Over the last few years a number of “personalization” engines have entered the market – and rightfully so. Personalization is the holy grail of merchandising, the magical ability to present the individual, with that product or products that matches his or her needs perfectly.
The problem is that information isn’t distributed equally between the actors in this play. The shopper in question knows a lot more about themselves than the site. The e-commerce site, only knows what it can know, and most times, that is not very much about an individual shopper, even if they force you to login (ala Amazon).
So with this asymmetrical information distribution, the goal is not so much perfection, as it is to best optimize available information.
Amazon claims they have recommendations just for me...
So with this asymmetrical information distribution, the goal is not so much perfection, as it is to best optimize available information.
Before we dive further into the concept of personalization – lets take a quick step back, and talk about ‘merchandising’. Most retailers have a product/product-line that draws in the crowds, but where their margins are very slim. Camera’s for most electronics retailers, for example. These retailers are hoping to sell you accessories, warranties, professional services, whatever they can around the primary product in order to actually realize a profit in the transaction.
Many times, this is encouraged, and achieved via a simple Product X goes well with Product Y type relationship. A good example of this is someone like SamAsh. See the Things You’ll Want section below.
Samash - Things you'll get VS Things you'll want!
This helps the customer surmount the major irritation of waiting excitedly to receive his acoustic-electric guitar only to find that he doesn’t know how to tune it, or that he needs an Amp to sound like Hendrix… stuff that any salesperson in a store would’ve told him right away.
The problem, even before this one however in our story line described above, would have been finding the right product (in this case a guitar, but could as easily have been a camera). That’s where merchandising zones like “People who viewed this, also viewed” (Overstock.com) – or “People who viewed this ultimately bought…” (Amazon.com)
Overstock.com - People who viewed this, also viewed that...
So, thus far, the story goes - a person entered the store, wandered the aisles, answered a few questions the salesperson asked, and voila, found a product. While looking at this product, “we” mentioned a few other products she might be interested in, and in doing so, helped her find the product she really wanted.
Then once we’d found her product, “we” showed her all the other stuff she’d need in order to get the optimal usage out of it.
Twice, in that interaction, we “suggested” products to her. How though, do we know what products to best suggest to someone? Now - we’re at the crux of what personalization does.
Let’s go back to the store metaphor. Most times, when someone walks into a store, a salesman can ask a few questions, and assuming he knows his stuff (I know, that is indeed, someting increasingly rare these days), he should be able to lead you to a product of your choice. If he’s been around a while and he knows your name then he might also recall what other equipment you have purchased, or own, and suggest accessories accordingly.
Why can’t e-commerce stores do this? If we remove the intangibles that an e-commerce store can never know (what the person looks like, what they sound like, what their body language is saying etc.) we are left with several tangible data points that can be used here – past purchases, intent, location, budget, wish lists, other behavior patterns on the website, other people similar to this person etc.
So the idea of personalization is to optimize what we know about the person based on the tangibles, and then present to them a product that we think is best suited for them.
This is the second problem with personalization. Let’s say it’s an ideal world, and I have a ton of knowledge about the person browsing my site – I know her age, her location, the products she’s bought in the past, the average order of her past orders, what price bucket she usually picks while navigating my site (cheapest, middle, high), which shipping option she chooses (needs it today, tomorrow, or no rush), so on and so forth (and there’s a LOT of stuff in the forth part that we haven’t discussed yet)
So now I know all this – let’s say she picks products that are usually in the middle price range. When suggesting a product to her on a Product Detail Page (like SamAsh’s above) can I take this into account? Do I know for every product that I have whether it is a low, mid or high price range product? If I wanted to calculate this, what other products would I compare it against? Do I compare the chromatic tuner to all other chromatic tuners? Or to all other guitar tuners? Or to all tuners including tuning forks?
Does her gender matter? Do I know if women have picked a certain type of tuner over another, all other factors being the same?
As you can see, personalization has multiple aspects to it – we’ve briefly broached the following
- knowing as much as possible about the customer,
- knowing as much as possible about the products
- knowing as much as possible about how customers interact with the products
Whatever data you have on the three points above, then needs to be mangled into a complex algorithm to try and determine the most personalized suggestion that the store can offer. This is where personalization engines started. They were in depth algorithmic approaches to the problem.
However, they soon noticed that retail stores actually had very little mined data about the third point – customers interacting with and purchasing products. So personalization engines extended themselves to listener’s as well – they now sit there, and record all these interactions. Purchases, clicks, links, emails etc. They are collecting and mining this information based on what data the algorithm needs to make decisions. Based on cookies, and logins they are also gathering information on the first point above – the customer.
For the middle point – data on the products, they rely on the retailer.
So now, we can guess at what the the ‘secret sauce’ is? It is the ability of personalization engines to mine behavioral, and purchasing patterns for data, and their ability to combine this with the daily feed on products that they receive, run it through an algorithm in order to spit out the possible best fit for an individual.
As Moore’s Law continues its onslaught, and as individuals continue to trade privacy for better services, the rapid rise of personalization will continue. This is great for both customers (who will get better suggestions) and vendors (who will, or so the theory goes, sell more products). Personalization has a long way to go (insert Semantic Web prophecy here) but it is definitely an area where more and more resources will find their way.
