| Author: | Date: May 5, 2008
In-Brief: In response to the comment Medio Systems’ co-founder Michael “Luni” Libes recently posted on MSG I tested Verizon’s Get It Now service. My analysis reveals what Verizon delivers – and doesn’t.Mobile search is about allowing users to find what they already know they want. It’s important to get search right, but even power-search does little to encourage impulse buying behavior. That’s where recommendations come in. Unlike search results, recommendations suggest similar content to users based on their search queries. It is also possible to base recommendations on other factors such as browsing patterns, purchasing habits or intelligent segmentation – but the mobile industry, for the most part, isn’t there yet.
What the industry can do is match content recommendations with keywords. At least, that’s what I assumed when I began my study of mobile search/recommendation on the T-Mobile USA and Verizon portals. However, I am forced to conclude that the industry has a long way to go before it properly implements content recommendations. To be fair, the industry has to get search right first, (which it is currently struggling to do), but I believe the industry should tackle these issues in parallel as the technologies are intertwined and mutually beneficial.
In response to my earlier post Luni correctly pointed out that the searches were performed from within Verizon’s Vcast Music application. He stated that “for a better experience on a Verizon phone, subscribers (today) must download the “Get It Now Search” application, which will search not only the VCast Music catalog, but also the ringtones, ringback tones, wallpaper, games, and applications catalogs, all from a single search box.”
Curious to see if there would be a vast difference in the quality of the results I downloaded the Get It Now app to my handset and conducted mobile searches for a variety of content, the results of which I have documented in this post.
To be clear, the Get It Now experience – which is powered my Medio – is much improved over the Vcast service. However, I am still concerned that the filtering technology, which delivers 200+ items in response to a music search for “Frank Sinatra”, presents the user with too much choice.
It’s a lot of results to wade through and negatively impacts the overall end-user experience. If the industry is serious about delivering results within 2-3 clicks, then filtering should be fine-tuned to do just that. I should like to add that it is not a problem for Medio or for Verizon; it is an issue across many mobile search services available today.Recommendation is a different matter. Here I see tremendous unfulfilled potential on the part of Verizon (and its partner Medio) to deliver a much more satisfactory user experience. Case in point: My search for the music group “Black Eyed Peas”. Medio delivered an abundance of search results (all well and good), but it also recommended tracks by Il Divo (at the bottom of the image under Recommendations – Suavemente). A closer examination raises some serious questions about the linkage.
Black Eyed Peas is typically classified in the following genres: Rap, Pop-Rap, Club/Dance and Alternative Rap. Il Divo, on the other hand, is typically classified in the genres: Vocal, Classical Pop, Classical and Vocal Pop.This (mis)match leads me to believe that these recommendations are the result of a clash between Verizon’s tagging of the catalog and Medio’s categorization of music. In fact, I would hedge a bet that the service maps the music against all the other music in the available Verizon catalog and not the music in an organized, comprehensive ontology.
It’s acceptable to hand-code recommendations, but the human factor here means more mistakes, less uniformity in categories and categorizations, and a range of issues around scalability and speed.In contrast, automated systems avoid many of these errors to deliver superior recommendations. (As one of the co-founders of a recommendation engine company, Caboodle Networks, which I later sold to a mobile search platform provider, we consciously chose to develop an egalitarian and automated approach to recommendations for precisely these reasons.)
And finally – in view of the excitement around approaches such as Yahoo’s Search Assist that enable users to input search queries faster because they complete the query for them based on a few letters – I decided to take a hard look at what Verizon offers. My take: The mobile operator is leaving money on the table by not making it clear how the partial word match works – and how users can accept the suggestion in the first place.
As the example shows I can see the word “Madonna”being completed for me in the search box, but I cannot click it on to accept it as my search query. Instead, I get “No matches found. Please try again.” It’s not apparently obvious that in order to activate the word completion that you have to click on a right direction key, before selecting OK. You figure…Peggy adds:
Thanks for deep-dive Eric. BTW: I reported
during CTIA on Medio’s “predictionary” technology
, which is built to tackle this problem. In fact, Medio hinted it went one better than word completion or word match because it could also somehow figure the user segmentation/profile into the equation. As Medio CEO Brian Lent explained: If the system somehow determines that a user is more a prosumer than a consumer, then the act of typing in only an “n” triggers the predictionary to suggest “news” and not “Nelly Furtado” because prosumers as a group are more likely to search for information rather than ringtones. I’m not sure where Medio is on this – but I’ll certainly raise the question. By coincidence, I have a briefing tonight with John Kim, Medio’s VP, Product Management and Product Marketing.