect of that ad to be gone. Just as each radioactive element has its own ‘half-life’, so every advertising treatment/product/response measurement will have its own half-life.

The Adstock concept agrees with common sense. If all ad effects were instantaneous, you could never build a campaign because every new exposure would be working on a blank sheet, and `brand equity’ (what ever that actually means) would depend on nothing but brand experience. Adstock is given `reality’ (calculated) by modeling sales panel data or awareness/image tracking data against the corresponding advertising schedules. It is central to the econometric modeling of advertising effects over time.[2]

It is Adstock’s introduction to TV scheduling that is the problem. Here Adstocks have the planner move GRP’s on the flowchart from when they actually run to when the response (for example, sales) is expected to occur. It is advertising’s phantom version of Industry’s ‘just in time’ scheduling.

Plan GRP’s are redistributed based on a presumed half-life effect of the advertising on sales. For example, 100 GRP’s run in week one with a half-life of four weeks, might be flow-charted as 20 GRP’s in week one, 15 in week two, 10 in week three and five in week four, totaling 50 GRP’s over the four week half-life. The remaining 50 GRP’s would be assigned to the weeks that follow at a steadily decreasing rate.

When this is done for a year, the Adstock redistribution can have earlier GRP’s cascading over several months. This leads to questionable recommendations like ‘heavy-up GRP’s in lower cost periods to allow the Adstocks effect to cover the higher-cost periods that follow.’ Or ‘Adstocks show us we can safely skip several weeks in a schedule and still maintain advertising pressure.’

So, because Adstocks shows advertising effects continue across time as shown by sales modeling, the idea is presumably that we can afford to cut GRP’s now because we had advertised before. This somehow allows us to get the same result for less money. This is a fallacy. It is misleading, because it confuses cause with effect.

Adstock itself does not assume delayed GRP’s. It does not take anything away from the immediate effects of ads. What it says is that in addition to the immediate effect there is a further, continued effect decaying at a certain rate (until the next ad). The Adstock is extra to what you put in originally. The chart above gives the truer picture of what we expect from 100 GRP’s in Week One, if there is a four-week half-life. [3]

Adstock is an attempt to explain the mid-term effects of advertising. Let’s simplify and say brands advertise to increase sales. Advertising moves towards that goal in two ways. It both reminds and teaches. It reminds ready-to-purchase consumers about the brand in order to influence their immediate brand choice. And advertising teaches brand awareness and salience, which makes it easier for future advertising to influence brand choice.

The two effects are not short-lived or independent. For example, today’s advertising may result in purchase of a brand, which may result in repeat purchase. And today’s advertising may also reinforce awareness of that brand which may result in future purchase. These ideas are central to current US planning.

The puzzle for an Adstock scheduling model is ‘at what point in time do the sales and awareness created by current advertising become a property of the brand and cease being an effect of the advertising?’ It appears to happen quickly, as the von Gonten data show.

These data suggest that the presumed continuing sales-effect of advertising is in most part consumer repeat-purchase following advertising's initial effects. Since repeat purchase is usually driven by consumer satisfaction with the functional properties of the brand, the carry-over effects of advertising may have little do with it.

Adstock’s use in scheduling is at odds with the near universal observation that today immediate advertising effects dissipate rapidly with time. For example, Mike von Gonten’s sensitive X-10 model which tracks brand penetration in a scanner panel based on the startling insight that at any start-point no purchase has yet occurred. It then compares the rate of penetration growth to the brand’s current level, with and without advertising.

The table above is the typical pattern for a successful TV commercial. Weeks on-air (shaded) show an average rate of penetration increase bump, which drops to average as soon as the advertising stops. The only carry-over effects the von Gonten data report are similar bulges each purchase interval out from the advertised week, reflecting repeat purchase (not shown). There is no evidence that this is an exclusively US phenomenon: if the same analysis were done on similar products in the UK or elsewhere, we suspect a similar pattern would be shown.

Von Gonten’s analysis challenges the idea that the stimulus effects of advertising continue for weeks past the exposure. It does not `disprove’ Adstock, which is a way of modeling some of advertising’s obvious longer-term effects. A quite different thing. These longer-term effects were dramatically proved by the famous IRI experiments summarized in AdWorks One.

It is a pointless semantic exercise to worry about how much of these effects are advertising memory and how much are brand experience – probably it is mostly the latter. The simple point is those who had been exposed to the advertising continued buying at a higher rate than those who had not, even for two years after the advertising had stopped.

How then should we look at Adstock? Does it have any proper place in media scheduling? Simon Broadbent is the father of Adstock modeling and the original inventor of the concept and procedures for doing it. So to answer this question, we turn to Broadbent’s `When to Advertise’, published in 1999.

His book makes clear that the primary use of Adstock is for establishing the budget, rather than spreading the dollars. Budgets do not (or should not) come out of the air. They depend on understanding what the situation of the brand demands, which requires sophisticated analysis of past information.

Broadbent shows how one can estimate the value to the advertiser of GRP’s in each week in the year, when one considers, a) the cost of advertising in that week, b) the value of the advertising pressure obtained in that week, c) the value of continuing pressure from previous advertising (as measured in Adstock half-lives), and 3) the value new advertising is expected to deliver in future weeks (as measured in Adstock half-lives).

The half-life estimates require modeling from real data, choosing the measure of response that matters for the advertiser’s objective (often sales, but it may well be something else).

Adstock analysis helps determine how much advertising pressure is needed to achieve a goal, a) ideally and b) scaled-down to what the brand can afford. Broadbent is clear that other considerations come into play when deciding on how to schedule media weight, whether `one is enough’ or whether there is a meaningful response function, seasonal pricing of media and the constraints imposed by industry buying and selling.

If Broadbent’s explanation is thoroughly grasped, it is clear that it is totally wrong to use the idea of Adstock as an excuse for reducing exposures or increasing the gaps between them. Its value is the arguably more important one of helping us to estimate how much we need to spend to do the job.

Certainly advertising has immediate, intermediate and longer-term effects. Recency deals with the immediate, which is simple. Advertising runs and sales respond or they don’t.

Adstock attempts to deal with the intermediate, which is dicier, because it is more difficult to ascribe an effect to a cause, like sales to advertising, with confidence when they are separated in time. Too many other things have happened.

Measuring the long-term effects of advertising is even more chilling. We worry about them because advertising usually doesn’t payout unless they are included. This leads to metaphysical claims like ‘the longer-term effects of advertising include all of those things that would not have happened had the advertising not run.’ Metaphysical, because apart from repeat purchase, it is impossible to determine exactly what those things are.

It is clearer to think of the longer-term effects of advertising as two separate effects. An increase in baseline sales from repeat purchase and an increase in marginal response to advertising from greater awareness and saliency.

The increase in response to advertising (which is what the Adstock concept captures) dissipates if not reinforced by more advertising.

This is important to planning, because if what Adstock describes as a GRP’s future effect has already occurred as an increase in baseline sales and responsiveness to advertising, there is no sensible argument for substituting Adstock for fresh advertising weight. That is simply ignoring opportunity and allowing awareness, saliency and the base business to erode.

There is a disquieting timidity in the use of Adstock as a scheduling model. It seems to promote a defensive advertising goal of steady-state sales, (another phrase borrowed from the physical sciences), not an aggressive plan for increasing them.

If the advertising grows a brand then the Adstock is likely to appear strong (long half-life). This will lead a successful campaign to diffuse its advertising weight. To our minds, substantially wrong advice. Brand experience shows that when advertising is working the proper thing to do is advertise more heavily.

We think what modelers call ‘Adstock’ is a systemic effect of advertising, not a carry-over effect. An analogy may be useful here. Advertising is like vitamins. It has a transformational effect on the brand, which extends across time by making everything work a little better.

To wrap this up, to modelers Adstock is a way of calculating how much advertising pressure is needed to achieve an advertising goal. Its essential application is in budgeting. That is quite useful.

What advertising people call ‘Adstock’ is ‘the carry-over’ effects that advertising has on sales or awareness. That is vague but harmless.

The mischief is in taking the assumed carry-over effects of advertising and turning them into phantom GRP’s -- as if advertising we had scheduled in the past was just now appearing -- and then rescheduling current weight to compensate. That seems at best confusing and at worst a dangerous double counting.

[1] Recency planning acknowledges this in explaining that the single exposure that can have an effect on purchase is actually the last of a series of exposures (Ephron, 1997).

[2]This was first done by Simon Broadbent, who modeled Cadbury awareness data (from Millward-Brown tracking surveys) in the early 1980s in the UK.

[3] Another problem is the data. Most of the Adstock calculations used in media planning seem frail. Media plans deal with future advertising. How are the Ad-stock half-life values for a future campaign obtained? We are assured each cam-paign, each medium, each TV day part and commercial length has its own unique Adstock half-life often reported to decimals. How do we forecast this? Is the underlying dataset sufficiently detailed and varied to justify the different values and this kind of precision? What is the damage to the brand if we guess wrong?

Bibliography

Broadbent, Simon, ‘One Way TV Advertisements Work’, JMRS vol 23 no.3, 1979.

Broadbent, Simon, ‘Modelling with Adstock’, Journal of the Market Research Society vol 26 no.4, 1984 (pp295-312).

Broadbent, Simon. `When to Advertise’, Admap Publications, Henley-on-Thames, Oxfordshire, UK, 1999

Donius, James and Michael von Gonten. ‘Advertising Exposure and Advertising Effectiveness: New Panel-based Findings.’ Presentation at the ESOMAR Managing Media Data for Market Profit conference, Rome, November 1996

Colman, Stephen and Gordon Brown: ‘Advertising Tracking Studies and Sales Effects,’ JMRS vol 25 no.2, 1983.

Ephron, Erwin. ‘Recency Planning.’ Journal of Advertising Research, vol 37, No 4, July-August 1997

Goerlich, Bruce. ‘A Consumer’s Guide to Marketing Mix Models’, Admap, Volume 30, No 11, issue 423, December 2001

McDonald, Colin. `How Advertising Works: a review of current thinking’, The Advertising Association, London 1992

- April 22, 2002 -
Originally published in Journal of Advertising Research, July-August 2002

^ top of page ^      < back <