After a pleasing meal, if the waiter or waitress asks me, in a conspiratorial way whether I would like some dessert, or if she recommends a particular digestive, a coffee or whatever - then, more often than not, I’ll happily go along with the expert advice and order pudding, glad that the staff are attentive to my appetite.
What happens if a restaurant just brings you a crème brûlée without first checking you want it? What happens if tonight you’re in the mood for ice cream?
Indeed, if I’m a regular at the restaurant, then I would have every reason to expect that the staff are familiar with my specific likes and habits and that they have anticipated my desires for me – "your usual crème brûlée, sir?"
Good advice, given in the right tone at the right time by the right people is good customer service. It isn’t presumptuous hard selling, is it?
But what happens if a restaurant just brings you a crème brûlée without first checking that this is what you want? What happens if tonight you’re in the mood for ice cream or for nothing at all?
It is difficult to consistently anticipate the desires of end-customers 100% of the time.
Amazon's predictive play: neural network or simple sales forecasting?
Amazon has been very astute, quickly making a sales suggestion based on your own historical browsing and buying habits, and those of other similar customers.
It could be that Amazon has taken this data and this approach to the next level of neural network-powered Artificial Intelligence, where forecasts are made in a non-linear way: where the complex inter-relations of various key prior purchasing attributes are compared to similar patterns of behaviour across millions of previous website visits and transactions.
We doubt Amazon will be using neural networks here; it is more likely clever use of standard demand forecasting by region - perhaps enhanced by 'see how long the cursor hovers over the product' data
If you have browsed product A and B and previously bought a Z, then, according to this kind of algorithm, there is a compelling % of probability that you would like to buy Gizmo X.
Well-trained neural network algorithms are able to learn very well from accurately recorded historical data, and, when fed with new inputs can predict future events with great accuracy (but not 100%, all of the time).
A gold trader might use algorithms based on the interactions between different elements of historic data to predict today’s gold price. My company, Epagogix, uses neural network algorithms built upon historic movie and TV performance data to predict Box Office revenues or audience share.
However, my own neural-network-building colleague at Epagogix - brilliant in this field - has looked at the new Amazon service, and he doubts that Amazon will be using neural networks here, as their functionality does not lend itself to the data that Amazon has publicly discussed.
He thinks that their operational development is more likely founded on clever use of standard sales/demand forecasting by region - which can, in theory, be enhanced by Amazon’s additional and noted "see how long the cursor hovers over the product" data.
Same day delivery for books
My expert colleague suspects that "anticipatory shipping" may be useful for specified SKUs such as new book releases for day of launch delivery to customers. However, for the majority of products, it just won’t be practical for Amazon to "anticipate" purchases, as it stands at the moment.
There has been some surprise in the tech blogosphere that Amazon obtained a patent, as it was considered by some experts to be insufficiently novel
In fact, there has been some surprise in the tech blogosphere that Amazon obtained a patent, as this development was considered by some experts to be insufficiently novel – which is not to say that it won’t work effectively, and delight customers who just can’t wait for the latest Lee Child (I’m one of them).
At the time of writing, it appears that Amazon anticipatory shipping is intended to deliver specified numbers of units into regional distribution centres, having worked out that a certain ratio of (as yet un-named, but probably prompted) individual customers in that area will shortly place an order.
This will ensure ready supplies are nearby for addressing and quick local delivery, once you and I have decided – as the algorithm sales forecast, and as prompted by clever marketing - that we want this new product asap.
There is the possibility that, in the future, neural networks might be able to correctly identify the likelihood of specified individuals making highly particular purchases – within a certain margin of error. If anyone can do it, Amazon can.