IDinsight is working with Indus Action to build an outreach system that connects people in India with government welfare schemes. To learn more about major components of the outreach system, and how we deployed it as a serverless application on Amazon Web Services read our full report here.
Photo by Yogendra Singh from Pexels
(FULL REPORT) Matching eligible people to social welfare programs in India - 622 KB
India has many government welfare programs designed to transfer money to some of its most vulnerable people. But for a number of these schemes, people either don’t know they are eligible, or don’t apply and many of the funds allocated remain unspent. For example, only 40% of INR 50,000 crores has been spent to date in a fund dedicated to the welfare of construction workers. The Building and Other Construction Workers (BoCW) Act, implemented in 1996, provides a range of transfers including health, education, maternal, and pension benefits, for workers in this industry. However, less than 50% of the estimated construction workers in the country are registered with labour boards, and a smaller fraction has accessed these benefits to date.
The registration and claims processes for social benefits can be complicated and onerous, which is one reason why few eligible people access these benefits. A second reason is that the government does not know which citizens are eligible, and in turn, individuals don’t know about all the schemes and their eligibility.
A number of other groups are working to improve people’s access to services by creating platforms that help them determine their eligibility for different social programs. However, even if people are aware of the service, they may not have enough digital access or literacy to effectively use them. The people who are most vulnerable are least likely to get online to check their eligibility or have funds to pay for a service.
Indus Action, an India-based non-profit that supports households to gain access to legislated benefits, overcomes these challenges by reaching out to this population directly. Their outreach approach includes phone calls, SMS messages, and in-person visits. Through their direct work with individuals and collaborations with state governments, they have rich databases with information about individuals who may be eligible for different services.
Given Indus Action’s available information about an individual, could we identify the set of benefits they would be eligible for? If we were able to build a system that answers this question of who might be eligible for what, we could then develop a smart outreach campaign that could channel resources to where they would be most likely to reach eligible households.
Indus Action and our IDinsight team recently built this outreach system that, given a set of citizens and their characteristics, is able to predict the probability that a citizen is eligible for a benefit.
A major feature of this system is that it can still make predictions about an individual’s eligibility even if their data in the system is incomplete. The first set of benefits targeted through this system were those offered by the Building and other Construction Workers Act (BoCW) in Delhi.
Often, multiple eligibility criteria need to be met by an individual to qualify for a benefit. For example, to qualify for the educational benefit under BoCW, a construction worker should be registered with the labour board for at least one year and should have proof of admission of their children to a school. However, the government doesn’t know whether a worker has children, as they did not collect this detail at the time of registration. How can the government now use available information to help the workers access the benefit without asking them to provide more data than they have already provided? To help provide that information, we developed an outreach system that uses a roster of benefits and their requirements and the attributes of a list of individuals. The system then returns eligibility scores for all the (benefits x individuals) combinations.
While matching individuals with benefits with incomplete information, we considered two kinds of reasoning. First, given our existing knowledge of each individual, we were able to determine who is eligible/ineligible for some benefit whenever the criteria for that benefit matched known information about the individual. Second, if the information needed to determine eligibility was missing along some axes of benefit criteria, then there was uncertainty whether an individual was eligible for criteria. In such cases, it may be possible to infer this individual’s eligibility for the benefit using other information we know about this individual.
However, for decision-makers to prioritise which individuals to call, we needed to balance certainty and uncertainty. To help facilitate this, the outreach system computes two scores.
The first score that the system produces is the Proximity Score. This score indicates the certainty of a person’s eligibility. If the value is -1 an individual does not qualify for a benefit; if the value is 1 the individual does qualify given existing information. Furthermore, it indicates how close an individual is to fully qualifying for a benefit. For each benefit, if we do not know whether the individual qualifies or not, the score is the percentage of benefit criteria an individual is known to fulfill.
Figure 1: Calculating the proximity score
The second score that the system produces is the Probability Score. It represents the likelihood of an individual qualifying for a benefit given (a) the incomplete set of characteristics of the individual; and (b) the distribution of characteristics of other individuals. The system employs matrix completion techniques to impute the missing individual attributes.
Figure 2: Calculating the probability score
Neither of these scores are perfect. But when used together, they can inform the design of outreach campaigns. Decision-makers can prioritise individuals to contact by balancing known eligibility with uncertainty.
As a major part of the project was to take the outreach system from a proof-of-concept (PoC) to a live production system, we needed to understand how Indus Action wanted to use the outreach system to support its programmes. When building any sort of software, it is important to put the end-user first and build around their needs. A standard method to build an understanding of the user is through user stories, which are descriptions of how a solution or feature would drive value for the user. Through co-design sessions, we worked with the Indus Action team to develop three user stories.
To create a proof of concept for these user stories, we designed a serverless solution1 in AWS for the outreach system. Post which we used AWS Serverless Application Templates (SAM) with GitHub actions to deploy our solution to the cloud. This enabled us to spin up CI/CD (continuous integration / continuous development) with minimal effort as building the PoC took place in parallel to the creation of other features for the solution.
Figure 3: Development workflow
Changes in local deployment, when pushed to the cloud, automatically refresh deployed resources in AWS to reflect those changes. This enabled large-scale feature changes such as the inclusion of bootstrap imputation to take place over a few days as opposed to a couple of weeks.
Our solution on AWS used three resources, AWS Lambda for computation orchestrated by AWS Step Functions with flat files stored in S3. This minimal architecture supported the scoring of hundreds of thousands of individuals.
The challenges – matching eligible households to welfare services – that this project was designed to address are not specific to the BoCW Act in Delhi. These barriers contribute to the low uptake of many schemes in many states. Maternity benefits under PMMVY are one example. Fortunately, the solution we created is configurable enough that it can be extended to other such benefits easily.
While we believe that this approach will make a huge impact on its own, steps to simplify the application and related processes could also make a big difference. Specifically for BoCW, Indus Action and state governments are diagnosing documentation-related challenges for migrant workers, setting up user-friendly interfaces for application, and strengthening processes for doorstep delivery to increase registration of construction workers.
12 July 2020
16 June 2019
20 February 2020
15 January 2021