Multimedia and multi-channel effects
Paul Baker joined Ohal in 1975, when it was a small company exploring - and also leading - in the area of how advertising works on sales. He became a director in 1977 and managing director in 1994. Ohal is now the largest econometric group in Europe with a significant office in the US.
Admap - October 2004, Issue 454
The simple answer to the above question is 'Yes'. The reason is culture.
I will discuss each of the key elements - 'measure', 'disentangle',
'multi-media', 'multi-channel', 'brands' and 'ten years' time' - in turn,
albeit in a different order.
TEN YEARS' TIME
I have been modelling the effects of media on sales for 30 years. Thus in the context of 'ten years' time' I have already experienced three such time spans. What can be learnt? Three things immediately come to the fore.
- Ten years seems a long time, but goes quickly.
- It seems that not much changes on a year-by-year basis but, looking back, a lot does.
- Most importantly, change is accelerating.
Consider fmcg (packaged goods) markets for 1974-1984. In any category the launch of a new brand or variant would take years to develop, be heavily researched, and then launched with extreme caution. The results were that markets (as defined by Nielsen) added an extra line about every two years. And mentioning Nielsen, it did not change either. The data were bi-monthly, presented six weeks later.
We move on to 1984-1994. More aggressive marketing was being carried out into extending new products and brand variants, and whereas 1984 was not much different to 1974, 1994 was significantly different to 1984. Nielsen also got competition, and data became available on a monthly basis.
However, the statement above about acceleration has become very apparent in the 1994-2004 segment. Again referring to fmcg, we had two major, but related, changes. The cause was the technology of barcodes. First, this enabled retailers to be far more aggressive and innovative with promotions. Second, there was an explosion in data. Most markets have at least 1,000-1,500 lines in a Nielsen or IRI database and 5,000 is not excessive. In 1974 there were only 12 lines in the toilet tissue market!
There is a simple fact about measurement. It is that if an activity - say advertising - makes little or no difference to sales, then it cannot be measured. Thus when we report back that on an activity (a local radio campaign) no measurement has been made, it is often construed as a 'failure' to measure, rather than that any effects were so small they were indistinguishable from random data movements. The converse is also true. If sales change by an amount, which can be regarded as outside the statistical 'norm', then there should be a reason. Disentanglement is covered later, but in principle it should be possible to identify the factor(s) that caused the change in sales. This is a fundamental part of good modelling. It is not just a case of putting the data into the computer and finding the best fit. There is a lot of detective work in examining the data and formulating hypotheses about how the market works.
This is very important in the context of this article. There is an implication that communications will become so complex that no single medium (or combination) will have a large effect.
The previous section talked about change over time and, in particular, referred to data and how they have changed. What has actually happened is that they have improved. Not only for fmcg. Retailers know daily sales of each store (indeed hourly for each product, if required). Credit card suppliers know all about patterns of expenditure across product fields. Services such as Nectar can cross-reference purchasers' behaviour across retailers and use it to maximise return for their partners.
Yes, they can do all of the above, and yes, they do a lot of it, but what has happened in the past few years is a change in culture. Marketers are not scared of finding out the return on investment (ROI) of their activities, because they want to understand how to improve it. In the next ten years this desire to know the ROI is going to increase, and of that there is absolutely no doubt (it is the key certainty of 2014).
Yes. Markets, media, data will all become more complex, but the desire to know will overcome these difficulties. How will they do this?
In most cases, the increasing quality of the data sources will simply overcome the difficulties. If not, top management - it is top management who want accountability - will demand that data are made available to carry out ROI measures. It will not always be possible, but there will be more than enough to make modelling a very buoyant industry.
Disentanglement is the principle of econometric modelling. I have no intention of going into detail here, other than to describe what it does.
Figure 1 shows that 'sales' can be determined simultaneously by a combination of factors. Factors here include those that, in 2004, might be expected to influence sales of an fmcg, but other diagrams could be drawn up for other types of market.
There is no doubt that demand for econometrics (or disentanglement) has increased recently. Over the years three factors have determined the ability to carry it out:
This has been available for well over 50 years. Theory is always being refined but, in reality, it is practicality that drives development, rather than new statistical tests.
This was a real problem in the early days, but by the 1990s it ceased to be so relevant. Inevitably, technical requirements have put pressure on the required capacity, but the barrier has gone.
There is absolutely no doubt that the amount of data currently available is greater than it was ten or twenty years ago, and there will be even more in ten years' time. In most cases data will be better.
Currently, and in the recent past, there have been brands and markets that are suitable for modelling, and those that are not. An implication in the lead-in to this article is that a number of these markets may become unmodellable - from a media point of view. Remember that the function of this article is to explore the complexities of media and channel changes upon modelling. Markets and brands will change and add their own complexities. Promotional changes for fmcg, internet shopping for retail, legislation for pharmaceuticals and alcohol will all pose their problems to the skilled practitioner, but these problems are not directly media related. I would argue that, if anything, the opposite will be true. Consider the following key categories (chosen on the basis that they represent over 90% of Ohal business during the past 20 years).
Modelling will become more complex, but the potential for results will become greater. At the moment, we can provide answers on, say, Kellogg's Cornflakes, for ROI both by medium and interactive effects across media and for other activities, such as promotions. We can even take the results further by finding out how many exposures are required to trigger a sale. However, these models cannot identify who is responding to the media. With the power of the card databases of Tesco and Sainsbury's, it will be possible to answer these questions.
Large retailers have comprehensive databases for each of their outlets - size, turnover, local competition, transactions, sales by category, and so on. As a result, they can target very specific media by outlet or groups of outlets. Posters and DM specifically by outlet catchment area, and even TV, radio and press have regional differences. Shopping patterns will change, but the whole retail industry will not.
This has always been a difficult area to model, and will continue to be so. There are two reasons for this, the first being statistical, in that the number of individual purchases is relatively small. (The number of packets of Kellogg's Cornflakes purchased in one week is significantly higher than the number of Ford Focuses in a year.) The second is partly a red herring - that the purchase decision is said to be very complex. Of course it is, but consider the sequence shown in Figure 2.
It is equally possible to fit Andrex/toilet tissue or Ford Focus/medium cars into this sequence. It is also possible to see how communication can affect numbers 2 and 5 in this process. The difference is time. The former has a timescale of about one week, whereas the latter is probably one to three months. Given the statistical issue above and the fact that car manufacturers spend very heavily behind brands, precise measures are more difficult. In most cases the ROI of total media campaigns can be identified and accurate media splits can be obtained. Good modelling can also measure factors such as short term effective frequency, although saturation can be an issue.
This covers a range of products from savings, mortgages, current accounts, credit cards and loans. When good data are available, most of these areas are good for modelling. The caveat is that data are usually kept for financial rather than marketing reasons. This is for statements, interest paid, charges collected, etc., rather than any aggregates for analysis. Media and communication changes will take place in this area.
This category is very similar to fmcg, but has more issues relating to legislation. Given trends in the US, there is far more likely to be heavy media spend behind prescription drugs.
Off-trade (take home) is very similar to fmcg, but on-trade data have always been patchy. As electronic tills have become more widespread, data are improving, and will continue to improve. It is difficult to forecast movements in legislation, but they could lead to major changes in media use.
This is similar to finance. There are a number of product ranges and data are often retained for reasons other than marketing. It is probable that data will improve over a period of time, but equally the market will probably be different - based on the changes over the past ten years! The issues of disentanglement of media will be small in relation to those of disentangling the market.
Entertainment and travel
Could be an area where modelling will improve, but identifying the effects of different media may prove more difficult. For travel, there will be greater knowledge of who, where and when purchases were made, as well as the price, but pinning down the stimulus may prove more difficult, as more internet purchases take place.
These are the main areas where consumer modelling involving media takes place. We are finding that there is interest growing in other categories, possibly as more traditionally based marketers are employed by other companies. However, when looking ten years ahead, it is sensible to consider first those areas that already represent the core business.
Multimedia and multi-channel
The distinction here is that multimedia is where a brand uses a number of distinctive media. For example, TV, radio, outdoor, press, cinema and DM. Multichannel means what might be described as other forms of communication, such as sponsorship, PR and ambient.
We are often asked about the effects of a multimedia campaign that has been run for a brand. When we examine the data more closely we find that 95% has still been spent on TV, and 5% diverted to a secondary medium. Unless the latter is enormously effective, it is too small to measure. There are definite trends, particularly within fmcg, for brands to be exploring more multimedia campaigns. This will make measures easier, rather than more difficult. There will still be the same market problems in the categories covered earlier, but these are not specific media problems. We can continue to hope that media data will improve, but while it is fair to say that over the past ten years, development has been 'slow', the understanding of data has improved.
In the section on measurement, I said that if something made a difference it could be measured. I deliberately left out a caveat, which is that if it does not vary then it cannot be measured. An obvious (non-media) example is that if price has remained constant, then it is speculation to ask what would happen if price were to change. PR and some forms of sponsorship fall into this category. The former tends to be continuing underlying 'noise', which almost certainly has a positive effect, but which is part of the base. Perversely, the most regular measures we get are when significant negative PR takes place, such as toxins in a food product. In terms of sponsorship, it could well be measured for weather forecasts sponsored by a hay fever product (specifically time-related), but not for a five-year contract on the Six Nations rugby tournament. Ambient falls into both camps on the issue of measurability, in that it is often a relatively small and opportunistic expenditure or an underlying continuous campaign.
Ten years ago an employee handed in his notice to me saying that he believed that there was no future in econometrics. He was palpably wrong.
When I was given the title of this article, I was not quite sure what the conclusion would be. However, a few key points have emerged.
- Most problems of measurement will be market rather than media-related.
- It if works it can be measured.
- Data will improve.
- Most importantly, the culture will demand it.
Thus the answer to the question posed at the beginning of this article is 'Yes'.
Download the full report