Category Archives: Business analytics

How is multivariate data analysis used in marketing?

Published / by Simon Foster

‘Multivariate’ means ‘many variables’ and in the context of marketing it usually means analysing multiple variables from customer records to get a deeper understanding of the customer base. This increased understanding of customer behaviour permits the development of customised offers, relevant creative messaging and more accurate media targeting – particularly with techniques like email and behavioural targeting. Very strong offer targeting will significantly increase your response and sales conversion rates.  Any company that has a database of more than around 5,000 records should be using multivariate data analysis to analyse customer data and improve marketing performance.

The most common forms of multivariate analysis in marketing are cluster analysis and hierarchical analysis. Cluster analysis uses statistical techniques to allocate customers into segments based on how similar, or dissimilar, they are to each other. So for example, if you had 10,000 customers and you were clustering by income and home ownership, you would be able to define groups of customers with similar levels of income and home ownership status, or those with high income and low home ownership status, or those with low income and high home ownership status. The number of clusters generated depends on how you set up your cluster analysis and of course, what patterns actually lie within your data. You can set up your analysis to produce either a large or small number of clusters, but most marketers can’t practically service more than about fifteen clusters.

Hierarchical analysis breaks customers down into sub-sets of the whole customer base. Results of hierarchical analysis are often shown as dendograms or tree diagrams. In a tree diagram, all customers belong to the ‘root’ and segments of the customer base are called ‘nodes’, nodes are connnected to the tree by ‘branches’.  So for example, all customers can be divided into males and females. Then the males and the females can be divided by age, and then by income and then by spend. You are then able to see what proportion of the whole base is composed of customers with certain characteristics.  Here are some examples of customer segments defined using hierarchical analysis:

  1. Spend more than £250 per year and are aged 18-34 and female and do not have children
  2. Spend more than £500 per year and are aged 25-44 and male and do not have children and earn between £20,000 and £30,000 and have a mortgage
  3. Spend more than £1000 per year and are aged 35-54 and have children and have a mortgage and live in the South East

Whichever technique you use, it is likely that you will see a small number of segments account for disproportionally large amounts of sales revenue or sales potential. When you have identified these segments you can leverage what you know to develop tailored offers, messages and targeting. Over and above this you can identify customers who have the characteristics of high performance segment membership, but are not spending at the rate they could be. You can use this information to target your marketing messages to the sales prospects with the highest untapped potential.

What is predictive modelling in marketing?

Published / by Simon Foster

Predictive modelling is a term with many applications in statistics but in database marketing it is a technique used to identify customers or prospects who, given their demographic characteristics or past purchase behaviour, are highly likely to purchase a given product. In this context, ‘predictive’ does not simply mean predicting the future; it means identifying the quantitative factors that can be used to predict buyer behaviour. Predictive modelling is a powerful data analysis technique that can be used to target email and direct mail activity, and to some degree behavioural targeting in online media.

Here’s an example: Let say you sell 10 products. It may be the case that all purchasers of product 8 are: 1) in a certain geodemographic group, 2) married with more than one child and 3) own more than one car. All these factors can be analysed and combined to predict the likelihood of any consumer in your database buying product 8. Usually this combined measure is referred to as a ’score’ i.e. a figure which represents the presence or combination of certain variables in the consumer record. Once you have developed your scoring model you can rank all customers by their score. When you’ve stripped out those who have already bought product 8, you are left with a set of high potential prospects.

Predictive modelling can also be undertaken based on transactional information about past purchases. Going back to the 10 products, it may be the case that 80% of people who buy product 7 have previously bought products 2, 5 and 6 and in that order. So we can say that people who have bought products 2, 5 and 6 (in that order) but who have not yet purchased product 7, are much more likely to buy product 7 than everyone in your database. Again a score is attached to these behaviours and that score can be used to rank your prospects in terms of untapped sales potential.

Of course as well as predicting purchase behaviour, these techniques can be used to predict risk. In credit assessment for example, it may be the case that those customers who have certain demographic characteristics combined with a certain type of past purchase behaviour are highly likely to default on a credit agreement. This is sometimes referred to as credit scoring. If you are rejected for credit at a bank or in a shop it will be because your data has been analysed and your credit risk score is deemed too high or low to meet the criteria of the lender.

These predictions can help you target your communications very efficiently and also help you control commercial risk in customer behaviour. What’s interesting about these techniques is that they help both the marketing department and the finance department. Marketing delivers customers who are both highly likely to convert to sales or high lifetime value whilst at the same time, producing customers who are less likely to cause problems for the finance department. Overall, this means that the resources of the business are being better utilised.

Data planning and market research – mind the gap

Published / by Simon Foster

I once attended a research debrief to report the results of a survey into the communication effects of a direct mail campaign. The survey asked if the target group had received the direct mail piece and what they thought of it. The survey results were not good. According to the research, hardly any of the respondents could recall seeing the DM pack and even fewer claimed to have responded. There was disappointment; it was a big mailing and a strong offer, surely someone must have seen it and been motivated to respond. But all was not lost. In reality, away from the results of the survey, the campaign had in fact been very successful. I knew that the campaign was in the process of beating all its response, conversion and sign-up targets.  From a hard data point of view this campaign was on track to become one of the most successful DM campaigns ever run by the client.

So why was the recall in the research so low and the actual response so high? I can think of three explanations:

First, we were targeting a large group of the population. It was possible that even though the hard data results were good, we were drawing our DM response from portions of the population that simply hadn’t been included in the sample.   If we had a 25% response then that was a record-breaker from a DM planning point of view, but it still meant that the vast majority of the target – 75% – hadn’t responded. Those that had engaged with the mailing were far more likely to recall it than those who had not. So if our sample happened to comprise of 85% or 90% of those who did not responsd, then the recall results would be much lower than the response actually experienced.

The second explanation is more intriguing. Could it be that even though 1 in 4 of the target had responded, those that did respond had failed to make the connection between the what they’d actually done and what the research was asking them? In this scenario the sample is accurate and reaching our 1 in 4 respondents, but those who had responded forgot that they had done so when asked in research. Had they failed to connect the research question to the campaign and to their response behaviour?

The third explanation is that some of the respondents deliberately disconnected their actual behaviour from the answers they gave in the research. In other words, they did respond, but they didn’t want to say so.  They were using the research as a communication channel to share a point of view along the lines of ‘I’m not going to tell you exactly what I did. What I am going to tell you is that I didn’t like being perceived to be in your target audience, or perceived to be the sort of person who would buy the sort of product you were offering’.

Whatever the explanation, this taught me an important lesson; market research and behavioural data can say very different things. Asking people what they did, or think they did, can be very different to what they actually did. If market research tells you something, take it as an indicator not a fact. If it’s something big, do more digging around the research before you act on it.  But if hard data tells you something, whether it’s good or bad, whether you like it or not, you can be sure that it reflects changes in actual behaviour, the ultimate measure of marketing success or failure.

Marketing data analysis gets you closer to customers

Published / by Simon Foster

Smart data analysis can be a major source of campaign insight and even competitive advantage for brands and advertisers. The customer data owned by a brand advertiser can reveal

  • Exactly who buys a given product or service
  • Detailed information about the characteristics of those buyers
  • Which other products and services they buy
  • Which product and service offers they find most attractive
  • Which buyers buy more of certain types of products
  • How you can find more buyers with the same characteristics

These data analysis techniques can be applied to all types of customer data – whether it’s for a retail business, an online business or a call centre based business. Insight from data analysis can be applied across a wide spectrum; from adding inspiration to a creative brief through to changing a company’s entire business strategy.

You may think the claim that data analysis can change the destiny of a business is rather grandiose. But I can can think of two examples of breakthrough data insight from the same category that ended up contributing millions in additional brand revenues.

Sainsbury’s  – Sainsbury’s agency AMV were tasked with increasing the then ailing retailer’s sales by £2.5bn over a three year period. A seemingly impossible challenge until viewed as a data question. The AMV team calculated that £2.5bn equated to £833m per year which in turn equated to £16m per week.  It still looked like a big number until the AMV team considered that Sainsbury’s handled 14m customer transactions per week.  Then the target equated to just £1.14 per transaction. The brief to increase sales by £833m per week could be redefined as increasing each existing transaction by just £1.14. Now the target not only looked attainable, but this data insight led to the idea that lots of small changes could make a big difference.  From this insight came the campaign idea that consumers should “Try something new today”. By asking customers to ‘try something new’ they were able to persuade customers to spend at extra £1.14 every time they shopped.

Tesco – The Tesco Clubcard is now legendary as both a customer loyalty card and a source of information about customers.  Up until the introduction of the loyalty card, many retailers didn’t know who their customers were. And if they didn’t know who they were it was difficult for them to gather the data that allowed them to understand individual customers better. With the Club Card this all changed. Tesco were able to develop individual data driven relationships with their customers.  They were able to understand customer needs better and in doing so they gained competitive advantage over their rivals.

Advertising Frequency and Diminishing Marginal Utility

Published / by Simon Foster

diminishing-marginal-utility

Economists have a concept called Diminishing Marginal Utility. This means that each additional time a consumer consumes something they get less satisfaction from consuming it. So, if I have one coffee, I find it very satisfying, two could be OK, but by the time I get to three I’m not getting much additional satisfaction, infact, I’m going off coffee pretty fast. And if I were to drink ten coffees I’d feel like I was being tortured.

Now let me apply this thinking to the world of TV advertising and in particular, sponsorship. In the UK, quality drama is a favourite for sponsorship. One of the reasons for this is that these programmes attract a high quality loyal audience who make an appointment to view. Certain drama strands can be sponsored heavily in a cross-programme deal covering different programmes in the same genre. Whilst this may appear to present great media value it can mean over-exposure for both brands and consumers. Seeing a break bumper a couple of times is fine, but seeing the same branded break bumper ten times in the same evening can seem like drinking that tenth cup of coffee.