A scenario that accurate de-duplication can help prevent...

Imagine you are developing a new product and need to conduct a correctly ‘sampled’ test mailing to determine if your offer, pricing and anticipated response make financial sense. This is before rolling out mailings to an estimated 20,000 customers across the UK.  You’ve spent time defining your target audience based upon some ‘rules based’ selection criteria - fields such as gender, Postcode, RFM purchase history (recency, frequency and value) and excluded those who are bad debt risks. You also only want to mail one person per household.  N.B. If you’re Business-to-Business marketers, please apply some lateral thinking. 

What could be wrong so far?  Apart from the targeting, which would be improved through the application of data mining and customer dynamics, you are reliant on some basic database information and trust in your data.   Lets say this test promotion is to offer a high-value scratch-card to your loyal customers on a bi-weekly basis - females who spent over £1000 on clothes, have traded more than five times in the last two years, live in M, PE or B Postcodes and who have never defaulted on payment. You are testing two pricing offers based on total spend, rewarding more profitable customers accordingly. 

Now suppose your database is still riddled with duplicates and maybe orphan data (transactions left stranded after their parent records have been removed), despite prior de-duplication attempts using matching processes or rudimentary de-duplication software. Also some addresses are incomplete but there’s worse to come. The gender field has been computed based on some rudimentary ‘first name’ checks and ‘title’ field, which was never mandatory and defaulted to ‘Mr’ if left blank.  

So, what negative effect could this have?  Well, there is a strong likelihood that some customers will get mailed more than once - possibly with conflicting offers as transactions may be aggregated over separate records, in effect portraying different ‘total spend’ amounts for the same person. Next, because gender is missing or wrong in many instances, good prospects have been omitted or males targeted instead. Incorrect or missing Postcodes mean the geographical split is skewed, RFM computations are inaccurate due to the omission of ‘product’ data and duplicated records, and finally, bad debt flags have been applied to only some records in a duplicate set leaving a person with high credit risk still being mailed.  

What about financial loss?  If, despite your response analysis telling you otherwise, the new product test fails from a return on investment perspective and you then mail your 20,000 loyal customers. Not withstanding any wasted mailing cost, you upset 2.5% of those customers. This equates to 500. Given they have each spent over £1000 in the previous two years and there is no reason to doubt they would not repeat this over the next two years, you are looking at a potential £500,000 of lost business!  If your margin was just 10% then poor data has cost you £50,000!

As a last point - what if your analysts had been spending weeks or months working with duplicated data and incorrect computations such as lifetime value. Think of the cost implications - especially if this flawed analysis forms the basis of important marketing and business decisions!

To summarise, poor quality data does little for customer loyalty, skews targeting and makes the results of your test promotion questionable - especially from a financial aspect given multiple responses from customers who later turn out to be duplicates make a mockery of response analysis. 

Tech4T can help you avoid the scenario detailed above. Each of the solutions we deploy is based on a mix of tried and tested methodologies and software, developed over the last 14 years and fully utilising our unique skill-set that bridges marketing, data, statistics and IT - all under one roof.

As such we can provide you with a one-stop solution to your de-duplication problems!