He heard ‘spots’ and got his hopes up. Sorry, Barry. But you still made it into the blog, eh?
We don’t do spot-matching at MetaMetrics, but we have been asked to in the past and are familiar with the insights it gives you – and the challenges involved in making it work. Here’s our take on it.
What is it?
A technique for attributing response (sales, visits to a web sites etc) to a particular spot. Used predominantly in TV, where second-level spot data is available (in the UK), and where the ad and the product are direct-response friendly, i.e. not baked beans.
See it, buy it – ‘second-screening’ has made response very quick to direct response advertising
How does it work?
In principle, it’s very simple. It looks at the response data, let’s say sales, within a set window, e.g. 5 minutes, of a spot. A spike in sales occurring in this window is deemed to have been caused by that spot (hence what we said about it being a direct response tool).
Most spot-matching solutions have an algorithm that sets a baseline for response to ascertain what the response would have been in the absence of the spot. This could be done, for example, by identifying the level of sales at the same time of day, on the same day of the week but when there were no spots on air.
Subtracting the baseline from the observed spike gives the number of incremental sales associated with that spot.
In practice, the task is compounded by the vast number of TV spots, and so spot-matching systems tend to be fairly automated ‘black-boxes’ which use a lot of computing power to churn through the volumes of data involved.
Is it as easy as that?
Well, no. We’re glossing over many problems. Some of the more common ones are as follows.
Zero-rated spots
These are real spots which are rated as zero impacts using the BARB survey data which is standard in the UK. This happens because BARB data is survey-based and niche channels with few viewers may not be represented on BARB. Of course, it’s a real spot – an ad does actually go out, and is seen by real, sentient people, some of whom will respond to it, causing a spike in sales. Spot-matching systems therefore have to be able to deal with zero-rated spots, e.g. by imputation (basically guesstimating the number of impacts for a zero-rated spot).
Overlapping spots
Frequently, more than one spot will go out within the same response window. The problem then is to associate the right amount of sales with each spot. Again, providers will have developed their own more-or-less sophisticated ways to deal
Consolidation
In the UK, TV data takes about 9 days to consolidate, meaning that the latest spot-matching results (i.e. the amount of sales attributed to each spot) are not available for that period of time. If you’re in a hurry, that might be a problem.
Where does the data come from?
In the UK, TV spot data (impacts) is usually taken automatically via an API feed from industry systems. This data comprises a list of spots, each with spot time, estimated impacts for each (from the BARB survey) and descriptive information, detailing, for example, the channel on which the spot aired, the programme it appeared in, programme genre, film code, the position in break and so on. This descriptive data allows you to cut the data in various ways and look for patterns among similar types of spots e.g. spots on a certain channel which ‘work’ better.
Example data for a single spot (not real data)
The response data comes from you. You will either send a time-stamped data file of your chosen KPI to your spot-matching provider, or, much better, set up an automated feed. Note that if your chosen KPI is sales as is usually the case, you will also need to decide how to deal with the time lag between first (e.g. a web visit) and final response (completion of a sale).
Does it work with media other than TV?
Basically, no. It suits media for which accurate, second-level spot data is available. That essentially describes TV. Radio is a possibility in principle, but in practice most spot-matching tends to be done in relation to TV, probably because of the sums of money involved.
The REAL challenge
Using the data.
Spot matching is fairly well developed in most advanced media markets, and, whilst a certain amount of pain is involved in setting up the data feeds, once it’s up and running it will produce a steady stream of data, with minimal ongoing effort.
The challenge now is to analyse this, more or less continually, to identify patterns, and to act on them by adjusting the TV mix. On one hand, that involves a lot of time and effort for a competent data analyst. On the other, it involves the contractual and organisational flexibility to be able to alter your TV mix. If you get your TV from a big agency, they will probably supply it from their pool, their contractually negotiated ‘chunk’ of TV air time for all their clients. This may limit the degree of flexibility you have in practice to make in-flight adjustments, so to speak, to your TV mix.
Does it work with offline sales?
No. And even if you were to respond to an advert immediately, the time it takes you to get in your car and visit a shop would put you well outside the response window.
How does it take into account other media and non-media drivers of sales?
It doesn’t. You have been warned. For this, other techniques (econometric modelling!) are needed. For this reason, spot-matching tends to be used to optimise within TV, rather than between media channels.
The quality paradox
One other thing to be aware of – low-quality programming can have a better response than high-quality programming. The reason is, if a programme is good enough to capture your full attention, then you are less likely to pick up your tablet and Google the thing you’ve just seen advertised.
How do I go about setting up spot matching?
Choose a spot-matching provider (there are many in the UK), agree a KPI with them, have their technical people talk to your technical people, press ‘go’.
But before you do any of this:
- Make sure you talk to the person who actually buys your TV (i.e. not the account exec or the head of investment) to find out how much flexibility you have to vary your laydown, and the lead times involved. If your spot list is pretty much dictated to you by your agency, then you will gain little from knowing which spots work best in that list.
- Make sure you have a competent analyst lined up to do the heavy-lifting, and on an ongoing basis – really, someone needs to ‘own’ the task of deriving insights from spot-matching data
Are all spot matching systems equal?
No. All systems are model-based, and are only as good as the underlying models, some of which may perform better in certain circumstances than in others. It is almost pointless, therefore, to compare the results from different spot-matching systems. They will give different answers and you won’t know which one is correct. Choose a provider who has clearly grappled with the technical issues involved, and who is responsive to your needs.
As we said at the start we don’t do spot matching, but we will happily point you in the direction of someone does. Feel free to get in touch.
Philip Gaudoin