The Science of Advertising: Targeting you with the things you need

This article first appeared in the May 2008 issue of it magazine. This is part of the Folk Technology series that I write every month.












Do you know how advertisers plan their Campaigns? They want to target the right audience with the right product in the right way. To do that they study audience behaviour. Read on....

Do you think you are affected by advertisements? Do you recognise the companies with the slogans, ‘Information technology, today, tomorrow’, ‘Just do it’, ‘Connecting people’, ‘Fly the good times’, ‘Wherever you go our network follows’, ‘Connecting India’, ‘The nation banks on us’, and ‘Taste the thunder’? If you recall even some of these, then you have to agree that advertisements do have some impact on you.

Companies spend crores of rupees on advertising because they know consumers are affected by advertising. The goal of advertising is to influence the purchasing behaviour of as large a population of consumers as possible. So how do companies plan their advertising expenditure?

Ad pricing

On traditional advertising media like TV, radio, print and billboards, the pricing is based on how many eyeballs or ears the advertisement captures. This is the cost-per-view model. For TV or radio, ad rates go up during a cricket match, because there will be a larger audience. Print ad rates are also based on the total readership of the edition.

In the world of online advertising, pricing is more complicated. Not only do you have cost per view options, but could also have cost per click (based on the number of times the ad is clicked on); or, if it is an online merchandising site, then it is based on ‘cost per action’—the number of times a product is purchased. The pricing model can be fixed or dynamic. The most popular dynamic model is the search-term-based advertising that has been popularised by Google. An advertiser has to bid for a keyword in an online auction. Popular keywords like ‘job’ go for many thousands of rupees; less popular keywords like, let's say, ‘lalu’ go very cheap. Have you noticed the sponsored links to the right of your Google search results? The advertiser showing up at the number one spot bid the highest for the keyword you used in your search; those who bid less are ranked below this. Every time you click a sponsored link, the advertiser has to pay the amount that they bid to Google.

We are watching you: We know what you need

Advertisers today try to collect as much information about you as they can—they can cash in on this knowledge by serving you ads tailored to your preferences or positioned to take advantage of your plans, as determined by your online activities. For example, when you search for a tandoori restaurant, the search engine dishes out related sponsored ads based on your search ‘interest’—your apparent plans to eat out. Similarly, when you read an e-mail about summer vacation plans in Goa, your friendly e-mail provider serves you hotel and airline booking ads, hoping to capture a booking or sale for their advertisers. This is true of SMS messages too; every SMS you send is automatically analysed to learn your interests, and to serve you relevant advertisements.

Getting icons to advertise

How do advertisers convince you to buy their products? One approach is to get a social icon—for example, Shahrukh Khan or Preity Zinta—to be their brand ambassador; you would tend to buy products thus advertised, because of your admiration for the person endorsing the product. But this is a pretty expensive proposition, because brand ambassadors come at a premium.

Social network analysis

So what's the next best thing to brand ambassadors? Analysing your behaviour and trying to find out which of your friends influence your purchase decisions! Man is a social animal and that fact is only now beginning to get exploited by advertisers—they try to influence you through the people whom you admire, with the added advantage of minimal expenditure.

Do you know me? Well, the chances are that you and I don’t know each other directly but have mutual friends. I may have a friend who has another friend, and that person is your friend—so you and I are within 3 degrees of separation. Experiments conducted in a limited setting show that, on average, only 6 degrees of separation exist between any two people. Others have extended those results to imply that anybody in this world is within 6 degrees of separation from anyone else.

If you have an account on a social networking site like Orkut, LinkedIn, BigAdda, or Facebook, then you can experience this first hand—you may directly know only 30 to 40 people, but count your contacts' contacts and further, and you can see the number of people grows exponentially with each degree of separation.

A social graph is a mapping of every person in this world, and how we are related in a given context. For example, if it is known that I am your school friend, and that both of us like cricket, then maybe only the more outgoing between us needs to be targeted—that way, if you are influenced to buy a Tendulkar autographed cricket bat, then most likely, I will also buy one, following your example.

Advertising to groups

Advertisers have long known that referral is the best way to influence consumer behaviour. When someone you know recommends a product, you are likely to be influenced by that. Today, on many social networking sites, people with shared interests form communities. Facebook, for example, wants to exploit this by targeting user communities. From your interactions, it becomes amply clear who your friends are and what your interests are, so ads that address your needs are served to you. That means that if you belong to, say, a bathroom singers group, then ads for music CDs, iPods and other music-related items would be thrown at you. The more influential in your group are enticed into buying first, and then recommending it to others.

Telecom graphs

Telecom companies are beginning to exploit ‘Who’s calling who’ graphs of millions of users, each with their own characteristics. Let's say you make an average of 25 calls per day. Some of these are to people you call regularly. Telecom companies analyse your calling behaviour, and that of your group, to decide calling and marketing schemes for you. For example, if the majority of your friends' group subscribes to caller tunes, and you don’t, then that automatically makes you a candidate for marketing caller tunes. After all, birds of the same feather flock together! Telecom companies keep track of group leaders. If you are always the first in your group to try out new offerings, and others follow your lead, then your clout can be exploited to reach your friends.

Mathematical modelling

For mathematicians and computer scientists, this is an interesting problem - second guessing consumer behaviour, based on past actions and current friends. Companies who have the customer data usually do not have the inhouse capability to analyse it, so analysis is outsourced to companies that specialise in Social Network Analysis.

Consumers get irritated if they are served too many irrelevant ads, so the idea is to serve them only the most relevant ones, and not bore them with ads that they are unlikely to follow up on. But this is an enormous problem—can you imagine the size of the telecom graph of a typical service provider in India? Even if we take just 2 million active users, that graph is huge! Analysing this graph to find potential leaders is really difficult.

Telecom companies are beginning to exploit ‘Who’s calling who’ graphs of millions of users, each with their own characteristics, to analyse their calling behaviour, and that of their group's, to decide calling and marketing schemes.


Forming graphs and identifying key players

This could be a ‘Who’s calling who’ graph, which shows how a set of people are connected to each other. The telecom company will be most interested in Juhi and Rajat, who seem to be central characters in their groups. They seem to be central links holding together their respective groups. Advertisers would like to use them for targeted ads, and to influence others. The telecom company will go out of the way to keep such central characters within its fold.

Also of interest are Sid and Tara, who are connections to outside groups; by targeting them, new groups can be reached. If Juhi’s group is with Vodafone and Rajat’s group with Idea, then for Vodafone to get new customers, it has to target Sid to get his Idea friends along. It also means that Sid is a candidate for leaving Vodafone and going to Idea, so either way, he needs special handling.

Telecom graphs can be huge, with millions of connections. Automatically drawing and analysing such large graphs to find the key players is a challenge for computer scientists.

Forming groups


Advertisers form connectivity graphs of people based on who they interact with. Then groups within them are identified, based on specific likings that are learnt based on the behaviour of the individuals. For example, if a person chats with his connections about cricket most of the time, if his site has pictures of cricketers, if all the YouTube videos he has tagged are cricket clippings, or he bought the Indian team T-Shirt online... that person must be a cricket fan. Learning these likings is challenging, and involves special software to track activities and map each activity to a group, using a set of rules. Rules can be hand-coded, or they can be learnt based on a limited set of examples. Once it is known who likes what, they can be served relevant ads. Though they may sometimes make mistakes and target us with wrong products, analysts are learning fast and trying to keep up with our changing needs, interests and likes.

(Part of the monthly Folk Technology Series that I write for i.t. magazine)