When I first got into Marketing, it was product-focused. As market technology became more prevalent - first as Database Marketing, then as CRM - the insights forced a shift to becoming customer-focused. And that made sense - and produced a lot of revenue.

I’ve been pondering the customer-focus with the advent and expansion of social media. Advances in crawling, text mining, and aggregation tools have brought us as close to one-to-one marketing as we’ve ever been. So that got me to thinking, if we truly know who the customer is, then isn’t it time to tell them about our products?

Think about it. We became customer focused because we didn’t know much about the customer. Now that we know so much more, perhaps we should try to sell them something too. After all, if we know so much let’s just tell them about our products and services.  Just a thought.

For more inquisitive insight, visit me at the DMA in October- Booth #1204

There was an interesting article in a recent Wall Street Journal. It discussed a study that concluded eating chocolate is linked to depression. ‘How is this relevant to me’, you may ask?

The objective of the model was gain insight as to what foods may contribute to depression. But that’s not important for this blog post. What is important is the process and methodology used to arrive at a conclusion.

The study indicated that several factors were introduced into a statistical model to see what impact each factor had on the stability, reliability, and predictability of the model. What the model discovered is that introducing foods other than chocolate had little or no impact on the model’s metrics. That’s exactly the same process we use to build and evaluate our predictive response and purchase models.

Like researchers, we too seek out factors that ultimately contribute to our understanding of who may buy something – or may not.

A couple of days ago I read an article about how data from social networking sites is being used by marketers. The article discussed how chatter amongst consumers is being tracked by marketers. And they are tracking who is chatting with whom - essentially the social network for that particular product or service.

Medical research on behavior modification concludes that behavior modification is achieved through social influences. And that got me thinking: impressive gains can be made if marketers can find a way to harness the social ‘chatter’ that I view as an indication of behavior.

Analytically, this means exciting times are upon us as social media-based data is now becoming available. The ability to add this data to the already existing banks of demographic and lifestyle data will enable highly refined target marketing.

I’ve always said that the objective of marketing is to invoke an emotion in the prospect. Hopefully that emotion is to buy your product or service. The best way to invoke that emotion is by connecting your message to their behavior.

Back when I first got into this business, I applied MicroVision codes to my customer database, did a frequency analysis, figured out which cluster codes my company was primarily selling to, bought prospects with those same codes, and got reasonably strong results. But that was almost 20 years ago when few marketers were doing anything like that.

Why was it successful back then? I surmise that so few marketers were doing any type of targeting that I was getting to prospects better than my competitors.

Interestingly, some marketers still use this technique and until recently have met with some degree of success.

A current client of ours fell into that category. They came to us because their response rates have been consistently dropping.

After a brief conversation with the client it occurred to me that it might be more ‘ROI’ effective to build a response model - something the client had previously considered but concluded it would be too expensive. I was able to illustrate how the economics of a response model favorably compared to the economics of cluster code marketing for the size of their direct mail campaigns.

For clients that are doing rather large scale direct mail, the unit cost of doing a model drops dramatically when compared with the variable cost of doing cluster code marketing.

It is something to consider if you’re not currently doing any sort of predictive model-based marketing.

I’ve been in this business since before database marketing was called database marketing (let alone CRM). When Don Peppers and Martha Rogers et.al. first introduced the concept of 1:1 marketing, it more or less remained a concept. Why? Because at the time, neither the technology nor the data existed to cost-effectively implement such programs.

As I read the current issue of DM News it is clear that more and more marketing services are rapidly moving toward this concept of 1:1 marketing. They are acquiring technology companies. They are acquiring analytical companies. They are acquiring creative services agencies.  Why? There are four reasons:

1) The technology to process nano bytes of data has been refined and costs reduced. 

2) The depth and breadth of behavioral data is vastly improved and readily available - both from traditional sources and now from the social media channels.

3) Modeling software can now swiftly process millions of calculations. 

4) Variable Laser printing technology, PURLs and dynamic web page capabilities.

Although the ability to converge data, technology and analytics has existed in marketing for decades, it has not been able to be as precise as 1:1 because of the cost and time involved.  That has changed.  Additionally, the cost of personalization far outweighed the benefit.

Now with laser printing technology, PURLs, dynamic web page capabilities, integrated with behavioral data and predictive model analytics,  it is now faster and cheaper to bring personalization to the doorstep (or desktop) of each individual in your targeted market.

Why is this important? Recently I’ve developed several regional models. The results clearly indicate that there are regional differences in any given market and that incremental gains are achieved when marketing is targeted regionally rather than generically and broadly.

If you take this a step further than it is reasonable to assume that by targeting on a 1:1 basis yet more incremental gains will be achieved over a regional model. 

Conventional wisdom has it that if you target market to prospects that look like your clients, you will have great success.  Conventional wisdom also has it that by doing something - anything - different you will likely see a change in performance - hopefully a positive change.  All of this is true and I have experienced it many times myself.  However, this approach merits a closer look to understand what the impact is on acquisition costs and therefore your return on marketing investment.

In a traditional marketing campaign, customer attributes such as demographics, lifestyles and behaviors are compared to the same attributes in the market.  If they see a high presence of prospects that share the attributes as their customers, they market to them. Ultimately this technique generates new customers as intended.  However, there may also be a higher number of nonresponders which, of course, they’d like to avoid.

If you consider that on a good day you may get a .5% response rate, then clearly 99.5% of the prospects will be non-responders.  Hence by merely comparing attributes of customers to a market - without regard to how those attributes are integrated to tell a fuller story - may inhibit your ability to gain an incrementally higher rate of  responders above the number of new customers that were acquired.

For example, let’s assume that your customers’ average age is between 35 and 45.  So, you buy a list of prospects that are age 35-45.  Unbeknownst to you, although that demographic likely contains many propsective responders, it may also have a higher ratio of nonresponders to responders.  The end result is that although you get many new customers, you may also get a higher number of nonresponders and consequently your acquisition cost increase.

How do you get around this?

A predictive response model considers how a full plate of customer attributes is integrated to better differentiate the potential responders from nonresponders.  By eliminating those that have a low probability of responding, you increase your overall average response rate thereby lowering your overall acquisition costs.

Although a predictive response model also adds incremental expense, your marketing campaigns may be large enough you need to carefully weigh the financial impact of the savings from not marketing to the low-probability-of-response prospects against the cost of the model.  You will be surprised at how cost-effective a predictive response model may be.

A client recently asked AccuData to develop a predictive model to help target a large national direct mail and email campaign. Notice the request was to develop “a” predictive model - we actually wound up building ten, and despite the incremental cost of building ten vs. one, they drove sky-high ROI for the campaign.

Why did we do this and why did it deliver economically?

Predictive models built for a national prospect universe assume that individuals or households with similar characteristics behave the same way in each MSA. They don’t. Many marketers approach this problem by asking their analytics team to include a geographic factor, essentially asking their team to make sure the predictive model includes a factor for geography. But this is often a mistake.

First, in many cases the geographic factor is weak compared to other key elements (e.g., demos, lifestyle factors). In this case the geographic factor falls out of the model. Alternatively the geographic factor could be so strong that other important variables fall out of the model.

What works?

In our experience the best way to address the challenge is to build distinct models for each key market. In this way a weak, but potentially important geographic factor, is incorporated into your targeting. At the same time a strong geographic factor is incorporated because you will develop models at the MSA level (or some other geo level).

If you are a large marketer you must fight for every 1/100th of a percent of response. In our experience, building localized models for key markets delivers critical competitive advantage.

The prevailing perception among many marketers is that all buyers of your product or service are alike.  Hence they typically focus on age or income or other basic segmenting factors.  The fact is that there are regional influences and behaviors that clearly earn a place in differentiating those that will buy from those that don’t.  For example, in some recent models I’ve built, there were buyers from both urban and rural areas.  The rural buyers tended to like Gardening and Fishing.  If we were to apply that natioanlly - without regional geographical recognition - we would erroneously use those attributes for targeting purposes. We would completely miss densely populated areas that likely do not offer Fishing and Gardening opportunities.  Clearly, the opportunities offered in an urban setting attract a certain market segment whereas the opportunities offered in a rural setting offer different opportunites. 

Recognizing these important regional differences will increase the probability of a purchase.

I find it interesting that there appears to be a direct correlation between data quality and model quality.  Fundamentally this might make sense to modelers and data gurus alike.

Interestingly, on the client-side there are still many organizations that do not adequately capture data on prospects and newly acquired customers - often no clear linkages between the two. This creates a disconnect.  Consequently, the patterns that normally can be extracted usually weakly present themselves.

Bottom-line: it is critical that organizations that want to improve their marketing results, focus on getting their marketing database in shape.

Many times, perception is not reality.

There are many hidden gems in our clients’ marketing databases.  These gems gradually get uncovered as we go through the Exploratory Data Analysis phase of our model builds.  This is further refined by applying not just one - but several different modeling techniques as we build the strongest model possible.

Recently we used a Classification and Regression Tree method to build a segmentation model for a client.  The client had a specific image of their customer profile.  Our model - obviously using the client’s customer data - told us something different.  Needless to say, the client was surprised and delighted that we had uncovered a hidden gem - a ‘buried’ market segment that clearly had an affinity for the product being offered - and a previously unrecognized growth opportunity for our client.

This reminded me of one of my first projects for a large advertising agency (long ago - longer than I care to remember).  The project involved creating a catalog of services for the biggest telecommunications company in the US.  Assumptions were made about whom the buyers would be and the graphics and copy written to reflect that.  The catalog underwhelmingly performed.  Mayday: client not happy!!!

My back-end analysis discovered that the actual market was 180 degrees opposite of what the assumptions and perceptions were.  Based on my analysis, the catalog was overhauled and ended up exceeding expected results.  Alert: Client very happy!!!

Lesson learned: Let the facts drive your marketing decisions. Perceptions can be misleading and may lead to disastrous results.