A lot of metrics used to measure digital are not output metrics: Dolly Jha
Nielsen India last week launched Custom Mix Modeling (CMM) in India. The ROI measurement solution enables FMCG advertisers to evaluate the impact of all media platforms on traditional trade retail sales.
The weekly retail panel has been launched in 8 major metros for 100+ brands across 16 FMCG categories. Nielsen uses their globally accepted Marketing Mix Modeling (MMM) technique on the weekly offtake trends to provide sales ROI measurement across all traditional and digital media.
With the CMM solution, apart from the impact of media platforms on sales, Nielsen will be able to provide the industry with benchmarks for digital ROI insights. Furthermore, with this solution Nielsen can set up quick tests that evaluate sales impacts of different creatives, ad formats, and placement decisions at a sales ROI level.
According to Dolly Jha, Head - Media, Nielsen South Asia, “This solution that we have created can help FMCG marketers and advertisers with insights at multiple levels on a medium that is today almost 20 per cent of the total media portfolio. This will directly aid in media planning, in turn helping advertisers gain overall efficiencies on media investments.”
In conversation with Adgully, Dolly Jha, Head - Media, Nielsen South Asia, sheds further light on the workings of the CMM technique, the state of measuring ROI in India, challenges of measuring digital and more.
How will Custom Mix Modelling measure ROI?
The media mix is undergoing an evolution, with marketers upping their expense on digital. The question that arises is when I’m spending money across India, how well am I spending it? Which essentially means that the kind of returns that I am getting by investing in various media, how good are those returns? And specifically when it comes to digital, can you give me some more granularity? With different publishers and different placements, what is the kind of yield that they return? Typically, a modelling exercise is done where you can look at your output, that is, your sales data or it could brand health data, whatever you want to look at. We are talking about ROI, which is why the output data is the sales data and the input data can have different variables (promo spends, price, any other marketing lead, etc.) For this modelling you require 30 to 35 data points, which is not a problem as there is media data that can be used weekly or monthly. But for the FMCG category, since we are a very traditional market, the sales data is available only at a monthly level.
So, what we really did was that we set up a custom panel of retail outlets in 8 metros, where we saw weekly sales trend in various categories. Nielson runs monthly retail audits, where we have monthly sales optics coming in. So, we have retail tandem and we use this data for modelling. And therefore, the problem which I had was solved as we could use this data for modelling and to the advertisers I could answer what is the ROI that they are getting on scale on TV, print, digital, publishers within digital, placement A Vs placement B. Thus, this provided a very objective and hard matrix on a scientific basis.
What is the reason for starting this off in India?
India is a traditional trade heavy market. The phenomena of trade is the heaviest in India, whereas in the more developed countries it is more digital, so it is easier that way there. Therefore, this kind of solution is required in India.
How the mix model typically works is that they find a change in the pattern of sales and see to it that it advances. Neilson does it globally for hundreds of advertisers. So, the methodology isn’t new, but we made it suitable for the Indian context.
How big is the sample size/ panel?
This panel runs in 8 metros and it covers 15 categories where the questions are coming from. The panel has around 1,250 retail outlets. Along with the massive retail panel that we have, this is what we use.
Where are you getting the 30 to 35 data points from?
Input data is available. All advertisers have their media agencies, so if you look at media data, then they are able to give us that data which is available at a weekly level.
Neilson normally also measures data on a monthly basis. So, based on the monthly data output that you get, if you have to build a model where you look at output of sales and media and see which caused what, the model required at least 30 data points, which means I have to look at, at least one year’s data for 30-35 points. With weekly data, it shortens that cycle so that I can go back 6 months and in 6 months you have 24 points and if you go back one more month you will get 38 points. It can also be built much faster so that you can get insights on the digital platforms.
While the building period is separate, it takes a month to model the data. This allows Neilson to collect the data and yield the model weekly.
Are you comparing this with the retail index that you are comparing per year?
We are not comparing, we are trying to establish an O/I relationship on the output side. We have variables that we consider; we are looking at sales as output variable for ROI and spends across variable on input side.
How many brands and publishers do you have on board?
Neilson does not talk about their brands, but Mondelez was one of the first brands to deploy. As far as publishers are concerned, all publishers are measured on that platform, so you read across all publishers.
With such a huge number of digital platforms, how can you get the sense of it because one platform has likes, shares, comments, while another platform has different method of measuring it? How do you get a hold of these engagements and how are you able to compare them? What is your goal metric?
If you look at how measurement happens, there are a lot of metrics that people use to measure digital that are not really output metrics. Likes and shares do not necessarily mean that it had a brand impact on the consumer. Similarly, they might like and share, but may have not bought it at the store. So, the idea is that you look at the end metrics and outcome which is a brand impact or a sales impact and use all of this as input and then place it in the model and see if sales have moved. That is how the model works. The metric is really the change in sales or the impact in sales.
So, if I’m able to deploy a campaign on social media, I am able to find out from which social media you are able to get the most metrics?
You will be able to compare not just across social media but also between publishers. Then there is also TV, print, OOH, etc. So, you can input all these data and get to see which drove more sales. Thus, it is not just digital. In the monthly model, you might not get only digital reads. If you go back to, say, one or two years, the digital spending may not have been that much and not be able to get a read specifically with the publisher. So that level of granularity is not introduced.
Could you tell us about the Marketing Mix Modeling (MMM) technique?
It is about contextualising it. We are rising on the MMM foundation. If you have the recency of spends, then its original avatar will not work. You need a certain amount of time to run that. So, you will need a suitable adaptation of that to the context with which the advertisers are faced with today. An advertiser might want to know how their campaigns had fared in the last 6 months. That is where the adaptation comes in, it is a custom mixed model.
There is customisation of the data collection across 1,250 retail outlets in 8 metros.