Blending research and real data

In Uncategorized on November 1, 2018 by sdobney

A current hot market research topic is the blending of ‘real’ data from observations or behaviour with survey and market research data in order to build pictures and patterns and from that try to forecast what customers will want into the future.

Statistical analysis of behavioural or observational data – say website data, tracking app use, e-commerce data or even just plain old-school sales and database data has turned into the massive data science industry, far larger than the older and more established market research industry.

In simple terms, the aim is to classify data in different ways, test for relationships and to use these statistical views of the data to identify and act on predictors in order to make the offer more relevant, or more likely to be chosen, by customers. For instance, identifying the characteristics of purchasers of different products, so as to better focus advertising, or linking data about the customer together in order to present a more relevant offer.

Simple targeting has always been done. Small, lower priced cars for young singles, larger but more staid and practical cars for older people with families. The difference now is the ability to mine for hidden relationships, and to mix the data sources together, and then to target at the individual level. Someone who posts about environmental issues would then get advertising more focused on the product’s environmental benefits and energy efficiency, than someone who perhaps tracks Formula One on their feed.

If you apply a time element too, then someone looking at houses or new accommodation now, might be a prime target for furniture sales a little down the line. Or, at the point that children reach 18, suddenly there are a whole set of life changes that impact the 18-year-old and their parents. Both get a mass of financial decisions, changes to lifestyle and outlook and both get more independence.

Understanding and mining the data therefore helps businesses identify new potential customers and customers at decision points, where the company can best offer new solutions, new services or new products.

However, the challenge with behavioural and correlational data, such as website data, e-commerce data, or just old school sales and database data is that it is typically backwards looking, and correlation can’t always be explained. It tells you want people did in the marketplace at some time in the past, but it can’t tell you why, and it can’t necessarily help with views about the future, where that future is different from now – for instance how the market will react to a new product.

For a practical example, an observation from data might show that car purchase by residents of Barcelona are typically smaller than those from outside the city. However, this doesn’t necessarily explain that part of the reason might be because Barcelona has large amounts of underground parking which makes larger cars more of a liability than small ones. The ‘why’ explains the data.

Similarly, if a company is planning to introduce a new car sharing service, say, the backwards looking data will show the relationships between trips and travel and individuals, but as it’s looking backwards, it can’t quite project forwards to the new service without going and asking – the data has to be created from scratch.

This is where blending of real data and survey data becomes massively useful. Not only can existing patterns be explained, and from those explanations will come new ideas about how to build the links, but also views of the future can be melded with actual older data to estimate not just the impact of the change, but also where and how that change will affect existing data patterns (and so perhaps lead to competitor responses and further change).

Now ‘blending’ is something of a new catchword, but actually companies have been blending research and real data for a long time. In most B2B studies existing customer data is often used to enhance the market research data, or used as keys for modelling research findings back to the database for tasks such as customer segmentation. In consumer markets, surveys and real data can be ‘fused’ statistically to build propensity models – a little like credit scoring.

The interesting aspect now though, is the ability to do this in near real time. So we might track a (willing) customer through their use of an e-commerce facility and then follow up immediately with questions about their behaviour and what they looked for. One interesting finding is that it seems customers often remembered how they looked – the journey – more than what they actually bought.

Once this is overlaid with the ability to conduct sample-size experiments where features and offers are modified to selected customer groups, who are then followed up with survey research to refine and iterate the offer.

This blending of research, behaviour and forecasting is the next frontier for customer insights.


Rethinking Questionnaires – fly menus

In Uncategorized on November 2, 2016 by sdobney

The second of our series on Rethinking Questionnaires is now at LinkedIn. As mentioned, our focus is on re-imagining what and how a questionnaire can be. Many of the research tools we use today are online replicas of surveys that could have been carried out 70-80 years ago; in reality not that much has changed. With the availability of computer-based interviewing via web-surveys the lack of innovation is quite surprising.

Our second demonstration of fly-menus which are used for drill-down and non-linear approaches to questionnaires can be found here:

Fly-menus are a structural approach to a problem that the linearity of questions in a questionnaire (from A to Z) is determined by the researcher and not down to the respondent saying what they are interested in, or what they would give their time to answer. A fly-menu approach says to the respondent, here are the topics, you choose which topics you want to tell us about, and in what order. This is why we also refer to this as a non-linear questionnaire structure.

It actually breaks quite a big taboo. In surveys we are fully aware that there are order-effects. That is the order in which you ask questions can affect the responses that we get. For some areas like prompting where we prefer to find out what someone knows or thinks spontaneously or semi-spontaneously before introducing the ideas, do still rely on question order. But for approaches such as customer satisfaction research, it should be the respondent/customer who determines the order of questions and prioritises what they want to order.

More specifically, we know that more attention is applied to earlier items when answering questions like scale questions. Though this illustrates a weakness of scale-based approaches in that simple re-ordering changes the answers, it also suggests that we should work with the respondent on the areas that they want to give attention to because then we will have better answers.

The original LinkedIn article on fly-menus is here.


Rethinking questionnaires – hot-cold scales

In Uncategorized on November 2, 2016 by sdobney

Over at LinkedIn we are have included some posts and demonstrations on Rethinking Questionnaires with new ideas for question design, questionnaire structure, blending surveys with websites and social media. A demonstration can be found here:

The first looks at hot-cold scales. These are use buttons and images to allow respondents to say what they like or don’t like, but crucially, we don’t force a response. Because of our work looking at choices and choice making, we feel that forcing tends to mix weakly-held with strongly-held views. Choices on the other hand enable a respondent to review a large group of items rapidly and naturally, much as they might do with an on-line or actual paper catalogue, only focusing on the items that are of interest.

The original LinkedIn article is here.


MR and Customer centricity

In Uncategorized on May 5, 2016 by sdobney

Edward Appleton in his entertaining and provocative market research blog asks “why, given that Market Research and Marketing been preaching customer-focus for decades, so often one experiences lousy customer experience?”

Read More »


How big is Big Data?

In Uncategorized on November 24, 2014 by sdobney

In a seminar on Big Data in Barcelona this week, a consultant from Ernst and Young (EY) Consulting suprised me by suggesting that 43% of Big Data projects involved less than 10,000 records running up to 71% of Big Data projects that used less than 1 million records. This was a surprise as my assumption of Big Data was that it was really N-million record type projects.

Read More »


Orthogonal designs in conjoint analysis

In Uncategorized on May 12, 2014 by sdobney

Conjoint analysis (or discrete choice estimation/stated preference research) broadly has four main components. The attributes and levels that make up the product or service that we want to test, a statistical design to choose combinations of attributes and levels in order to convert them into product profiles that reflect the decision space, a choice method – usually direct choice but it could include an estimation of volume (eg number of prescriptions for medical subjects), a ranking, a multiple selection eg of items into a consideration set, or a rating in a more old fashioned conjoint, a method of analysis – normally Hierarchical Bayes or MLE and a modelling. Of these, the element that is most troublesome is the statistical design. With large numbers of attributes and levels, it is impossible to test all combinations, so we have to choose a subset (a fractional factorial design). Read More »


Comparing Open Office, Libre Office and Microsoft Powerpoint for market research charts

In Uncategorized on March 6, 2014 by sdobney Tagged: , , ,

Charting is one of the main reporting methods for quantitative research, and most companies use Powerpoint. With Open Office and Libre Office now reaching version 4.0+ we’re discovering that Open Office is now getting good enough for productive chart production for market research. Read More »