Optimising Marketing Mix
The initial calculation reports on marketing effectiveness, but it doesn’t tell you why these results occurred.
To understand this, you need to conduct a series of tests using the data you collected.
Let’s look again at how your marketing dollars are converted into sales, leads, and new customers:
Response is the measure of audience who took a voluntary step to learn about your product. In SEM, this will be the number of non-bounced visits, in Youtube – number of views over 5 the mandatory 5 sec or number of visits associated with watching your ads, etc. While there is no universal measure, it is possible to construct channel-specific measures to determine how well people respond.
For the tests of audience-message fit, you should focus solely on conversion from impression to response, ignoring costs and subsequent conversion to sales. When you have different marketing messages, it is the one that generates higher response per impression that fits better with the audience.
Response rate should be used as a measurement device and not as a target to maximise. Clickbait ads might generate higher response but will not ultimately translate into higher sales. The best use of this metric is to test 2 or more marketing ideas with all other things (formats, presence of CTA) being equal. You should also exercise commonsense and check if the ads attract the right audience for your brand.
There are multiple audience segments within each channel, and analysing response allows you to drill-down into those segments to determine best fit. Say you are selling bicycles – do women respond better to “woman on a bike” ad or “family on bikes”, or do they not care at all? This is the kind of questions you can answer with this test.
InContekst software makes this analysis easy as you would already have all the necessary data in the system. You will need to select the measure for response, measure for impressions for the appropriate channel and run them through “the algorithm” to get response per impression. To answer the question above about women’s preference in bike marketing messages, you would need to select placements that target women. In the system, there are functions to quickly tag placements, and these tags later can be used to separate such placements into a separate marketing input.
An advertising campaign usually includes ads of different formats with the same message. There are long and short TV and radio commercials, print and digital ads of various sizes, etc.
With longer / larger ads being more expensive, you would want to use them less in the ad mix. However, the marketing message might not fully fit into shorter / smaller ads, so they end up working more as a brand reminder.
How little full-format ads can you afford to have in a mix on the same channel?
To answer that question, you would need to use data with varying levels of different format ads in the mix. If your business is substantially online, you can use non-bounced visits to the website as a target marketing outcome, because if the prospect took the time to find your site and look around, it really doesn’t matter how long was the ad they saw.
Within InContekst system, enter two separate marketing inputs – spends for the larger and the smaller format and run the model calculator. It will report how much response does either format generate per dollar spent – and the share of the lowest-cost one should be increased.
You might ask: can’t I just run a test and compare results at the face value? It is really difficult to conduct a clean test in real life on one parameter with all other parameters being equal. The model controls for the influence of those parameters distilling the effect of the parameter you study.
An advertising campaign would include a mix of premium and low-cost placements / programs / context. The higher quality context is more expensive, but it rubs off on the brand, making it more trustworthy.
What is the optimal percentage of premium placements?
To answer this, you need to conduct a test similar to the format test above but tagging the placements as premium vs low-cost rather than by size / duration. You can use several categories of premium-ness, e.g., top rating program, premium position within a commercial break, program sponsorship. The number of categories is limited by the amount of data you have.
InContekst helps flagging the placements on the basis of costs, TV or radio daypart, or by other criteria.
The rest of the test and the interpretation of the results is similar to the format test above.
All channels have limited audience and it is quite possible to run out of your target audience on a channel.
If you have an initial estimation of how many times your message need to be shown to an audience member in order to get noticed, watch for the average frequency reported by the channel. Some channels report reach and impressions instead; average frequency = impressions / reach.
If you don't have the desired reach in mind, run a series of analyses that track response as a function of impressions, starting with say a week, and gradually extending the analysis period. Calculate marginal effectiveness as response of the current test period less the previous one. When marginal response declines to, say, 20% of the original value, this might indicate that the campaign indeed ran out of audience. Note the average frequency reported by the channel for that test period - it would be the optimal frequency to stop the subsequent campaigns on that channel.