The natural choice for search, advertising and everything else


I’ve often discussed the importance of contextual awareness in delivering a good mobile experience. This can range from understanding the user’s physical environment to knowledge of their brand preferences. Quite simply, the more you know about a user, the more likely you are to be able to provide an experience which meets their needs. Achieving that requires a balance between automated analysis of customer habits and enabling the user to explicitly indicate their own preferences.

As an aside, I’ve felt for some time the industry needs to do more to facilitate the latter. The development of simple tools for self-personalisation of the mobile experience is an under-addressed area. Companies like Sonopia are taking steps in this direction, but are still very much in the early stages. I’d be interested in hearing from anyone working on this kind of technology – please email me.

Building personalisation into the value chain is also a key part of our MEX conference agenda in London next week, 2nd – 3rd May.

However, the main subject for this article is the automated part of contextual awareness. If we imagine that every mobile interaction can be defined as a ‘mission’, we’re looking at techniques which enable us to understand that mission and help the user complete it as quickly as possible.

It is important to establish early on that these mobile missions differ from traditional computer-based search, which is often an exploratory exercise, where the availability of screen real estate, better input tools and time available allow for a broader, immersive search experience. Missions are about succinct, prioritised requests, measured on a timescale of seconds rather than minutes.

The simplest method for understanding missions is keyword search. You input a couple of words relating to what you’re looking for and the search engine tries to present you with relevant pages of information from the web. The Googles and ASKs of this world keep their search algorithms a closely guarded secret, but the basic concept involves looking for pages that contain the keywords, applying some page ranking and word association methodology and serving up a list of pages in order of relevance.

The initial result provides the user an overview of resources which might contain the actual information they are looking for. An appropriate analogy would be a hungry diner hovering over a a large buffet – they’ve defined that they are hungry by going to a restaurant, they’ve narrowed down their preference by selecting an establishment serving a particular cuisine, but they are still some distance from actually obtaining the piece of food they really want.

This is fine on a desktop PC, where the input and viewing mechanisms allow for easy exploration of this additional layer of information, but in the mobile environment, where navigation of each layer can take many precious seconds, it is a frustrating roadblock in the mission path.

A lot of companies specialising in mobile search are focusing on the results part of this process to improve the experience. The underlying methodology is basically the same as web search engines, but instead of serving up a list of web pages containing the keyword, they bump items like relevant ringtones, maps or stock quotes to the top of the results page. They’re not really doing a better job of understanding what the user is looking for, but they are working on the assumption that most mobile users prefer particular categories of results over others, i.e. a map instead of a text page describing a location or a ringtone instead of a link to a singer’s homepage.

Using this methodology, there is relatively little chance you will give the user the answer they are looking for first time. It might be somewhere within the results, but the user still has to search within the page to find what they’re looking for.

It is a cost effective way of processing billions of search queries and relatively easy to monetise by selling keyword-based advertising. However, it’s not a great way to help users complete their mobile missions.

I find the most effective way to obtain the information I want in the mobile environment is through 82ASK. It involves no graphics or Java downloads and is available on even the most basic mobile handsets. You simply text your question to 82275 (in the UK) and they send you back an answer.

Each question costs GBP 1.00 and it can take several minutes for an answer to arrive, but the experience is superior for several reasons. Firstly, the answer is almost always exactly what you’re looking for and, secondly, the time delay is asynchronous. The interaction method of SMS is perfect for the mobile environment, because you can quickly input a question and then put your phone away and forget about it until it beeps to alert you to the answer. When you’re walking down a street or standing on a train, this is a much better way to request information than the synchronous continuity of the browser environment.

As highlighted in my previous article on latency, even the most efficient browser or Java search engines can see their overall experience crippled by the poor performance of the client software, network delays or coverage issues. The result is that 82ASK often gives you the answer you’re looking for much more quickly than even the most advanced mobile search client; it does this not because it has the most impressive technology, but because it is designed from the ground-up for mobile.

It uses every natural advantage in the mobile world to increase the speed of mission completion – a familiar SMS interface to reduce ‘user induced’ latency, a native SMS client application already optimised for best performance, the SMS delivery channel to avoid packet data network delays and an integrated alert mechanism to tell users when the answer has arrived.

82ASK achieves this speed by combining absolutely simplicity on the client side with the world’s most complex information architecture at the back-end: the human brain. 82ASK employs a team of human researchers to vet most of the answers they send out. They do this for a very simple reason: there is no better mechanism for answering the questions posed by a human brain than another human brain.

I spoke recently with Paul Butcher, CTO of 82ASK and a man with extensive experience of the mobile industry from roles at SavaJe, Argogroup and Smartner. He described the importance of this human element in delivering a great experience, but also its long-term value for the development of computer-based search.

82ASK has answered millions of questions since its formation in 2003 and, in doing so, has built up an extensive database of how users ask questions, matched against a a list of exact, human-checked answers. The value of this should not be under-estimated.

Typically natural language processing systems work by annotating data. Each word or phrase is assigned additional meta data which enable the system to get better and better at working out exactly what the user wants. MIT has a web-based example of this called START.

There are a couple of ways you can add this meta data. One is to start with a small chunk of information as a template and teach a computer how to annotate this information. It is quite crude, but relatively inexpensive and quick.

The most accurate approach is the one taken by 82ASK, where almost everything in the database has been checked by a human researcher. The company’s business model, where it charges GBP 1.00 per question, has enabled it to fund the construction of this valuable resource for contextual understanding.

It can help with understanding users as individuals and more general language techniques. Let’s looks at an example question: “What’s the score in the blue’s game?” The chances of a traditional search engine understanding this are virtually nil.

However, a natural language processing system would be able to draw on several resources. It could look at the words ‘blue’s’ and ‘game’ and work out that in previous questions, these had been associated with Chelsea Football Club. It could also look back through the individual’s question history and see that the user had previously asked questions about this team and therefore draw a conclusion that answer they really wanted was: “What is the score in Chelsea FC’s current soccer match.”

That ability to personalise results based on a user’s previous behaviour and share knowledge of that behaviour to enhance the experience for other customers is incredibly valuable.

One of the long-term concerns for a company like 82ASK, a small organisation which has grown organically with little funding, is the competition from online search giants like Google and Yahoo. However, Google’s business model is built around its ability to automate search and adding a human element as 82ASK have done is counter to their ethos. 82ASk’s CTO Paul Butcher also believes that any company, no matter what its size, will struggle to replicate the valuable insights held in their database without spending several years building it up by hand – it’s something you simply can’t rush.

This has enormous implications for the mobile business. As more and more companies look to advertising to fund their business models, the importance of contextual awareness and natural language processing will increase rapidly.

Currently services like 82ASK (and its UK-based competitor AQA) are limited by the requirement to charge consumers about GBP 1.00 to cover the cost of answering questions. There are only so many people who can afford to use such a service on a regular basis. However, given their extensive understanding of customer behaviour and their direct channel to the mobile device, they are among the best placed companies in the world to fund their model through advertising.

I’ve always believed that mobile advertising needs to be about helping users complete their current mission rather than the distraction method traditionally employed by the advertising industry. In fact, this is one of 10 manifesto statements that form the heart of our MEX conference agenda on 2nd – 3rd May (the topic will be addressed by Antti Ohrling, co-founder of the advertising-funded MVNO Blyk). Companies like 82ASK have a unique capability to understand those missions and help users complete them.

I suspect it may not be long before the online giants who are serious about mobile start taking a serious look at what these small companies are doing.


4 Comments

Add yours
  1. 2
    kg

    you make several claims, but fail to provide enough data to back them up. Also, I think you fail to reference some of the latest improvements to mobile search offered by the searhc giants.

    An example of a claim for which some backup data is required: “When you’re walking down a street or standing on a train, this is a much better way to request information than the synchronous continuity of the browser environment.”

+ Leave a Comment