defining match strategies for MDM

cleanse, match, merge ...

 

i've heard it many times.  on an mdm project this is an important mantra.  it is the mdm process in a nutshell [ref] maybe it is more accurate to say cleanse, match, merge, pub, subscribe? [/ref].

but there is a lot of logic in those three words.  not to mention that how you match data is often a complex decision that dictates a lot of your success.  so let's go over a few strategies for building in as much success as you can.

clean it up

one of the most effective ways to increase your match success is to cleanse your data.  this is why it is part of the core mantra.  sounds simple, right?  well, it depends.  cleansing data implies you know what is not clean. sometimes that is obvious, but not all the time. so here is a list of cleansing best practices:

1. know what data is important to mdm and what is not

cleansing every piece of data is a bit like boiling the ocean and, well, no one has time for that so know what data matters and what doesn't. for example, if you are implementing a customer master hub, focus on things like name and address and let product data stay dirty

2. know what clean is and what it is not

this one sounds simple but, believe me, it isn't. every one will have a different definition here so you need to gather stakeholders together and hammer that out. for example, marketing will likely want to keep Hollywood as a valid city name even though it is merely a vanity address.

3. validate those addresses

customers know where they live and the postman may really know where they live, but you might not. use an address validation service and find out for sure [ref] and do some research on the validation service, because they are not all created or managed equally [/ref]

look before you leap!

you might think you know how best to match your data, but you'd be surprised what you find when you take a close look at your data. the best way to do this is to profile your data prior to defining your strategy.

for example, if you are building a customer hub you might logically assume that name and address is the best way to match them. in the majority of cases you'd be right. however, i have been on more than a few projects where, for one reason or another, addresses where not regularly captured. in this situation you need to move on to the next most identifying piece of information.

in that particular instance, email was regularly captured and was used in stead. i have also seen data where the name and address data was so similar that even though it was regularly captured, it was not a good match predictor. this happens when you use common terminology in account names and when those account holders reside in locations that are referred to in a common way [ref] like rural route or RR. those types of terms drive so much commonality that they all almost always match [/ref]

just as with cleansing data, do your research with profiling tools because those are not all created equal and you want one that can run against your particular data store and produce reports that help you get to the details quickly and easily.

here is a short list of profiling best practices

1. only profile what you need to

just like cleansing data, profiling everything is a waste of time and resources, so only profile what you really care about.

2. look in the shadows

unless there is something majorly wrong with your data collection, your data issues should be in the minority. so look for them by sorting your profiles that way. if you are looking at most common values, your doing it wrong.

3. write it up

 you are not going to be the only one who wants to look at the profile results, so do yourself a giant favor and write up a document of what you find while you are reviewing the profiles. you'll be sharing that thing in more meetings than you ever anticipated too, so make it look nice and explain why you are noting the particular issue.

failure to plan is planning to fail

if you profile and clean your data you are far more likely to generate true positive matches during your mdm implementation and truly deliver the return that an mdm project promises. i have jotted down some things that have helped me in the past few years. it is not a comprehensive list but there are some of the most important tips i can pass on.

include them in your plan and you'll be more successful. blindly determining a match strategy without following these steps is failing to plan which is planning to fail. i've seen it happen and it is very hard to untangle the mess after it is already implemented.