Recently, in a fit of quixocity, I took the overnight sleeper train to Melbourne. I am not sure what motivated me to try it again after quite some years, maybe positive experiences with trains in the rest of the world had raised my expectations, had led me to give it the benefit of the doubt, had led me to assume that a first class ticket (no cheaper than the plane) would lead to a first class experience.

Naturally, I was disappointed. I could rephrase that to say “as expected, I was disappointed .. my expectations were not met”.

There is much of semantic interest here : there is an implied back-of-mind lingering mistrust, an implied attitude of “give them another go, but ..”.

OK, would I travel with them again?. No, I would not. And that seems to me to be a big problem – repeat purchase rates amongst full fare paying first class passengers would, I would have thought, be of prime concern. (However, they probably won’t even know of my intent .. any monitoring programs in place?)

So, what went wrong, and what has this got to do with data analytics?. Well, I could give you a litany of complaints and I will a bit later.

Measuring the Negatives - the “litany of complaints”

It is the “litany of complaints” that interests me.. the concept is that it (the litany, the complaint set) is large (that is they are numerous) and that the complaints are detailed and highly specific.

It is also suggestive that the complaints have a subtractive role .. that is, one (the customer) starts with a reservoir of goodwill (a suspension of mistrust, a positive prior expectation) and that “things happen” to eat away at that goodwill, with consequent diminishment of repeat purchase intention or positive word of mouth.

That is rather in contrast to the usual and rather anodyne/bland “customer service” model, where all the positives are “measured”, usually positive abstractions (e.g. the quality of the room) and these are added up in some manner. Working with such data is often rather unsatisfying, partly because the question measurement scale is often poor (most people give ratings at the high end), but partly I suspect that the data collection is focusing on the wrong problem.

People are not good with abstractions. And they are not good at telling you what is good about something (unless you are really skilled in the questioning), but can readily enough volunteer the specifics about what went wrong, how their expectations were disappointed.

Note that I am emphasizing SPECIFICS, almost in a qualitative “anecdote and incident” sense.

That is going to make it harder for questionnaire designers and analysts, but so be it.

Better that than collecting data about abstractions which are neither directly actionable, nor meaningful to customers nor management. Monitoring nonsense abstractions only alerts you to changes in nonsense: it does not tell you what you can, specifically, do about it.

Supposedly “the devil is in the details” .. in this case, we might say that salvation is in them. If you have “incident reports”, you can always aggregate them in some meaningful fashion but simple minded measurements of abstractions will not allow you to disaggregate those.

Bottom up, rather than top down, OK?

Here is my litany of complaints

• There was nowhere to wait between trains (I had two hours to kill). For a first class fare, I expected a first class lounge (with power for my laptop). My (admittedly inchoate/vague/latent) expectations were not met, not nearly met .. to the extent that disappointment moved to frustration thence to anger.
• The train was 1 hour late in departing, and announcements were uninformative about the likely extent and effect of the delay, and were self serving and unapologetic and contradictory about its cause .. I expected close to on time running, I expected that staff would be genuinely apologetic about the situation, I expected it to be a rare event (and not treated as a common occurrence of little import)
• I expected, reasonably enough since I had specifically enquired about this, to get dinner on the train. I was disappointed when I was told that, due to the late running of the train, there would be no meal, just “snacks”. (I also expected that there might be some modest effort to cater to different dietary requirements .. in this, I was also disappointed).
• On the return trip, which was on time (but again no waiting facilities, no early boarding), there was indeed “dinner”, which was to be delivered to the room. Delivered it was, in a flimsy cardboard box, an hour late, microwaved to the point of being dangerous and near to inedible. My modest expectations (and they were modest), were again dashed
• Breakfast again was poor .. no attempt at presentation, a flimsy low quality cardboard box with not even a logo printed on it (not exactly a bento box), little choice, supermarket bread
• The trip was a little noisy, the bed squeaked, various things went bump in the night ..but I can live with that. No points, of consequence, lost there.
• The cabin was adequately clean, but only just. Not a significant source of complaint.
• In spite of the fact that half of the cabins were empty (later occupied by staff, I suspect) I was assigned a rear facing seat. A forward facing seat, or whatever was my preference, should have been locked in at the time of booking.
• Staff collected towels half an hour before arrival .. for their convenience, not that of the customers

OK, that is my complaint list, reasonable or not. Others will have different, or less numerous, or less serious complaints.

The Concourse of Complaints

Which brings me to my next point .. how to develop a “concourse of complaints”, how to structure this into some sort of (hierarchical) factor structure, how to measure it and how to model it.

Remembering that we want to retain the fine grained actionable nature of the beast .. we don’t want to get too far away from those detailed incident reports, we want the details and we can then summarize the details. The summaries for the overview and for monitoring, the details for direct action.. for fixing things.

We also want to model this.. that is, we want to know how serious (in terms of impact upon repeat purchase intention and positive word of mouth) these incidents (or incident types) are.

Of course we can simply ask customers for a “seriousness” measure on each incident, and that might suffice at the concourse development stage, but an inferred importance weight (inferred from regression of repeat purchase intention against incident type) is arguably better and reduces respondent load. (There are various hybrid designs that could be adopted).

I suspect also that a simple model of the NUMBER and FREQUENCY of things that went wrong (as a predictor of repeat purchase intention) would be informative.

Now, to build the model and a monitoring framework, data is obviously required (as in, a simple monitoring survey of passengers, perhaps mailed to their home address) and the design of that exercise needs some care.

But before that , we need to develop the “concourse of complaints”.. the universe of all the things that can go wrong from which we can sample and develop highly specific questions that are “orthogonal”. To go from “incident report” to fine-grained measurement instrument.

To do this, we need to go into the world of Q methodology, lines of communication, and semantic factor analysis.

Another day.

Case Based Explanation

btw, there is a connection between this bottom-up, “incident and anecdote” approach and case based explanation.

Case based explanation asks the question “what is the cause of ..”, somewhat equivalently “what is the explanation of..” and makes quite heavy use of analogous situations and extrapolation therefrom.

Obviously if our question is as broad as “why does State Rail have low profitability on its interstate services” then there are multiple factors (causes, explanations) at work, and so case based explanation (CBE) does not map directly. But I am still interested in CBE, specifically how “anecdote and incident” could be used to create prototypical cases and explanators.

If you are interested in CBE, there are some books around .. “Inside Case-Based Explanation” by Roger Schank is OK, but there may be better ones. Also, check out the area of “analogical reasoning”.

Obviously, working with the positives is good too

and some companies are mining their customers for stories/ anecodote see for example

1. http://marketingroi.wordpress.com/2007/02/26/the-stories-loyal-customers-tell/
2. http://customersrock.wordpress.com/2007/02/23/stories-and-the-personal-touch/

but that is mostly commonsense marketing communications, not a real attempt at building a meaningful customer service metric.