Well, using actual purchases is the best bet, but also the least targeted. With the scale of Amazon the biggest win is in providing the best recommendations If you're a small shop, on the other hand, you probably want a slight bias towards higher margin products because those do the most for your bottom line. Further, you'll probably want to highlight product X as part of a promotion.
So for a small shop, you probably want to have some sort of transaction-based recommendations as the base layer, but add the ability to seed those recommendations with manually specified items as well. As a half-way, seeding each product category with specified items would be another option. So when it came time to list, you'd check for any related items specific to the current item, then items based on multi-item transactions containing the current item, and finally any items bound to the current item's category. Obviously you'd stop once you found a sufficient number. As the scale increases, specific product bindings become both less valuable and impossible to maintain, and category stuff drifts to irrelevance very quickly after a new product is introduced, so you're left with only the transaction-based stuff. Also worth mentioning that if you're doing "also bought", that's quite different from "similar items", so your product and category associations need to reflect the right one. E.g. if I'm shopping for a new game console "also bought" should contain a second controller and some AV cables, while "similar items" should contain other versions/packages of the same console and perhaps other consoles as well. So make sure what you're building is well understood by all the stakeholders, including the eventual site visitors. cheers, barneyb On Tue, Apr 7, 2009 at 10:07 PM, Mike Kear <[email protected]> wrote: > > So Phillip, you say that those "people who bought this also bought > ..." are based on ACTUAL purchases? > > I would have thought that would be simple on a high traffic site like > Amazon, but I was thinking more about the mere mortals in the > ecommerce world? Sites that have a lot less traffic than Amazon. > How do they do this cross-selling thing? Still based only on actual > purchases? > > How would you set up such a thing in a new site? > > Cheers > Mike Kear > Windsor, NSW, Australia > Adobe Certified Advanced ColdFusion Developer > AFP Webworks > http://afpwebworks.com > ColdFusion, PHP, ASP, ASP.NET hosting from AUD$15/month > > > > On Wed, Apr 8, 2009 at 1:47 PM, Phillip Vector > <[email protected]> wrote: >> >> 1) Find someone who bought product X. >> 2) Find an order that person made. >> 3) Pick a random item from that order. >> 4) ????? >> 5) Profit. >> >> On Tue, Apr 7, 2009 at 5:21 PM, Mike Kear <[email protected]> wrote: >>> >>> How do e-commerce sites do those "people who bought this also bought >>> this .." selections for cross-selling? I have been doing this >>> with a lot of work on the categories in my databases, so that when >>> someone selects one item, they are shown several other (randomly >>> selected) items in the same category. >>> >>> Is that the best way to do it? I guess a vast site like Amazon has >>> enough traffic to very quickly assemble enough transactions that they >>> can actually HAVE a 'people who bought this also bought that' because >>> for even very new items, its highly likely that someone will have >>> bought it. But what about lesser sites where new items might not make >>> sales for a day or two? How do they assemble this information? >>> >>> Is it in the categorising? Product keywords? A random sele > > ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~| Adobe® ColdFusion® 8 software 8 is the most important and dramatic release to date Get the Free Trial http://ad.doubleclick.net/clk;207172674;29440083;f Archive: http://www.houseoffusion.com/groups/cf-talk/message.cfm/messageid:321437 Subscription: http://www.houseoffusion.com/groups/cf-talk/subscribe.cfm Unsubscribe: http://www.houseoffusion.com/cf_lists/unsubscribe.cfm?user=89.70.4
