Monday, 15 December 2008

The British Post Office

At the end of last week, many news bulletins in the UK carried a story that British Postal Delivery workers ("Postmen" or "Postwomen") were unhappy with a new computer system which had changed their delivery routes. According to the stories, some postpersons were having to walk at 4mph for three and a half hours. (Example story.)

The stories were short on detail about the "computer system" except that it had details of 27 million addresses and came from Canada with the name of Pegasus Europe. It seems to have been around for several years, but evidently there have been recent changes in the system which have led to the unhappiness.

Reading the story, it was obvious that there was a mathematical model behind Pegasus of a familiar kind -- optimal (or efficient) route planning for vehicles (postpersons) with constraints (loads, time). Someone, somewhere had been using operational research to help the postal service. And as an operational researcher, I ought to feel proud.

But I don't. The news reports showed that something was lacking in the O.R. process. I can recognise several possibilities for what was wrong, but without further information I can't give a full diagnosis. Maybe someone from the post office can help. So, in no particular order, here are my observations and questions:

(1) Was this system tested and developed for the UK postal service, or was it an off-the-shelf system into which British data was inserted? If it was the latter, then did anyone verify the assumptions that had been made by the designers?
(2) Given the size of the database, it seems likely that the system is largely, if not wholly, deterministic. If so, what sensitivity analysis was carried out? And if so, what changes were made to the data and the model as a result? If not, why not?
(3) How much communication was there before, during and after the development of the system? Who with? Did the creators/users of the system discuss what they were doing with staff at all levels of the system?
(4) Was the objective simply cost-based? Or were there other criteria?
(5) Did anyone concerned with the data collection, input, modelling and recommendations actually go out and test the results? Or, to put it bluntly, would the modeller trust the model's output if they were asked to do a postperson's job?
(6) Did the modellers collect feedback from the postal service about the implementation?

For years, I have taught that an essential part of O.R. is a feedback loop, that an O.R. project is not properly implemented until it has been accepted (and possibly welcomed) into the practice of the organisation. It seems, from the press, that this project lacked this.

We are often reminded of places in companies where there is "OR inside". I feel that this is a story of "OR Inside" which omitted the essential, friendly face of OR outside.

Where was Genchi Genbutsu?

eBay and the AHP

I am an eBayer (or whatever the appropriate term is) to buy items for my stamp and postal history collection. From time to time, I have also sold on eBay. As such, one becomes aware of the size and complexity of the organisation that hs come to symbolise online trading in many countries. From time to time, once one has registered with eBay, one becomes aware of the data analysis carried out by the company's statisticians. I was emailed to ask if there was anything wrong, as I had not bought as much in the previous few months as I had in a similar period earlier -- in other words, my behaviour was outside their forecast limits.

Last week, I was invited to take part in an eBay survey and said yes. I am not sure what behaviour of mine had prompted this particular survey; it may have been another action that was outside their forecast limits, when I bought in a "Buy it now" sale; I invariably use auctions for my collecting.

There was something familiar about the structure of the questions, but I confess that I didn't recognise what it was until after I had answered the last questions and submitted the online survey. eBay was using the AHP (Analytic Hierarchy Process) to find out what I felt about some of the factors that vary from vendor to vendor. I was asked to imagine that I was considering an item at a particular price (30 British pounds) in a Buy it Now sale. That fixed one variable. Then I was asked to consider variations of delivery cost, photo and description quality, speed of communication, first in a ranking, and then by a succession of "Which do you prefer?" comparisons where two sales with differing profiles were offered. I think that I was reasonably consistent in my answers, although the example sales that were offered did not fit into my profile of eBay use. (For stamps, postage is seldom more than GBP1 within the UK and usually less than USD3 internationally; the examples offered had postage figures of up to GBP6.)

Wanting to find out more, I Googled for "eBay analytic hierarchy process" and came up with the answer that eBay does use AHP and ANP (network) and its use is mentioned in the citation of an INFORMS award to Thomas Saaty and in a publicity release put out by Decision Lens, the consultancy used by eBay for their AHP work.

It is strange to contribute to an OR study and only realise that you are doing so afterwards!

Monday, 1 December 2008

Genchi Genbutsu

Genchi Genbutsu is a new expression for me. It was posted in a discussion board, as relevant experience for anyone in O.R.. The wikipedia summary is:
Genchi Genbutsu (現地現物) means "go and see for yourself" and it is an integral part of the Toyota Production System. It refers to the fact that any information about a process will be simplified and abstracted from its context when reported. This has often been one of the key reasons why solutions designed away from the process seem inappropriate.

I don't know why there is so little literature drawing on the parallels between Genchi Genbutsu and practical O.R.; good O.R. requires the modeller to understand what is happening in practice, and you do not learn that by sitting in an office. I suppose that there are eqivalent expressions in English -- "go and see for yourself", "management by walking around", "walk the line". But too many managers (and a few O.R. people) rely on second-hand information.