Upcoming Jan 11 Web Event – IAIDQ

January 6, 2011 by
Filed under: Events 

I will be presenting on business impacts of poor data quality at a web seminar hosted by IAIDQ this coming Tuesday, Jan 11. Register at http://iaidq.org/webinars/2011-01-11.shtml.


2 Comments on Upcoming Jan 11 Web Event – IAIDQ

  1. Tom Kunz on Mon, 31st Jan 2011 10:24 AM
  2. David,
    I participated in your webinar and have also downloaded and read the paper on this subject that you authored. While I take no issue with the concepts and methods you propose, it doesn’t really help me get any closer to actually calculating the cost of poor data quality. As I have thought about this, I have come to the conclusion that because there are so many variables, whatever cost is calculated will likely be incorrect. Assigning a high – medium – low to data errors is potentially doable, but when we do this we end up with an unprioritized list of “highs”.

    An example: When a delivery address is wrong, there are a number of outcomes: 1) the product is delayed for delivery. 2) the customer gets mad and ends the relationship and we lose the sale and the customer 3) someone catches the error before the product is shipped, so the only cost is that of rework 4) the delivery is late because the driver first goes to the wrong address, but then calls the company and finds the right place. 5) the product is not only taken to the wrong location initially, but uses the wrong route not suitable for a delivery truck and a fine results when the truck is stopped by the police. 6) etc , etc.

    So with one error I have at least 5 scenarios, all of which may have occurred. I could risk adjust the cost of each scenario and add the values together, but then the linkage becomes so remote that the business will not be able to recognize it.

    How do you deal with these sorts of issues in determining the cost of poor data quality?


    Tom Kunz
    Data Manager, Downstream
    Shell Oil Company

  3. admin on Mon, 31st Jan 2011 2:25 PM
  4. Tom,

    You are right that there are a number of scenarios, and your approach reflects one of the methods that we take when doing an evaluation. You have:

    A) Identified a data error; and
    B) Determined some business processes that can be impacted as a result of the error.

    Your example provides some potential outcomes resulting from the existence of the error, ranging from the extreme (the customer is lost) to relatively benign (the delivery is late, perhaps not even noticed by the customer). So to assess the (potential) loss of value, the next set of questions has to determine:
    C) How often the data error occurs;
    D) Of the times that the error occurred, what percentage of the time did each of the scenarios happen; and
    E) The (average) cost of each scenario.
    So let’s say the delivery address is incorrect 5% of the time. For all the times that the error occurred each month, it only resolved into a cost scenario 20% of those instances, and was irrelevant the other 80%. Of that 20%, half resulted in a minimal cost of rework (such as the driver having to find the correct address on the fly), one quarter resulted in rework and additional shipping costs (the shipment was returned and had to be redelivered), and one quarter resulted in a lost customer (and therefore a full loss of the customer’s future lifetime value).
    A rough estimate (presuming no changes to the environment) uses the historical aggregates as future probabilities. In this example:
    – The data is error-free 95% of the time;
    – The error occurs 5% of the time;
    – The error is only relevant 1% of the time (20% of the 5%);
    – 0.5% (50% of the 1%) there is a minimal rework cost;
    – 0.25% (25% of the 1%) there is a rework plus extra shipping;
    – 0.25% (25% of the 1%) there is a lost customer.
    If you have an estimate for the value of each cost scenario, you can multiply the cost by these probabilities and then sum them.
    Of course I can already sense the next comment, which is how do you figure out the cost of each scenario? Let’s leave that as a future topic for discussion.

Tell me what you're thinking...
and oh, if you want a pic to show with your comment, go get a gravatar!