While the online services provided by social security institutions enable tasks to be performed ever more efficiently, they do not usually allow for open-ended questions or inquiries. In this context, chatbots (automated inquiry services) are emerging as a tool that can add value.

A chatbot, chat robot or bot, is a tool that makes it possible to hold conversations with users, simulating conversation with a person and providing automated responses to the most common questions. The technology allows the user to hold a conversation, without requiring a physical person to respond. As a result, it can be available on a continuous basis.

Chatbots are a specific type of virtual assistant, helping people to complete tasks through oral or written communication. These tools are usually added to websites, business apps, social media or instant messaging services.

At present, there is a great deal of interest in this type of technology, which is increasingly being adopted by social security institutions since it can be set up within a few months, and at a reasonable cost/benefit ratio, enabling them to handle open-ended inquiries. This trend is apparent from the good practices and experiences reported by members of the International Social Security Association (ISSA) across all regions.

Drawing on these experiences, and through its own Guidelines, the ISSA promotes the use of this type of technology. In the ISSA Guidelines on Communication by Social Security Administrations (ISSA, 2019a), guideline 10 refers to the “strategic use of new communication technology” in all areas related to the use of chatbots in social media and messaging systems. In the ISSA Guidelines on Service Quality (ISSA, 2019), guideline 5 focuses on “understanding users’ needs and experiences” which can serve as a basis for gathering the information and knowledge base required to set up virtual assistants.

The main benefits of creating a chatbot

The introduction of chatbots to social services has a number of potential advantages.

Firstly, they increase user satisfaction. The chatbot can respond at once, swiftly handling users’ queries as they browse the institutions’ websites. They increase the availability of advice services, because the chatbot is available 24/7, 365 days a year.

Greater use of the service leads to an increase in the number of users interacting with the social security institution. It has been shown that the use of chatbots on websites, replacing online forms, increases user engagement with these sites, while minimizing in-person or telephone contact. This also increases the capacity to obtain information from users, because chatbots can handle thousands of users at once, searching for and extracting relevant information. The institution can use this information for statistical purposes or to analyze service provision, thus improving the user experience in future interactions.

Finally, it is possible to improve the advice services’ cost/benefit ratio, because chatbots do not require a physical person to be available full time to attend to users. The chatbot can help users in real time, at any time of day, leading to cost reductions and better service quality.

Chatbots based on artificial intelligence (AI) are also able to learn on a continuous basis, through interaction with users, which means that the range and quality of responses is constantly improving.

Types of chatbot and their classification

When deciding to set up a chatbot, it is important to decide on: the user needs to be addressed, the technology required and the preferred communication channel. This approach facilitates an analysis of the various types of chatbot and helps to identify which option would be best suited to the requirements.

There are various types of chatbot, which differ in terms of the level of sophistication, complexity, and their implementation and maintenance costs. In fact, chatbots can be classified according to a number of criteria.

On the one hand, chatbots can be classified according to their implementation techniques and the extent to which they rely on AI. Chatbots using AI, also known as cognitive or intelligent chatbots, make it possible to provide more sophisticated services with a response quality closer to that of a human agent. However, their implementation is based on training an AI system rather than the development of traditional software, and this in turn requires the availability of a dataset representative of user interactions. There are also chatbots that use traditional rules- and operations-based programming.

Chatbots can also be classified according to the means of communication, i.e. whether they are confined to text, can respond to spoken requests or allow for multimedia interaction (for example, video). They can also be classified by the purpose of the interaction (for example, contact, information, online help), and according to whether their operational capacity enables them to carry out transactions. The communication channel used to deploy the chatbot is another criterion for classification; for example, the chatbot could be used on websites, social media or messaging applications. Institutions can opt for a multichannel approach, adding in multiple communication channels to reflect user preferences.

All of these classifications and types of chatbot can be combined in different ways, giving rise to specific solutions designed to meet each institution’s needs.

Table 1. Categories

Categories of chatbot
According to implementation technique and use of artificial intelligence	ITR (interactive text response)	Do not require the use of artificial intelligence, as they are command based. They provide preset buttons and apply a sequential logic based on a predetermined menu of choices.
	Word spotting	Respond by recognizing keywords, which then trigger a scripted response.
	Cognitive – intelligent chatbots	Draw on artificial intelligence techniques such as machine learning. Are able to use natural language processing. They are context based, i.e., they are able to decipher user intent and context, and to devise responses from scratch. Cognitive chatbots learn from previous interactions and/or from new information supplied to them.
According to the means of communication	Text chatbot	Interacts through text messages, like a traditional chat tool.
	Voice recognition chatbot	Uses a voice interface. Understands users’ spoken requests and responds through the loudspeaker. Similar to an interactive voice response, but with the ability to understand speech and give responses in context.
	Multimedia chatbot	Combines text messages with more dynamic features such as images, buttons and other content, simulating a real human conversation through messaging applications.
According to the purpose of the chatbot	Contact	Used in place of online forms.
	User support and assistance	Used as a more interactive version of frequently asked questions.
	Online procedures assistant	Supports the user through the various steps of an online procedure.
	Social	Aims to provide personalized content in a conversational format.
According to operational capacity	Transactional	Enables the user to carry out a transaction or procedure. Must be able to interact with internal systems or third-party services to search for information or complete transactions.
	Non-transactional	Works like a chatroom. Generally used for frequently asked questions or to help users navigate a website.
According to communication channel	Website	Added to a website to interact with visitors.
	Social media	Used to increase user loyalty and respond to inquiries 24 hours a day.
	Instant messaging	Incorporated into messaging applications used by the organization.
	Multichannel	Included in several of the communication channels listed above, to reflect user preferences. For example, a user could interact via social media while also obtaining information on the website and sending a message.

Whichever type of chatbot is used, the option to contact a human agent should always be available wherever this is required by the user or due to the complex nature of the issue. Ideally, it should be possible to continue the conversation on any communication channel. This means that the existing conversation and the surrounding context must be passed on to a call centre, which must have multichannel capacity in order to incorporate all communication channels consistently.

Incorporating chatbots in social security services

Presentations of experiences given at the ISSA Good Practice Award for the Americas 2020 and in the context of other ISSA activities show that social security institutions are tending to make more use of chatbots. Some specific examples are set out below.

Argentina

In Argentina, the Superintendency of Occupational Risks (Superintendencia de Riesgos de Trabajo – SRT) wanted to reduce the pressure on its customer service phone-lines and respond promptly to users, freeing up the more specialized customer service telephone agents so that they could handle more complex inquiries. The SRT set itself the challenge of reaching a larger number of users through a help service that could respond quickly and efficiently, on a 24/7 basis, to the most common questions.

To this end, the SRT launched Julieta, a virtual assistant (SRT, 2020), which focuses mainly on the occupational risk system, but also takes questions on the status of a file, joining an occupational risk insurance scheme and finding out at which office a particular procedure should be initiated.

The chatbot was developed gradually, with the first step being to generate a knowledge base with a limited number of questions and answers. There were 100 answers and some 3,000 different question formulations linked to those answers. In the second step, a focus group of employees was used to interact with the chatbot, which then made it possible to double the number of answers (200), generating 8,500 different question formulations. This development reduced the margin of error from 30 per cent to 7 per cent, i.e. 93 per cent of Julieta’s answers were found to be correct. In 2019, with a daily average of 500 inquiries from more than 85,000 users, 200,000 exchanges with Julieta were generated. The SRT is planning the next stage, in which the chatbot will be integrated with a messaging service.

The following figure shows the Julieta interface.

Brazil

In Brazil, the National Social Security Institute (Instituto Nacional do Seguro Social – INSS) set up a virtual assistant called Helô (INSS, 2020). The initial aim was to field inquiries about the use of the digital platform, Meu INSS (My INSS), and it is currently being extended to create a hybrid system with human assistance for more complex matters.

The COVID-19 health emergency led to the closure of in-person services and the use of Meu INSS increased accordingly. This situation led to a pressing need to provide relevant information quickly and easily, offering better guidance to citizens.

With Helô, the INSS achieves greater transparency and provide more ways to get in touch, offering a continuous service, available 24/7. In turn, this new service means that the INSS can make best use of its customer service assistants, minimizing human involvement in the more straightforward cases.

As with the SRT in Argentina, this initiative was implemented in phases. The first phase involved setting up a rules-based virtual assistant, using keywords. The second phase included the creation of a knowledge base, to provide a more targeted service to citizens based on their profiles, and to enable integration with other social media and messaging platforms. In the first month of operation, a million calls were received, serving an average of 32,000 citizens a day; 57 per cent of citizens using Helô said that it responded correctly.

The figure below shows the visual appearance of Helô:

Panama

The Social Insurance Fund (Caja de Seguro Social – CSS) of Panama set up RoVi – a virtual robot, (CSS, 2020) an assistant to automate customer care and answer the most frequent questions from insured persons. The challenge facing the CSS was to improve the quality of health services, introducing greater fairness, effectiveness and efficiency, as well as a more holistic approach.

In line with its strategic plan, which includes the creation of models based on new technologies that bring services closer to the citizen, the CSS has developed a set of initiatives. These included the development of digital consultation, the option of making online payments, and the development of RoVi as a way to quicken formalities for insured persons by providing a permanently available virtual assistant.

One aim of the CSS service digitalization was to provide basic medical care remotely, from any telephone device, a goal which was fully achieved thanks to the introduction of digital consultation. The system for making online payments of regular contributions was also well received. Over four months, it was used by 54,000 employers, processing 8,000 transactions relating to the employer-employee quota. As regards payments to retirees and pensioners, 42 per cent were completed online. Alongside these results there was an expansion and modification of the call centre, which was moved from a physical to a virtual office.

RoVi is part of this broader context of digitalization and makes it possible for insured persons to be offered guidance on the various services provided by CSS. In particular, it can provide updates on the status of requests for medication, saying whether they have been approved, are being processed, or are ready to collect.

RoVi is incorporated in the institution’s website, as illustrated in the figure below.

Uruguay

In Uruguay, the Social Insurance Bank (Banco de Previsión Social – BPS) set up a chatbot to act as a virtual assistant, analogous to a BPS staff (BPS, 2020). It is aimed at the domestic work sector, a difficult-to-cover group.

The chatbot – virtual assistant initiative is part of a broader strategy to promote the inclusion and formal incorporation of domestic workers in the social security system through the use of advanced technologies. The specific challenge facing the BPS was to improve the quality, accessibility and dissemination of information, and also to help employers with the relevant procedures.

The BPS developed a set of initiatives to improve the coverage of domestic workers, including the centralization of procedures to one organization; the encouragement of employer to use self-services via the mobile app; multichannel assistance with a free phone number; the sending of authenticated emails; assistance in-person and via social media; and the establishment of an intelligent chatbot, available 24/7, to provide help and advice.

The virtual assistant was launched with 70 responses (of which 19 were classed as “chit chat” or informal conversation). The user interface is enriched with suggested responses, animations and audiovisual material. The logic underpinning the virtual assistant also detects repeat attempts and can redirect the conversation to a human agent.

As a result of this strategy, 95 per cent of BPS procedures are handled remotely. Online procedures were used for 57 per cent of employers who registered workers and 42 per cent of payments. In 2019, a monthly average of 1,300 inquiries were sent to the virtual assistant. Of these, 97 per cent were handled solely by the virtual assistant, while the remaining 3 per cent required the involvement of a member of staff.

The following figure shows the chatbot screen.

Summary of results and key factors in implementation

All of the good practices submitted in this area aim to support and assist users, either by responding to common questions or through support to complete procedures online. They all involved additions to websites, though some organizations are also considering extending the service to social media and messaging applications.

Table 2. The following table summarizes the outcomes and classification of the chatbots described above

Country and institution	Focus	Main outcomes	Implementation techniques	Purpose	Operational capacity	Channel
ARGENTINA – SRT	Frequent questions Online guidance about transaction systems	93% of responses correct	Cognitive, based on artificial intelligence (AI)	User support and assistance Online procedures assistant	Transactional	Website Messaging (future)
BRAZIL – INSS	Questions about Meu INSS	57% of users satisfied with the response	Cognitive, AI-based	User support and assistance	Non-transactional	Website Social media and messaging (future)
PANAMA – CSS	Frequent questions	100% of devices covered	Cognitive, AI-based	User support and assistance	Non-transactional	Website
URUGUAY – BPS	Advice about domestic work	97% of responses produced solely by the virtual assistant	Cognitive, AI-based	User support and assistant	Non-transactional	Website

The main critical factors affecting implementation are to do with support from senior management and an institutional strategy to adopt new technologies bringing services closer to the citizen. This may be achieved by inclusion in the institution’s strategic plan or a preference for implementation strategies which make extensive use of technology.

Another important factor is the technical expertise of the team managing the project, both with regard to the social security issues on which the chatbot is to be trained, and in terms of their awareness of the intended audience, their needs and their desired responses. With cognitive chatbots, it is also vital to have the data available to train the algorithms based on artificial intelligence.

Finally, it is worth highlighting the fact that all of the good practices described adopted a stepwise approach, starting with those measures that were simplest to implement, either because of the scope of the issues involved or because of the communication channels chosen initially.

Conclusions

Chatbots can be defined as software tools that function like a virtual assistant and are able to maintain conversations with users, either by providing preset responses to a particular type of question, or by understanding user intent and inputs.

Social security institutions are increasingly using chatbots to provide uninterrupted customer services that can enable users to obtain advice. As illustrated by the cases discussed above, institutions are tending to use AI-based chatbots, which means that the chatbots learn for themselves and, over time, hold more complex and natural conversations.

The most highly developed chatbots incorporate advanced voice functions and natural language processing, enabling them to interpret questions regardless of the varied tones adopted by the users and to understand user intent. Some of them provide such an authentic experience that it is very hard to tell whether the response has come from a virtual robot or a human being.

The challenge is to enable chatbots to provide services which increasingly resemble what would be provided by humans. To achieve this, the chatbot needs to move beyond making logical deductions as to user intent. Many versions already use natural language processing and understanding, but they do not have the contextual information required to assess the user’s emotional state.

It has been suggested that in the future, artificial intelligence will incorporate vision and voice in a way that enables it to identify emotions on the basis of facial expressions, tone of voice or behaviour patterns. This would lead to a more personalized response which would be better attuned to each situation, creating easier, more natural interaction with users. The responses will then certainly bear a closer resemblance to those of a human being.

These developments can also help institutions to improve their capacity to innovate through the adoption of emerging technologies, such as artificial intelligence, which can then be applied to other areas beyond chatbots.

References