Re: [projectvrm] Is VRM an Ideologically-inspired Dead-end?

[sylvain willart < >]

I'll jump in. The invite was tempting enough...

> Both online tracking systems and VRM assume that there is economic value in
> making an effort to know in some detail and anticipate what people intend
> (or will in the near future intend) to buy.

There is a value, at least in the retailing industry. Sales
forecasting leads to better supply-chain management. If retailer know
what customers will buy, he will make the right purchase in advance,
so his aisles will be stocked with the right products, and he won't
face out-of-stock problem that leads to loss of sales (sometimes even
loss of customers)

> I think there is a non-trivial possibility that the alternative is, in fact,
> correct.  The underlying idea here would be that purchasing decisions are
> essentially a random phenomenon.  Not unpredictable in the aggregated, but
> individually random.

Totally agree with Joe Andrieu on that point. Consumer behavior follow
the law of large number and is pretty well predictable at the
agregated level (at least in retail industry).

> You can know that if you offer (say) $5000 off on a
> new car, more people will buy it, but you can't really know which more
> people will buy it.  (Think of nuclear decay.  You can know that on average
> X% of Uranium nuclei will emit neutrons within Y period, but it is
> impossible to predict which ones, or precisely when.)  The more
> high-falutin' way to present it would be to say that human decisionmaking
> arises from a chaotic, complex adaptive system. Very small changes in inputs
> can unpredictably lead to vast differences in outputs.

For long, there was a theory called "no-bridge" which you sum-up
above: the impossibility to predict individual level behavior using
aggregated methodology. But then came bayesian statistics, and
networks model, and geo-statistics. All trying -and succeding in some
ways- to forecast individual-level behaviors. But these methodologies
need quite big amount of -personal- data. And the more they have, the
more they fit actual behavior (a retail industry, in Belgium, even
managed to discover which women were attempting to be pregnant based
on shifts of their consumption - hopefully it didn't use this insight
because the risk associated with making false prediction was too high
and would have certainly lead to loss of brand image).

But in -too- many areas, the negative consequence of a false
prediction remains low (it's not a big deal if people jus't don't
click on your Ad). Hence IMO, retailers did not really take a close
look at margin errors (which are tricky in bayesian models because
there is no R-square, and p-values do not allways follow gaussian
distribution), and have started piling up all the data they could
gather.

Apart from that, I think retailers aren't the only ones to blame here.
AFAIK, Data crunching is not made inside retail companies, but rather
by consulting firms. These firms give results to their clients
(retailers), and when the results go wrong, they tend to blame the
quality and amount of data on which the model was built. Then,
retailers go back to their own customers and gather more data, and so
on and so on...

IMO, VRM could break this vicious circle. Retailers actually want to
know buying intent. If customers give it directly to them, through
VRM, retail industry will not need consulting firms to crunch the
data.
To put it straight: nobody actually need more data (except
statisticians to test their new fancy methodologies, which are quite
cool), but better data.

For now, retail industry has kind of a "follow the leader" strategy.
And the leader is Amazon (from a growth perspective). At Amazon, there
is lot of data. So many retailers want their "Big Data". Just like
chidren want to have their friend's toys.

sylvain

PS: just learned Adblock Plus was removed from Google Play Store. What
"don't be evil" really means exactly?