Covering Disruptive Technology Powering Business in The Digital Age

Login
Register
image
Shipwallet: How big data and predictive analytics could help slay Amazon
image
July 29, 2016 News

This is a situation that Piotr Zaleski, CEO of Shipwallet, knows all too well. He recognises that people are willing to pay for convenience and it’s this shift towards simplicity that he hopes will work to his company’s advantage. His idea isn’t a silver bullet to stop Amazon in its tracks; it’s more like a lifeline for its rivals in e-commerce.

By devising algorithms and developing machine-learning techniques, Shipwallet is focused on dealing with one problem alone. It’s a big problem, though, and about as fundamental as is it gets when it comes to e-commerce. We’ve all been there: you’re just completing an online shopping order, you’ve gone all the way to the payments—all good and then: how do you get it?

All too often the choices are stark, annoying, or quite simply not what you want as a delivery method. Wouldn’t it be great if, at that final stage, your shipping preferences were already known and that, for different kinds of item, you’d be presented with options tailored to suit you and your locality?

It’s an oversimplification, but in effect this is what Shipwallet does. Although it’s still early days for the Swedish company, the service is notching up 500,000 transactions a month by tapping into that convenience factor. Not only are Internet shoppers benefiting, but so are merchants who have been struggling for oxygen underneath the long shadow that Amazon casts across the online marketplace.

The idea is to aggregate all the various shipping options from hundreds of logistics companies and use the algorithms to offer up suitable choices, as Zaleski explains:

Consumers can configure their shopping with much more granularity, but we also have a predictive analytics platform on top of it, which calculates the best shipping options. So a new consumer comes to the checkout and we say “you should basically have this shipping option.” Using all the data we have from the shippers—from Google and also from the past behaviour of your neighbours—we can say “this is the best shipping option for your particular address.”

You don’t even have to build up a profile, as Shipwallet will have already a good idea of the logistical solutions people have been opting for. These choices will also vary depending on the product.

Zaleski gives the example of having a pair of shoes delivered: they would be too big to fit through a letter box, and leaving them on the doorstep poses the risk of theft. But by examining data on the delivery preference behaviour of neighbours in your area, popular options are defined and presented. Perhaps most people like to retrieve larger items from somewhere like a nearby Doddle (a parcel collection shop), instead of a home delivery, and Shipwallet would show that as a likely solution. By contrast, a book purchase that would slip through a letter box, wouldn’t need any elaborate choices.

The cloudspotter’s guide

When it comes to the technical decisions, receiving $100,000 (£75,000) from Google Ventures has been influential in Shipwallet’s development. This funding can’t be spent on salaries or the executive cocktail cabinet—only on Google Cloud Platform products. In essence, it unlocks access to an infrastructure stack of compute engines or virtual machines alongside database, storage, and networking systems. With these resources, Shipwallet’s technical team are able to configure customised tools, and have been busy building insights from their data sets and refining the predictive algorithms.

Technically, Shipwallet is built using the open source language Go, which started off as another one of those Google experiments. Given the Google environment, using Tensorflow was a logical choice for machine learning and, likewise, its Cloud Datastore NoSQL database and analytics tools.

The company will soon have access to massive databases from “a couple of really big merchants” but the size of the current data set is limiting. Zaleski openly admits that the machine-learning aspect is not yet mature. He explains: “Machine learning has a problem in that it has huge thresholds before it becomes something you can use. For instance, if your data is incohesive, you will need huge amounts of data to make any sense of it. If you have cohesive data, you don’t need that as much. So, we’re talking about incohesive data in terms of shipping, especially globally.

“However, when we analyse the data, using Google’s Tensorflow, we can see anomalies, we can see patterns and stuff, and a human mind will interpret that and make sense out of it and build algorithms. In a sense, we’re using machine learning to go half the way. Eventually, when we have millions and millions of transactions a month, then we can start using algorithms to do the prediction all the way.”

Cost vs. convenience

Customer data is just one side of the story, as Shipwallet also needs to grapple with all the available delivery companies and, of course, the costs. Shipwallet operates an “all shippers welcome” policy and leaves it to its algorithms to calculate the best options for the customer without bias toward any particular logistics company. It’s a consumer-orientated approach that favours transparency to gain trust in its business model, as the firm won’t be taking commissions or fees from the delivery firms.

What frustrates Zaleski is the fact that shipping companies don’t share data. If they did, he believes that they could deliver a whole new range of consumer-centric shipping options. Ones where you could cherry pick which company you use for what leg and for what route, just as Amazon does. He describes it as a “counterweight” to Amazon’s logistics excellence, with multiple partners in a delivery chain, instantly configurable as one logistical solution.

While this might sound like it’s just simplifying delivery, there is a sound economic case too. The shipping choices certainly appeal to customers to the point where they are less likely to abandon transactions when those delivery options appear at the final hurdle. It’s here where Shipwallet makes it money, with merchants paying €295 a month, and €0.10 per successful transaction. Zaleski admits it’s not a very hard sell, but the intention is to include everyone and, in the process, build a massive network.

“In the A-B tests that we’ve done with the merchants we have right now, we increased conversion by eight percent on the desktop and 15 percent on mobile. So that’s kind of a pretty huge deal. Going into a sales meeting with those numbers could kill any kind of argument.”

Such claims suggest that even with manual analysis to provide the insights, rather than machine learning going at full tilt, turning shipping into a convenience rather than a chore, certainly delivers.

This article was originally published on arstechnica.co.uk  and can be viewed in full

(0)(0)

Archive