I had to buy a new ride-on mower lately .. I didn’t want to, but it was a tradeoff between doing so, spending more money on repairs to the old one, buying a new “long motor” and installing that – or mowing 5 acres by hand. That tradeoff is another story, interesting too, but let’s leave it aside. Just let’s work with an assumed starting point of “I am in the market for a new mower”.

So, you could not give a fig for my mower problems, but the decision process is somewhat interesting and more so from a modelling viewpoint.

I know we aggregate things in data analysis. In doing so it sometimes seems that we somehow blur the edges so that linear compensatory models of one sort or another, or maybe tree structured models, end up looking OK.

And anyway, studying individual decision processes in a strict quantitative fashion is “too hard” . It seems like we would never have enough data from one individual to fit a per person model [although there are some Bayesian approaches, not that I find them compelling] and maybe we would not know what to do with all those models even if we had them.

“Too hard” that is, unless we actually asked the person and they could give us a coherent explanation of their choice process .. and if we could somehow map the verbal into the numeric. Right, that is going to be hard too.. but is it sensible just to collect data from people in order to fit models and be deaf and dumb to THEIR explanations?

The other interesting issue here is the role of the web. It featured exactly ONCE in the purchase process, and I will talk about how that worked. Now, admittedly in the “prior to being in the market” stage I did doodle around the web, ignored just about everything, was not in the mood to set up a spreadsheet to compare features and prices .. none of that was recorded or featured at all later on.

Now, I will just briefly verbalize the process I went through and then set it out in a nice little flowchart (Graphviz) .. which you will then see is highly sequential. No brand-price tradeoff, no tradeoffs at all.

It went like this.

I had a chat with my brother-in-law .. we’ll call him Bruce, because … Well, good ol Bruce has his ways but he does know a bit about engines and believe me he can help you while away a suburban afternoon discussing the many and varied pleasures and benefits of grass cutting.

So, I talked to Bruce .. or listened. Then he took me to a couple of his favourite mower shops… he would have taken me to more, but I demurred.

OK, first shop is run by a mower mechanic with grease on his hands (a good sign) and after he and good ol Bruce had had a good chinwag about the best type of bolt to use on a 53 Chevy manifold (or something) he (the-mechanic-with-grease-on-hands, name of Barry) turned to me and said .. this here (leaps on yellow mower, drives it around shop at tremendous speed, then faster same in reverse, and does a few wheelies) – this here is what you need : “Keep it Simple and Keep it Tough”.

Right.

Second mower shop the guy tries to sell me a streamlined red job from China. Cheap, but nope.

THE CONSIDERATION SET – A SINGLETON ! and flatlining through BPTO (Brand Price Tradeoff)

OK, at this point I had reluctantly decided to buy a mower and had only one brand – indeed one model – in mind. The price was significantly higher than I had mentally set my threshold at, but if I could not get a simple/tough combo then I might as well drop out of the market.

So, my “consideration set” contained ONE item. If I had been asked to do a brand-price tradeoff research exercise at this stage, I would have flatlined through it .. almost no price sensitivity information would have been gained from me, nor any information about the relative utilities of other brands.

Note the difficulty of obtaining any information from me about the brand-price interaction for the mower from China : I guess there might have been some point at which I was prepared to buy it, but it would be hard to elicit that data. And would this be best modelled as an interaction effect, or just a very negative brand utility which cold be compensated for by a substantial discount?

So, to the web.

CAN I DISCONFIRM?

I went home and looked at the manufacturer’s website.

Now, it was not pretty. It wasn’t grossly ugly, it did not scream “backyard” .. so, I could put that aside (just about). I browsed around, read some specifications and non technical explanations about why-it-should-be-thusly-and-not-otherwise (rare) and some history and some testimonials, reassured myself of the origin of some components and the terms of the warranty (not generous, but…). I still did not really want to buy a mower, but there wasn’t anything so negative there that I could reasonably say no (which would have been my preferred option).

You might say that the website allowed me to confirm my decision. Or would not allow me to disconfirm it.

And the other thing I did at the manufacturer’s website was to find the name of the closest local dealer.

I rang that guy up, never went to his store, the price was what I expected.. and the deal was done. Not only that, I bought a couple of ancillary items he suggested.

THE DECISION PROCESS AS A PRETTY PICTURE

Enter the WILD DIVERGENCE

You have probably seen something like this. It looks good in presentations, and makes us believe that we “understand the decision processes of the customer”.

Well, maybe we do – of one customer at least.

But market research and data analytics (predictive modelling) practice IS WILDLY DIVERGENT FROM THIS. Decision tree models come a bit closer but in general we are working with aggregated data which we then split or smooth or estimate marginal effects from. The resultant models look nothing lie that decision process above.

Maybe it is a pipe dream. But I’d like to see a model in which MY realities were reflected. And I’d like to see models more closely tailored to the individual, and a more formal use of verbal explanations. And maybe we can incorporate something from case based reasoning as well.