Riddle me this! Where can French, Italians, and Germans all agree?

Well, apparently, its in their tendency to pair wine with spicy Chinese food.

How did I come up with such a farfetched notion? The answer to “how” is “easily” when I started with a WP7 app like VinoMatch’s Mobile Sommelier that was instrumented with Runtime Intelligence for Windows Phone. But let me back-up a little.

On Nov 4, 2010 MSFT and LG announced that LG Owners Get Free Access to Popular Windows Phone 7 Applications. You can read more about the 10 premium apps and another 10 LG-specific apps on LG’s Facebook page here. They call the program “hAPPiness!”

Anyhow, it turns out that all 10 of the premium apps were instrumented with Runtime Intelligence and have been sending analytics since the January launch of the program.

Last week, Microsoft invited lead developers from all ten development teams on campus for three days to gather feedback as well as to hold some specialized training on a variety of technical topics – and I was fortunate enough to be invited to lead a discussion on best practices for incorporating mobile analytics into the development process. In support of this presentation, I was given permission to look into the runtime data that these 10 apps had been collecting to see what I could see.

The 10 premium apps that are free from LG through March 10th are:

1. Cocktail Flow by Gergely Orosz
2. Color Sprouts by Jarek Kowalski
3. Colorize by Kitron Software
4. Doodle God by JoyBits Ltd. (a Top 100 Paid App)
5. Envision by Dotnetfactory
6. Krashlander by Farseer Games (a Top 100 Paid App)
7. Mobile Sommelier by VinoMatch
8. Mr. Hat and the Magic Cube by BRAVO game studios
9. Talking Ragdoll by Spritehand LLC (a Top 100 Paid App)
10. Weave by Seles Games (a Top 100 Paid App)

While I can’t share the entire presentation here, VinoMatch and Gergely Orosz were kind enough to give me permission to include some of their mobile analytics in this discussion. I want to highlight two powerful analytical techniques using their real-world data.

Conversion feature analysis

Embedded within most apps are a few keystone features that drive conversion (from eval to paid, prospect to customer, fan to fanatic) and Cocktail Flow has one under development, Shopping Assistant. It does not yet connect directly to a store and, as a nascent feature, is not yet a centerpiece in their UX design. Here is what I was able to deduce from the Runtime Intelligence data:

• Only 5.5% of all user sessions used the shopping assistant feature.
• Of those, 49% used shopping assistant only once inside their session, and 51% used shopping assistant multiple times. Of those 51%, they used it for an average of 2.9 times per session. What were they doing?

With Runtime Intelligence, I am able to map out the full usage trail, feature-by-feature, session-by-session, and so I am also able to visualize:

• Where in the sequence of features of each user session shopping assistance was being called.
• And when it was used, was it a part of a “one use” session or one of those sessions where the user called the shopping assistant multiple times in a single session.

I can also see that the overwhelming number of Cocktail Flow users using this feature rarely used more than 10 features in their session and that virtually all lengthy sessions (taking more than 14 actions) were always churning (going back to the shopping assistant feature multiple times).

When this feature is functionally complete to the developer’s satisfaction, they will now have benchmarks to measure feature adoption (increase it over 5.5%) and session behaviors (when, how often, and under what circumstances the feature is used) all relating to what may ultimately be their most lucrative app feature.

Mutli-variable AB testing and cross-domain user profiling


In the Cocktail Flow example, we focused on a single feature. With Mobile Sommelier, we are going to add a few more dimensions.

The following graphic shows, as a relative percentage, how often ten selected features inside the Mobile Sommelier application were used over a 6 day period.

What we can see immediately is that there is very little variability over time. The population of users generally uses the application in the same way over time.

What can we do with this perspective?

If we were to release a new version of this App, we could immediately see any meta shift in behavior across the two versions (Runtime Intelligence will automatically roll-up multiple versions and then break them out when asked).

What CAN’T we do?

While we can see how often the “ShowTasteCard” feature has been used, we CAN’T readily determine which wine/food pairing users are most (or least) interested in. We can’t analyze custom, application specific data.

Luckily, Runtime Intelligence CAN grab this data – with no programming – all post-compile. For example, this (very busy) chart shows the how often (by percentage) users want to pair wine with specific foods (like BBQ or spicy Chinese) AND it is further broken out by country.

Of course, this is too busy to read in its current form, but the trend lines do immediately show that while all users may use the “ShowTasteCard” with generally the same frequency (as indicated in the last graphic), it is clear that they use the feature DIFFERENTLY, e.g. they care about different topics. Culture counts!

Let’s drill down.

Focusing in on just three countries (France, Germany, and Italy) and focusing on just a few of the food options (from AntiPasto through Chocolate Cake), we can see genuine differences in user behavior and preferences.

Perhaps it is not surprising that French are more interested in Caviar, Germans in apple pie, and the Italians in antipasto.

On the other hand, maybe we were not expecting Italians to show such a strong interest/preference for BBQ. Could this data point to a way to sell Zinfandel (a US export based on a grape that originated in Italy) into the Italian market? – as the US wine for US food (BBQ).

Some general best practices

At the end of the session, the developers generally agreed on the following developer patterns:

• Instrument key usage and application milestones (Not just clicks and page views)
• Capture user selections and preferences (Not just selection and preference setting events)

Other conventions and best practices

• Implement exception reporting
• Integrate runtime intelligence into your CRM, ALM, and Marketplace repositories

Application Analytics are not your older brother’s web clone for the phone!

These have been just a few small examples of the rich set of analytics that are now available (both on and off of the phone). So, if you think Runtime Intelligence is just another web analytic clone for the phone, think again – Runtime Intelligence provides application analytics – not web analytics (see my earlier blog Application analytics: a new game brings new rules).

In a nutshell, this means that these analytics include:

• Custom data (including complex objects)
• Session, feature, event, and method-level precision (not just event)
• Exception reporting (unhandled, handled, and thrown)
• Support for both Silverlight and XNA
• Opt-in policy enforcement, SSL transmission, and caching
• And were implemented using post-compile injection eliminating the requirement to change (or even recompile) source code.

Anyhow, I am ALWAYS eager to hear from developers who have employed analytics in cool and innovative ways – PLEASE reach out and share the wealth! Thanks