When people in the West talk about political campaigns they think of US presidential elections or the great struggles between traditional and new parties in Europe. Yet the world’s largest democracy, India just went through the biggest live experiment in how political communications is changing in the Internet age. Its most recent election saw a dramatic escalation in digital campaigning, both official and organically across social media. In this article tech entrepreneur Matilde Giglio gives her personal account of her work at the election frontline as head of the digital campaign for the National Congress Party (INC).
On my arrival in Delhi I had to come to terms with a party fighting for its political life against a powerful incumbent. We had the task of swinging the sentiment of 900+ million voters who varied from the sophisticated urban middle classes familiar with all the latest social media and mobile technologies to rural women living without Internet connections. The lessons I learnt came from the cauldron of India politics but, I believe, resonate around the world.
I knew I needed to understand who the target voters were, how to reach them and what to say to them. My job was to build the data-set infrastructures, to collect voter data and to come up with an action plan to couple that data collection with targeted engagement: the political marketing message.
I had to understand the complexity of the logistics of one of the world’s biggest information marketplaces and the special nature of its communication platforms. India’s democracy is vast and complex. It is a country of young people with a dizzying range of multi-hyphenated identities and varied opinions and concerns. It is also a country with a rapidly expanding middle-class and an accelerating adoption of digital technologies.
One example of India as a digital campaign test-bed is WhatsApp. As internet access surges in India with the proliferation of smartphones and cheap data, more than 300m Indians are now on WhatsApp, making the country by far its biggest market. WhatsApp has become the platform of choice for politicians because of its massive reach that goes beyond a party’s loyal voter base, but also because of the lack of gatekeepers. Messages forwarded through the system have no context about where they originate, but benefit from the trust of coming from a contact.
Like all Indian political parties we had hundreds of thousands of volunteers who spread messages over WhatsApp. Although each message was limited to 256 recipients it still enabled us to engage a much bigger audience overall.
In the end we used a complex tool-kit of different digital tools to engage with India’s huge and diverse electorate. We didn’t win, but in the seats we focused on our ad campaigns led to a vote share increase of between 4-8% in target constituencies and groups – according to online polling before and after the campaign and the exit polls. That was a great result, although not enough to change the overall outcome.
The campaign office was down a wide tree-lined Delhi street within view of the parliament, children playing cricket in the street lined by the carts of fruit-sellers piled high with mangos, lychees, and bananas. I had to pass through layers of security to enter the INC campaign office, a vast room full of over 100 campaign workers, engineers, political analysts, data scientists, video makers, designers, pollsters, and grass-roots volunteers.
We had to reach 400m online voters, as well as 500m voters who were offline but still could be connected via the grass-roots campaigners.
We would have to do it semi-blind. We had no access to the systems and data that most western campaigns might take for granted.
Our Digital Campaign Plan
How did we do it? We built it in house. That was my job.
We divided up the work into three main stages:
- Gathering data about the electorate: finding out who our target voters were. Those who could make the difference, the persuadable voters.
- Conducting research: finding out which issues and areas we should focus on, what messages and tone resonated with each segment and sub-segment.
- Engagement: The implementation of the marketing campaign and the content production phase.
The way we thought about these three pillars was actually a cycle, rather than stages. Each stage feeds into the other and they are heavily interdependent.
Data is Decisive
Running a campaign for India is like running a campaign for Europe. Not a single country, but trying to run a campaign across the whole continent. Except India is four times bigger. In India, there are over 900 million voters, over 100 different languages and 29 major states, with one alone containing half the population of Europe combined.
We started gathering data about the electorate overall. We took a lot of data from publicly and commercially available sources.
Publicly available datasets such as the electoral roll can give a lot of valuable information. A person’s name, gender, location (which voting booth they voted at), their voter ID and age, if you voted and when. I realised by talking to political analysts that this would help indicate their caste. Surnames can tell you 80% of people’s caste. Sometimes you can have a doubt between 2 castes, but the location of that person (constituency) helps.
As with caste, religion can also be determined by surname. So we asked our volunteers from different religions to share their own temple or mosque’s directories and we combined that with the voter registration records to build an even more accurate data base.
Caste and religion are key indicators of likely voter behaviour in India. Once we had those mapped, the next step was to include phone numbers.
Data protection laws in India are minimal except for finance and health. Data brokers and telecom companies sell phone numbers, alongside data on electricity usage, which it turns out is a remarkably good proxy for wealth. In India if you have air conditioning or a dishwasher then your bill is big and essentially means you are middle class.
By the end of the campaign we had about dozens of data points for each voter, enriched by the Congress Party’s data of donors and volunteers. If only we had this data in advance we would have been so much better prepared.
Once this data was gathered, the next step was to define which seats to target.
To do this we had to build a model of our own, largely from scratch.
The first thing was to look at the last three elections: two the INC won, one they lost (very badly). I got a list of all the constituencies and how we performed. Unfortunately, in India they often change the names of the constituencies every few election cycles. The only way to track the changes of the 543 seats was manually. We analysed the results of each seat and did a strength analysis of each. This was made up of a combination of how we performed historically on a scale of 1-5; how close we were to winning or losing; and then weighted to the most recent results in 2014. We came up with a marginal score for each seat. The model also needed to include the hardest variable of volatility: how much has your performance varied over time?
We then calculated the minimum amount of resources – money – you needed to win a seat. We had a set budget so we needed to focus on the seats we could win: how many votes did we need to win those constituencies based on who we thought our core voters were. We started with the people who had voted for us in 2014 but then had to define our swing voters including those who had ever voted for INC. That produced the figure for the maximum target votes. So we created a table ranking the constituencies on how many voters we needed to convince to win in each place.
There are massive differences in the size of the electorate in each India constituency so if you have a limited budget and time frame you need to focus on the cheapest seats you can win. Clearly it makes logistical sense to focus on the smallest winnable seats.
Research for Targeting
The other key component was research. The constituency data only told us which states we could possibly win. We needed to know who within those areas we needed to target. Who are the swing voters? What issues do they care about? A cornerstone of building the congress data sets was research: polling and focus groups. Polling tells you who the swing voters are – focus groups tell you what they are thinking.
YouGov’s work for us focused on online polling so we needed to correct for the online bias towards the urban middle class. We had a call-centre do live calls and another company do robot-calls to reach people offline, mainly in rural areas. I discovered that robot-calls were especially effective for women in villages. Completion rate was over 95%. Clearly those women were happy to stay on the phone replying to questions from a machine for half an hour.
Some of the questions we needed answers for were:
- Channels: Which media platforms did people use to get informed about politics – TV? Facebook? SMS? WhatsApp?
- Issue discovery: Which issues do they care about? This might be as simple as asking do your family or any friends have sheep or cows?
- Emotional attachment: how angry or hopeful were they about issues? We measured that by tracking live how fast they replied to a question online- questions about immigration or religion might be answered in a second, something like health or employment might take longer. It indicated their passion about that issue.
- Frequency: how many times do we need to hit a person with a message before they change their mind? We measured that by polling them before and after advertising campaigns.
- Prioritisation: which issues should be put first, second or third according to the emotional attachment they had. Should you start with the issue they cared most about so that you grab their attention? Or should you start with a lower priority to build trust gradually before catching their commitment with the key issue. The latter actually worked best according to the level of impressions or shares of our messages. The signups to our website rocketed when we used this approach.
We conducted polls that led to voter segmentation for content generation. It showed that we needed to focus on women, rural people and minorities: all people with a low propensity to turn out to vote so our campaign changed from a persuasion to a turn-out campaign.
We then conducted focus groups to identify the message and tone that could work best for different groups. The aim of the focus groups was both exploratory and pre-testing. We pre-tested our core messages with the narratives that we thought could work. For example, how concerned are Muslims about the security of their community?
Then we did exploratory testing of more marginal and broader messages for targeting groups. These were general questions like, ‘what do you think about when you wake up in the morning?’ or ‘what do you talk about with your family?’
What I Learnt About Indian Voters
What I learnt is the universal truth that all politics is retail politics. People care most about the small things that impact on their daily lives: the price of fuel, of food, the job that their friend didn’t get, more funding for education because I didn’t get on the course that I wanted.
I also learnt that voters don’t necessarily understand economics. In particular, they seemed to have a blind spot for inflation. “The price of milk is higher than five years ago, that’s why I am angry at Modi.” So one of the most effective campaigns we had was a very simple set of graphics where we compared the price of basic goods such as fuel and, of course, they had gone up over the past five years.
The second thing I realised that while our politicians wanted to talk about big issues such as GDP or the Rafale Missile Scandal, at the grass-roots voters did not really care because it didn’t have any direct impact on their lives.
We identified who we needed to target, where to spend our money and where we needed to send key politicians to campaign. The cycle of data, research, engagement was already spinning. We had built a ‘pathway to victory’. Or so we hoped.
Campaign message creation occurs only after the target audience is defined. Similarly, marketing messaging should be built upon established buyer personas created with extensive data, research and analysis of your audience.
Our messaging had been running for weeks but the actual official campaign now kicked off for real, just a few weeks after I’d arrived in Delhi.
Part of the engagement phase is also defining the platforms that you are going to use. One unique thing about India is the online vs offline breakdown. Although over 400m people have internet connection and data has become ultra-cheap in the last 3 years thanks to the mobile network provider Reliance Jio, the majority of voters in rural areas still lack internet connection.
Platforms and formats were decided based on whatever was widely available. Platforms are a function of what the electorate has. Some areas of India have mobile phones and cheap data and people are happy to watch short videos whereas in rural areas where you have no internet connection we used SMS, calls and TV ads.
We could now go to the content team with our results. We told them: these are the people you are talking to, these are the messages and this is the tone that will work. They then created thousands of different images, graphics and videos.
In India there are 100 different languages so every piece of content has to be tailored to those different groups. In practice, we were data-rich, but content poor. We created thousands of ad campaigns but still did not have the capacity to refine all the content in the time frames we needed it in the form and quality that we wanted. We were always running behind schedule. Indian elections run in seven phases over two months where different constituencies vote at different times. We rarely had the material for each phase or constituency ready in time.
The problem was made much worse by the Indian Electoral Commission. Every piece of digital content had to be approved by a bureaucrat in an office in Delhi. I visited them in a dark building with a hundred officials in little cubicles. I would take six colleagues with me to carry print-outs of each digital message. We would wait hours to see someone who might then take a week to give their approval.
I remember sitting there for four hours on my first visit to get a message approved. It blamed Modi for unemployment going up. The official looked at me as if I was crazy. ‘This can’t be approved’ he said. “The first rule of the Electoral Commission is that you can’t criticise the government’. ‘But what can I talk about?’ I asked, ‘We are the opposition – our job is to criticise the government!’. He replied, ‘Why don’t you talk about what great changes you are going to make?’. I said, ‘But every BJP advert I see attacks the Congress?’. ‘Ah’, he said, ‘But you are not the government’.
All the evidence shows that negative or attack campaigning is the tactic that works best, especially if you are the opposition. I realised that not all our problems were internal.
Facebook is a key platform for politics in India. As they are subject to oversight from the Electoral Commission it meant they would also reject anything from our official pages that was not approved.
We couldn’t say anything negative about the government but that would not stop us making videos with people saying what had gone wrong in their lives under Modi. For example, a woman saying that she didn’t feel safe in the street, that four years ago things were better, and that she would vote for change.
The Electoral Commission had to approve them because they didn’t attack the government directly. They were extraordinarily successful on all digital platforms.
What happened next?
In the target seats we focused on, online ad campaigns led to a vote share increase of between 4-8% in our target constituencies and groups – according to online polling before and after the campaign and the exit polls. Which was a great result for our efforts, although not enough to win the overall contest.
A campaign effect is only marginal and in the end politics is the art of the possible. It is about broader forces and fundamentals such as the economy. Modi had massive structural advantages: campaign resources: financial, human, wide control of the mainstream media and bodies such as the electoral commission.
After the result the junior campaigners were crying in the office. But the bulk of the team were phlegmatic. They would be there for the next four years. But lessons have been learnt. A pollster has been appointed to make sure that next time they have the data strategy in place before the campaign starts.
This was the most extraordinarily challenging and yet rewarding experience of my career so far. The future of political campaigning will have data and digital at its heart. India is an inspiring but hugely complex democracy. We can see that around the world, elections/politics are more complicated and volatile than ever. Understanding how to use these new tools to serve democracy and fight for power will be essential to anyone who wants to change the world.
This article by Matilde Giglio who co-founded a machine learning-driven journalism start-up after studying at the LSE. She is now an entrepreneur in residence at a London VC firm.
All articles posted on this blog give the views of the author(s), and not the position of Polis, nor of the London School of Economics and Political Science