Data Miner for Ogilvy Paris, Samantha Bilodeau has +10 years experience analyzing data and helping brands discover their consumers on a deeper level.

How does your job fit into the advertising process? Data mining is typically one of the first things brands want done, either as a prerequisite to an advertising or marketing campaign or as an audit of the overall health of their brand. I work mainly with account planners and strategic planners. I don’t normally work with creatives, however that is something I would be open to doing.

Data mining is the study of monitoring consumer behavior. I begin by uncovering:

• Descriptive statistics: developing a snapshot of the consumer’s current behavior – age, gender, demographic, which of your product(s) consumers purchase and how often, how many coupons do consumers cash in, etc.
• Predictive statistics: using probability to anticipate what consumers will do, what, when and how they will buy your product next and assessing your brand’s position in the minds of your consumers.

For example, if data mining reveals that your brand’s typical consumer goes shopping every week, but only purchases your product once a month, the brand must address the problem and then encourage consumers to return. For example, perhaps:

• Your package size is large enough that they only need to buy your product on a monthly basis
• Your brand is considered a parity product and consumers don’t actively seek out your particular brand

In my experience, brands usually have their own data and it’s my job to organize that data, combine it with other available information such as questionnaires or online information and then interpret that data into actionable solutions for the brand. For grocery stores it could be your purchase history. With tele-communications companies it could analyzing the amount of time you spend talking on the phone and communicating via text message.

[EDITOR’S NOTE: Refer to the book Emotional Branding by Marc Gobé for more on how brands collect information about you and how you can use it to better meet your consumer’s needs.]

For example, I once worked with a large hypermarket that wanted to create a personalized mailing list catered to each ‘type’ of consumers who were shopping at their store: Single? Married? With children? If yes, their children’s ages? Etc.

For predictive statistics, the company had data from several questionnaires as well as a fidelity card program where families could register in exchange for discounts. With this data I was able to determine the family status of each individual customer in their customer base:

• How often they shopped
• Which products they purchased and in what quantity
• Which products they bought faithfully and which they bought periodically
• The hours during the day they usually when shopping
• If they cancelled their fidelity card, was it due to bad service or through a competitor’s marketing
• Etc.

This snapshot then allowed us to personalize the brand’s mailing lists by predictively anticipating the best products and promotions to offer each consumer category, when to schedule more cashiers to meet the influx of consumers, and so on.

For example, if your business is run on a contractual or a subscription model, then understanding who is unsubscribing and why will give you an indication as to who is likely to unsubscribe in the future. If you’re a tele-communications company and you notice that an unusually large number of consumers who typically use their phones more to send written communications than they do to talk to a person directly, then you might be able to deduce several things:

1. The consumer didn’t use all of the services that came with the contract, either because the consumer didn’t care about the services, or perhaps because the consumer wasn’t sufficiently informed when they signed the contract.
2. Disruptive mobile applications (such as the free Facebook messenger) may have diminished the tele-com’s current offers to the point where those consumers now want the least expensive offer available.
3. A competitor, having conducted their own data mining and already uncovered the profitability of this developing consumer behavior may have launched a special offer aimed directly at that demographic.
4. That your competitor’s special offer is so appealing relative to your current contractual agreement that consumers are willing to end their relationship with you for your competitor’s offer.
5. That unless you act now, that identified percentage of consumers within your target demographic who prefer text messaging may soon be enticed to leave you as well.

Can you walk me through the data mining process? First, we sit down with the brand and create a brief to uncover the objectives of the company and the data mining process. I need to understand all of the problems the brand wants to solve. Typical questions we could ask include:

1. What target(s) specifically does the brand want me to address? To be more efficient? Because a lot of customers are leaving for an unknown reason? To follow up on an advertising campaign and determine the campaign’s conversion rate and profitability?
2. What does the brand think is the cause of the problem they’re having? This question is particularly important as a hypothesis to determine if the brand has any deeper underlying problems that need to be addressed such as brand mission or another internal inconsistency.
3. What data mining have they already conducted and what were the results? This helps us compare with the previous results, know if we need to present the data in a particular way congruent to the previous data’s layout and could save time by not re-analyzing data.
4. What data is available? How old is the data? In what format is the data?
5. Who at the brand is in charge of the project and who else in the brand’s company might be interested in knowing the results of our analysis? This question is important at the beginning when I am determining the different variables because if later it’s discovered the Sales department could benefit from the analysis.

Step two is retrieving their data, either via USB key or through access to their online database, and configure it for my statistics software. There are many different software available for analyzing consumer data which can come in the format of: .txt, .xls or .csv files, for example.

Reliable data mining software include:

• IBM SPSS Statistics
• IBM SPSS Modeler
• R
• KXEN
• SAS

For step three I confirm that all of the data files sent to us can be correctly configured into my data mining software. If their data files were incomplete or corrupted, then I try to salvage what data I can before contacting the client to explain the problem and discuss possible solutions.

I then follow up this step to ensure that the values are correct and that there aren’t any important values missing. With all that correct I am now confident I can data mine and answer the objectives outlined on the brief.

Step four is analyzing for the descriptive statistics.

Step five is analyzing for the predictive statistics I mentioned earlier.

Step six is turning my findings and recommendations into a PowerPoint presentation that answers the brief for the brand to understand and act upon.

And finally, step seven involves incorporating the model into the brand’s database for future use.

From there, the findings may go to the strategic planner who incorporates the findings into his branding strategy or to the creative directors and copywriters who then create an advertising campaign.

Typical data mining models include:

• Scoring – the probability a consumer would do something such as purchase a particular product, unsubscribe to a mailing list…
• Text mining – analyzing open-ended answers and customer comments against a dictionary of words to determine positive and negative feedback.
• Percentage – percentage of consumers who are happy with a particular product
• Classification – regrouping people who have similar behavior and profiles and sort and adapt them according to each grouping.

[EDITOR’S NOTE: The books Consumer.ology by Philip Graves and Buy*ology by Martin Lindstrom as well as my interview with Peter Spear discuss the risk of including questionnaires in your data mining models, notably how questionnaires subconsciously influence the consumer’s decisions, thus potentially rendering the data gleaned useless.]

What is the importance of investing in data mining? Understanding your consumer is the fundamental building block of a brand and absolutely crucial to a successful communications campaign. Imagine spending all your money creating an advertising campaign or paying a public relations expert to communicate the wrong message to your consumers and potential consumers!

Secondly, basic problems such as a low conversion rate or a mass customer exodus with no apparent reason are often symptoms of a larger problem(s) that data mining can help detect shed light on.

At what point should I consider investing in data mining for my business? Data mining can be quite expensive, but like I said understanding your consumer is crucial. So monitor your global statistics: conversion rate, rate of consumers joining versus leaving, etc, and when you get to the point where you have so many consumers that you can no longer monitor everything they are doing, then is the time to consider periodical data mining analysis just to stay on top and in touch with them. You could run probably some pretty good statistical models with as little as 1,000 consumers.

What are some misconceptions brands commonly have about data mining? « Tu peux faire tout ce tu veux aux chiffres ! » Translation : “You can say anything with numbers!” Meaning, some brands may approach data mining with a preconceived outcome that they want the data miner to validate, regardless of what the actual numbers indicate.  Like I said, it’s good to have a hypothesis so we can test your expectations against the actual data, but the objective should be better understanding your consumer, not making sure the numbers prove your point. If that’s what you want, then you’re wasting your time and money.

Another misconception is that data mining can be done quickly and easily. Rule number one is ensuring that all the data has been transferred correctly. That is the most important and time consuming step. If your raw data isn’t correct, then your results won’t be correct.

Another time consuming step is actually translating the results of the data mining into actionable information the brand can take and use.

How often should a brand conduct data mining on its consumers? It’s important to begin slowly and overtime invest more deeply.  The thing about data mining is more you learn about your consumers, the more questions you’ll have and the more you will want to learn about them.