The Retail Measurement Service (RMS) is a cornerstone of the Nielsen business. It’s used by marketers around the world to understand the retail landscape and shoppers’ detailed purchasing behavior. Currently, the sample includes 103 countries and more than 1.2 million stores that either receive monthly audits by Nielsen representatives or share their point-of-sale data electronically with Nielsen. A separate Retail Establishment Survey (RES) is executed to compile detailed information on the universe of retail stores in each market. This is necessary to support a representative and projectable sample, as many countries—particularly in developing and emerging markets—do not have reliable universe information.
The RES process involves an even larger number of data collection visits: more than 8.5 million annually. The data is collected for five to nine months, with a universe estimate update happening once a year in most of the countries. To generate a more accurate and consistent universe, Nielsen has been transitioning this process to one conducted on more of an ongoing basis. This new process is known as Rolling RES (RRES) and has been deployed in 54 markets already. It involves the collection of information throughout the year and updates to the universe every four months, or three times a year.
So why is this important? Rolling data collection procedures like the RRES are a marked improvement over periodic collection procedures for every business that deals with large-scale enumeration needs. They make it possible to develop samples that are more aligned with the realities of the marketplace. Their implementation, however, puts pressure on operational standards. First, it challenges the design of the sample; in the case of the Nielsen RRES, for example, the optimal sample has to be divided into three properly balanced subsamples – not a trivial task. It also challenges estimation methodologies (e.g., researchers have to account for new government policies that affect one of the subsamples but not the others). And it requires much faster rollouts, to make estimates available quickly after fieldwork and reap the full benefits of the system.
Those new constraints come at a cost. The designs and processes that underpin traditional enumeration projects are generally very manual and typically lack the flexibility to accommodate the type of testing and validation necessary to bring a rolling process to life. For instance, splitting a country’s sample into three robust subsamples may take a data specialist a full month to execute adequately. To support the deployment of a rolling solution at scale, it is necessary to standardize and automate the full methodology. A comprehensive production platform needs to be developed across all steps and processes to bring all the data together (and prevent the waste of time involved in transmitting and reformatting that data on an ongoing basis) and allow experts to experiment with it to test the impact of a variety of factors (e.g., the number of strata) on end results.
While successfully switching from a manual process (involving several disconnected platforms) to an automated process (leveraging efficient centralized databases) is not easy, it’s the right thing to do from a data quality standpoint. Standardization efforts are always challenging, but they open the door to new core competencies and create opportunities for data science to make a substantial contribution to a business’s bottom line. Nielsen is currently running pilots to automate its RRES process in countries across Europe and Asia, and the results are already extremely encouraging.