New Features On Avail SaaS

We are very excited to announce a slew of important new features on Avail SaaS. We will roll out the new features to all SaaS customers over a period of two weeks starting this week.

There are three new features which we have named the "Big Ones" as we deem that these will bring large benefits for our customers. The Big Ones are:

1. A/B Testing - Will let you easily set up A/B tests of Avail templates to find optimal settings.
   
2. Template Tracking - Allows you to track a lot of metrics on individual templates.
   
3. Improved Recommendations - An addition to the algorithm improves the quality of the recommendations.

I hope you enjoy the fruits of our hard labor!
/Henrik Schinzel
Co-founder & CTO Avail Intelligence

A/B Testing - Big One 1

A/B Testing will allow you to easily and rapidly set up A/B tests of templates. The purpose of A/B testing templates is to optimize the settings used. Example of tests that you might want to do:

• should the price rule on the shopping cart template state that the price should be less than 10, 15 or 20?
  
• should you should display three or four recommendations on the product detail page?
• should the third recommendation shown be a product on sale or a new release?

You can

• Create up to five variations of a template
• Assign a percentage to the variations that sets how much traffic the variation gets
  
• Set start and end date-time
• Edit the variations like a normal template

When the test finishes a report with the results of the tests will be generated that you can access in the Control Panel, and if you have set an e-mail address the report will be sent to you.

Some of the metrics available in the tests:

• Sales
• Conversion Rates
• Average price purchased recommendation
• Number of displays

Implementation Required 
Requires Template Tracking
Automatic

Template Tracking - Big One 2
Several customers have expressed the desire to have detailed metrics in Avail. We have heard you and are happy to announce Template Tracking. This will allow tracking individual templates on metrics such as

• Sales
• Number of clicks over number of views
• Number of views
• Average price of recommendation purchased from a template
• and more ...

The data will be visible in the Metrics area

with a lot of metrics to choose from

Implementation Required
The implementation required - assuming that metrics is already used - is to use the new methods:

• logClickedOn(ProductId, TrackingCode)
• logAddedToCart(ProductId, TrackingCode)

when a recommendation is clicked on or added to cart. The TrackingCode is returned with the recommendations.

Improved Recommendations - Big One 3
The quality of the recommendations has been be improved. To what degree will vary from customer to customer. What has been done is a more effective noise reduction. Noise is data that makes it more difficult for a computer to correctly classify something. If you have a product that has sold really well like the book The Da Vinci Code by Dan Brown this has noisy data and causes noise as it is connected to so many different people and products. The new algorithm more effectively reduces the noise from data that is not as relevant and therefore is able to produce better recommendations.

Implementation Required
Automatic

Summary of the Big Ones
To recap the three new important features:

• A/B Testing - Will let you easily set up A/B tests of Avail templates to find optimal settings.
• Template Tracking - Allows you to track a lot of metrics on individual templates.
• Improved Recommendations - An addition to the algorithm improves the quality of the recommendations.

Auto Categorize
Auto Categorize solves the problem of categorizing the recommendations and more importantly the fallback recommendations. If category filters are not used with recommendations requests on a product detail page and there are no real recommendations the fallback recommendations are a plain top-list with no relationship to the product currently viewed. For example, if viewing a goth metal album that no one has bought and no categories filters are used, the recommendations will be from the top list and typically be products like albums from Britney Spears and Justin Timberlake which will most likely not appeal to the goth metal crowd.

Auto Categorize applies to recommendations based on a single product, a set of products or a user. If used, the returned recommendations and fallback recommendations will automatically be submitted to a category filtering based on the categories of the input. In the example above, if the goth metal album belongs to the categories Goth and CD the returned recommendations will be from the top-list but within the categories Goth AND CD and thus much more relevant.

Logic
If the input is one product

• The returned recommendations will belong to all of the categories of the argument product.
• An AND-relationship between the categories of the argument product.

If the input is several products

• The returned recommendations will belong to all of the categories of argument product 1 or all of the categories of argument product 2 and so on.
• An AND-relationship between the categories of each argument product and an OR-relationship between these.

If the input is a user

• The logic of several product is applied to the purchases of the argument user.

Implementation Required
Automatic

Cart Recommendations
To optimize the quality of the shopping cart recommendations Avail now supports cart recommendations that is intended to catch cart specific behavior such as recommending a toner with a printer.

Implementation Required
When uploading transaction history there is a new optional column, order id. There is a new method getCartPredictions that returns shopping cart recommendations. To incrementally add more order data the method logPurchase now accepts an OrderId as optional argument.

Maximum Number of Recommendations from a Template
With the new release it will be possible to set the maximum number of recommendations a template returns no matter how many recommendations the subtemplates are set to return.

The reason for this feature is that in some implementations of Avail a limitation on the number of recommendations has been hard coded in the web layer. This hampers the merchandiser in that he or she cannot readily change the number of products that are returned from a template using the Control Panel. To change the number of recommendations displayed the merchandiser needs to request resources from the techies.
By adding this feature there is no need to implement a hard coded limitation on the number of recommendations rendered in the web layer. The merchandiser can control the number of recommendations returned without bothering the techies. Also, by adding this feature it is one less thing for the techies to do when implementing Avail if this behavior is desired.

Complex Templates
This feature also enables more powerful merchandising. For example, we can create a template that is set to return a maximum of three recommendations that has three subtemplates:

• Subtemplate 1 - returns three recommendations with Price < 5, no fallback recommendations
• Subtemplate 2 - returns three recommendation with Price < 10, no fallback recommendations
• Subtemplate 3 - returns three recommendations with Price < 15, use fallback recommendations

The result is a template that generates three recommendations that are as cheap as possible. This as achieved as the first subtemplate generates as many real recommendations as possible that cost less than 5. If there is not enough real recommendations that cost less than 5, real recommendation that cost less than 10 will be added. And lastly if it will add real recommendations that cost less than 15 and use fallback if necessary. This is just one example with what you can do.

Implementation Required
Automatic

Category Names
Avail now supports category names. If category names are uploaded these will be displayed in the Control Panel instead of category identifiers.

Implementation Required 
New upload: CategoryNames

Find My Twinsumer
A new method accepts a user identifier as argument and returns the user with the most similar behavioral pattern. This feature does not stand on its own but is a powerful feature to any e-commerce website, because it allows visitors to the site to quickly identify "people like me" (anonymously) and then implicitly ask these "Twinsumers" for product-related information and recommendations.
The way a retailer can use this function is e.g. to publish the following functionality on any of their website pages. "Find My Twinsumers (people like me), and then:

• get a list of products these Twinsumers have purchased (optional: in this category)
• get a list of products these Twinsumers have written a review about (optional: in this category)
• get a list of products these Twinsumers have given a rating higher than X (optional: in this category)
• get a list of products these Twinsumers would probably also have bought, had they come to our site today (optional: in this category)
• let me ask them a question (via anonymous email/messages that the site can distribute without disclosing personal email addresses)
• let me send them "invites" to any community of choice like Facebook, Twitter, LinkedIn, etc (via anonymous email the retailer distributes)

Implementation Required
New method: getTwinsumers

Clickstream Tracking & Recommendations
Avail now automatically collects data on which products visitors have clicked on and enables recommendations on this clickstream data. The products clicked on by a user is collected with getTemplateClickstreamPredictions and logClickedOn. The purpose of this feature is to enable more rapid integrations.

Implementation Required
Requires Template Tracking.
New method: getClickstreamPredictions

Multiple Currencies
Metrics now has multi currency support. With this it is possible to log purchases in different currencies. The purpose is to facilitate integrations. Note, there is no automatic currency conversion.

Implementation Required
New optional argument in logPurchase: logPurchase(..., [Currency], ...)

Expanded Identifiers (AlphNumString)
The data type for user, product and category identifiers (AlphNumString) now supports three more characters:

• at sign (@)
• full stop (.)
  
• hyphen-minus (-)

Implementation Required
Automatic

Recommendation Reinforcement - Beta
This is a beta feature which is disabled by default and will be evaluated over next few months.
Some products trigger customers to buy more than others. The purpose of this feature is to give a bump to the prediction score for products that have a high inherent conversion rate. This will make these high converters more likely to be recommended.
This is especially useful for products that are not so easily found. Some products are easily found because:

• they are prompted - on the first page, on category landing pages, on banners and so on
• rank high in the search results
  
• are famous on their own accord such as the iPod
  

There typically exists a set of products that are high-converters that has not been caught by the merchandisers and thus are not prompted. But when a user stumbles over these high-converters these are more likely to trigger a sale than the average product. By giving these high-converters a bump in the predictions score, Avail will automatically promote them in the recommendations.
Logic
The increase of the prediction score is the ratio: the number of purchases to the number of clicks. The maximum increase in prediction score is 20%. This logic is likely to change over time.

Implementation Required
Requires Template Tracking
Automatic