Starting with tools in the public domain and datasets that the Election Commission itself provides, data analysts can build huge datasets that a party can use for targeted advertising. With just the information on the electoral roll that is publicly available, analysts can get a person’s name, his or her father’s name, location (what booth they vote at), voter ID number and age.

In many large states of India such as Uttar Pradesh and Bihar, the caste of over 70% of the people can be determined simply by analysing their names. With a relatively simple code, a political party can identify the caste and religion of a large number of voters by analysing their last names (surnames), but there are some major complications in the process.

Different communities in different parts of a state sometimes use the same last name and some last names are not easy to classify. Certain last names like Yadav, Paswan, Singh, Agrawal and Shukla are easy to classify in most states but some are used by multiple castes in different parts of a state. For example, in Bihar, both Bhumihars and Pasis use the last name “Choudhary”.

The solution to this is relatively easy for a political party. The party has a resource that very few entities do – free labour.

There are lakhs of karyakartas who are thrilled to help out with anything that has them doing something other than simply distributing pamphlets and putting up flags. They can be tasked with manually classifying a few names in each constituency so that an algorithm can assign a caste to the entire dataset using their input. Local karyakartas usually know each family’s caste in their area, so this exercise wouldn’t be a difficult one for them.

The BJP had an edge when it came to volunteer engagement for such classification but the edge is slowly evaporating. A few years ago, the BJP was the only one with the cadre strength to be able to undertake massive initiatives like appointing “panna pramukhs”.

A panna pramukh was an individual karyakarta who was responsible for outreach and campaign amongst all the voters listed on one double-sided page of the electoral roll that was assigned to them, and which on average consists of sixty names. The Congress too has caught on to the technique and has started the practice of appointing such “page in-charges”.

Experimenting with the strategy during the elections in Punjab in early 2017, the party also used it during its campaign for the 2017 Gujarat legislative assembly elections.

Once a party has most of the voters’ castes and religions mapped against their names in the electoral roll through a combination of technological tools and manual labour, the next step is to include phone numbers. Data protection laws in India are basically non-existent for data other than financial, health and some other sensitive ones. This means that obtaining phone numbers doesn’t require the intervention of high-level employees of telecom companies. It is available with data brokers who have probably obtained it from low-level operatives of telecom companies or from SIM card dealers.

Other details that could play a role in determining voting behaviour can also be added to the database against each name. One such detail would be socio-economic status, which is a major determinant of the kind of messages people would respond to and the way they’d choose whom to vote for. A lot of datasets, like land records, BPL lists, surveys by the National Sample Survey Organisation (NSSO) and census data, can provide this information.

Much simpler proxies are also available, such as electricity bills, which can be used to judge a person’s economic status. A person’s electricity bill is directly correlated with their income since people with high income often consume more electricity with gadgets and appliances like air conditioners and geysers and hence have higher bills, while the poor have relatively fewer electrical appliances to consume electricity.

The extremely poor voters can be identified through the Below Poverty Line (BPL) list, so electricity bills are really useful for segregating the rich from the middle class, especially in dense urban pockets. With this data in hand, parties can create groups consisting of specific types of voters. To most casual observers, the groups might seem to be of just random people living in a specific area, but in reality they would consist of people selected on the basis of their age, religion, caste and economic status. This would allow for targeted messaging and micro-targeting in a way that hasn’t happened in any other country in the world.

As an example of what such groups could accomplish for political parties, let’s take the case of Uttar Pradesh.

Assume that a party (Party 1) already has upper caste votes in Uttar Pradesh but requires the support of other caste groups to win an election. The OBC population of Uttar Pradesh comprises approximately 11% Yadavs and 31% others.

Let’s say the Yadavs strongly identify with another party (Party 2) and they’re a group that is unlikely to change its vote. The non-Yadav OBCs on the other hand could be convinced to vote for Party 1 if it could provide them with an effective reason to do so.

The simplest way to achieve this goal would be for Party 1 to instill a sense of victimhood in the non-Yadav voters and alienate them from the Yadav voters. WhatsApp groups can prove to be an effective tool to accomplish this.

By gathering adequate data, analysts can, for instance, create groups that consist of only non-Yadav OBCs between the age groups of eighteen and forty, with low to middle socio-economic status. A close approximation of this group can also be targeted through Facebook and other social media platforms through a little-known feature of Facebook that allows for targeted advertising through the use of lists consisting of email addresses or phone numbers.

If messages stating how the Yadavs had cornered all benefits of reservation in educational institutions and jobs, thereby depriving other OBC groups, were broadcast repeatedly in different forms over such a group, there would eventually be an observable impact. Over a period of two to three months, the arguments sent over WhatsApp would become a part of people’s daily conversations and they would begin to get convinced that their rightful share of benefits had been taken away by the Yadavs.

After such priming, they could easily be influenced to hold the opinion that this “injustice” would not be rectified unless they were to teach Party 2 a lesson by voting against it, leading to an increase in the chances of Party 1 winning.

Cambridge Analytica probably couldn’t even dream of this level of targeted advertising. And yet in India, it isn’t just possible but is actively being done. This is also where fake news comes into the picture. Sometimes, the actual facts are just not enough to convince a group that it is the victim and that there is an enemy that it should unify against.

Targeted fake news and propaganda allows politicians to demonise particular groups. Parties drill the message into target groups that they must support the party to protect the group’s interest and that the party is the only one that would act against the common enemy. It also helps if the opposition can be shown to be close to the said enemy, ensuring that the target voter clearly knows which party to support.

Excerpted with permission from How To Win An Indian Election: What Political Parties Don’t Want You To Know, Shivam Shankar Singh, Penguin Books.