Implementing AI for Enhanced Public Services Gov.br: A Methodology

for the Brazilian Federal Government

ısa Kely de Melo

1,6,∗ a

, Silvia Ara

ujo dos Reis

1,7 b

, Vin

ıcius Di Oliveira

2,11,12 c

Allan Victor Almeida Faria

1,4 d

, Ricardo de Lima

3,12 e

, Li Weigang

2,12,∗ f

Jose Francisco Salm Junior

1,13 g

, Joao Gabriel de Moraes Souza

1,9 h

, V

erica Freitas

8 i

Pedro Carvalho Brom

2,10,12 j

, Herbert Kimura

1,7 k

, Daniel Oliveira Cajueiro

1,5 l

Gladston Luiz da Silva

1,4 m

and Victor Rafael R. Celestino

1,7,∗ n

LAMFO - Lab. of ML in Finance and Organizations, University of Bras

ılia, Campus Darcy Ribeiro, Bras

ılia, Brazil

TransLab, Department of Computer Science, University of Bras

ılia, Campus Darcy Ribeiro, Bras

ılia, Brazil

Ministry of Management and Innovation in Public Services, Federal District, Brazil

Department of Statistics, University of Bras

ılia, Campus Darcy Ribeiro, Bras

ılia, Brazil

Department of Economics, University of Bras

ılia, Campus Darcy Ribeiro, Bras

ılia, Brazil

Department of Mathematics, Federal Institute of Education, Science and Technology of Minas Gerais, Formiga, Brazil

Department of Business Administration, University of Bras

ılia, Campus Darcy Ribeiro, Bras

ılia, Brazil

School of Business and Management, Uberlandia Federal University, Uberl

andia, Brazil

Department of Economics and Business Administration, Brazilian Institute of Education Development and Research -

IDP, Bras

ılia, Brazil

Department of Mathematics, Federal Institute of Education, Science and Technology of Bras

ılia, Campus Estrutural,

Bras

ılia, Brazil

Federal District Secretariat of Economy, Bras

ılia, Brazil

Department of Computer Science, University of Bras

ılia, Campus Darcy Ribeiro, Bras

ılia, Brazil

University of the State of Santa Catarina, Florian

opolis, Santa Catarina, Brazil

Keywords:

Brazil, Design Sprint, Public Administration, AI, Machine Learning, LLM.

Abstract:

The website portal of the Brazilian federal government (Gov.br) consists of pages from almost 40 ministries,

180 public agencies and up to 5000 public services for all citizens, posing a signiﬁcant challenge in improving

service quality. This article presents an innovative methodology to implement artiﬁcial intelligence (AI) to

address these challenges, to enhance the efﬁciency, accessibility, and quality of services to the population.

The methodology combines elements of Lean Ofﬁce, Design Sprint, Analytic Hierarchy Process (AHP), and

advanced AI techniques, particularly Large Language Models (LLMs), making it ﬂexible and adaptable to

the needs of government entities. Developed in collaboration with project managers, public servants, and

stakeholders, the methodology includes a survey of demands, selection, and prototyping of AI projects in a

complex government context. The practical application selected the Gov.br portal for prototyping, involving

the development of an advanced generative agent to interact with citizens, clarify doubts, direct to the re-

quested services, and provide human interaction when necessary. The recommended practices offer a valuable

contribution to other developing countries seeking to integrate AI solutions into their public services.

https://orcid.org/0000-0001-8120-9778

https://orcid.org/0000-0002-1646-4454

https://orcid.org/0000-0002-1295-5221

https://orcid.org/0000-0002-4300-9334

https://orcid.org/0000-0002-9269-3558

https://orcid.org/0000-0003-1826-1850

https://orcid.org/0000-0002-8492-1645

https://orcid.org/0000-0003-0685-3082

https://orcid.org/0000-0003-3035-9738

https://orcid.org/0000-0002-1288-7695

https://orcid.org/0000-0001-6772-1863

https://orcid.org/0000-0001-5898-1655

https://orcid.org/0000-0001-9650-2993

https://orcid.org/0000-0001-5913-2997

∗

Corresponding author

Kely de Melo, M., Reis, S., Di Oliveira, V., Faria, A., de Lima, R., Weigang, L., Salm Junior, J., Souza, J., Freitas, V., Brom, P., Kimura, H., Cajueiro, D., Luiz da Silva, G. and Celestino, V.

Implementing AI for Enhanced Public Services Gov.br: A Methodology for the Brazilian Federal Government.

DOI: 10.5220/0012997000003825

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 20th International Conference on Web Information Systems and Technologies (WEBIST 2024), pages 90-101

ISBN: 978-989-758-718-4; ISSN: 2184-3252

1 INTRODUCTION

In recent years, the public sector has increasingly ex-

plored artiﬁcial intelligence (AI) to enhance opera-

tions and improve service delivery (Maragno et al.,

2023). There are at least 142 AI applications in fed-

eral administrations (Engstrom et al., 2020). Some

focus on making government information more acces-

sible, while others aim to improve citizen interaction

through Information and Communication Technolo-

gies (ICTs) (Kitsios et al., 2023). AI technologies,

such as machine learning (ML), natural language pro-

cessing (NLP), and computer vision, hold great po-

tential for automating tasks, analyzing large datasets,

and offering more efﬁcient, personalized services.

The disruptive potential of AI in the public sec-

tor is evident in three main areas: enhancing internal

efﬁciency, improving decision-making, and strength-

ening citizen-government interaction. AI applications

in the public sector include facial recognition in polic-

ing (Blount, 2024), recidivism prediction in criminal

justice (Wang et al., 2023), virtual agents for process

automation (Scutella et al., 2024), forecasting social

service needs (Dwivedi et al., 2023), chatbots (Chen

et al., 2024), healthcare (Yu et al., 2018), public trans-

port (Jevinger et al., 2024), education (Cohen et al.,

2023), environmental management (Fan et al., 2023),

and agribusiness (Kutyauripo et al., 2023). These

initiatives promote more inclusive services and en-

hance citizen participation in public sector activities

(Samoili et al., 2020).

Integrating artiﬁcial intelligence into the public

sector is evolving, with much still unknown about its

full potential and optimal strategies. Existing research

on AI adoption highlights three key values: trans-

parency, effectiveness, and efﬁciency (Rocha et al.,

2022; Faria et al., 2022; Chen et al., 2023). Rec-

ognizing the need for clear guidelines, the Brazilian

federal government aims to identify the best practices

for selecting AI projects in public agencies.

This paper presents the methodology developed

for the project “Artiﬁcial Intelligence Research Ap-

plied to the Prototyping of Solutions for the Fed-

eral Government’s Artiﬁcial Intelligence Ofﬁce.” The

methodology focuses on mapping, prototyping, and

documenting AI solutions for public administration

issues, aligning with the responsibilities of the Secre-

tariat of Digital Government (SGD) (BRASIL, 2023),

under the Ministry of Management and Innovation in

Public Services. The proposed solution supports AI

governance and the development of data models, pro-

cesses, and standards, including data mining, analy-

sis, and visualization technologies. These practices

enhance the public policy management cycle and the

provision of services. Given that these responsibili-

ties are part of other countries’ digital transformation

strategies, the methodology offers a valuable contri-

bution to governmental agencies developing AI solu-

tions for public services.

It also details the step-by-step methodology for

selecting and prototyping projects in AI Decision-

Making in the Public Sector. This methodology was

developed and applied in collaboration with AI and

project management expert researchers, members of

the SGD, and other bodies comprising the Brazil-

ian Government’s Artiﬁcial Intelligence Hub (NIA).

Figure 1 illustrates NIA, highlighting partner insti-

tutions’ diversity. The results of the methodology’s

application in federal government agencies are re-

ported, highlighting recommended practices and ob-

tained outcomes. The project selected for prototyp-

ing, arising from the demands identiﬁed by the public

agencies participating in the research, was the Gov.br

portal service (BRASIL, 2024). This portal com-

prises pages from nearly 40 ministries and 180 pub-

lic agencies, integrating up to 5000 public services.

Over 150 million Brazilians access it and records ap-

proximately four thousand accesses per second (Gov,

2024). The prototyping involved the development of

an Advanced Generative Agent to interact with citi-

zens, assist them, clarify doubts, and direct them to

the requested service pages, with human overﬂow in

necessary cases.

2 RELATED WORK

We present the following works that explore various

aspects of AI solutions for public services and the

techniques used in the proposed methodology. Ad-

ditionally, we contextualize the application of AI in

public services within the Brazilian framework.

2.1 Public Service with AI

Implementing AI in the public sector requires un-

derstanding its capabilities and addressing challenges

like explainability, data governance, and infrastruc-

ture. Human-AI collaboration can improve tasks and

create new jobs, but caution is essential to avoid

mistrust. Engaging stakeholders, such as citizens

and suppliers, foster innovation but demands effec-

tive ecosystem management and clear regulations

(Maragno et al., 2023). Currently, there is no uni-

ﬁed global approach to determining AI’s legal status

(Atabekov, 2023), and most countries focus on the

process rather than the ﬁnal result generated by AI

(Atabekov, 2023).

Implementing AI for Enhanced Public Services Gov.br: A Methodology for the Brazilian Federal Government

Figure 1: Brazilian Government Artiﬁcial Intelligence Hub (NIA).

There is a knowledge gap in academic research

concerning AI’s inﬂuence on public values and the

governance challenges that arise from this interac-

tion (Chen et al., 2023; Kitsios et al., 2023; Maragno

et al., 2023; Neumann et al., 2024). Insights into how

government employees perceive AI’s impact on pub-

lic values and governance challenges remain limited.

The debate over whether AI will transform the pub-

lic sector and its operational environments continues

(Kitsios et al., 2023).

The rising interest in AI within government spans

various scholarly ﬁelds. While this diversity is en-

riching, it can cause fragmentation in theoretical and

terminological approaches as researchers from dif-

ferent disciplines work independently (Straub et al.,

2023). Technical research often neglects broader per-

spectives and real-world contexts. Additionally, the

increasing number of typologies and taxonomies to

interpret AI’s role in government sometimes needs

to be more interconnected, reinforcing a fragmented

knowledge base (Straub et al., 2023).

Despite growing debate, AI diffusion in the public

sector remains limited compared to the private sector

(Neumann et al., 2024). Interdisciplinary AI scholar-

ship in government is still emerging, and theory has

yet to offer signiﬁcant value to practitioners (Birhane

et al., 2022). Humans remain crucial as decision-

makers and rule enforcers, including public safety and

fraud investigations (Chen et al., 2023). The abun-

dance of conceptual terms and lack of integration be-

tween social sciences and AI’s technical aspects leave

governments uncertain about which principles to pri-

oritize in AI governance (Straub et al., 2023). Glob-

ally, AI in government is still in early implementa-

tion, with key decisions shaping future expectations

(Straub et al., 2023).

2.2 AI in Public Service in Brazil

Brazil has advanced its digital agenda since the

1990s, starting with the 1991 Information Technol-

ogy Program, which offered tax incentives for R&D

in hardware and automation and promoted company-

university collaboration (Filgueiras and Junquilho,

2023). In 2018, the Brazilian Strategy for Digital

Transformation (E-Digital) (BRASIL, 2018) outlined

actions for advancing public policies, emphasizing

AI, big data, 5G, and modernizing production. It also

highlighted the need for ICT training and assessing

the impacts of disruptive technologies. In 2021, the

Brazilian Strategy for Artiﬁcial Intelligence (EBIA)

(BRASIL, 2021) was introduced, focusing on AI re-

search, public security, legislation, and international

governance (Anna, 2023). EBIA promotes innova-

tion by fostering knowledge exchange through coun-

cils and networks (Nonato et al., 2024).

Brazil has made signiﬁcant strides in innova-

tion potential (Anna, 2023), with growing social in-

terest in enhancing public sector innovation to im-

prove efﬁciency, quality, and accessibility (Guedes

and J

unior, 2024). The Brazilian government has

established AI centers of excellence, funded univer-

sity research, created collaborative networks, and de-

veloped ecosystems while ensuring privacy and data

protection (Filgueiras and Junquilho, 2023). The

public sector leads in policy formulation, promotes

data transparency, fosters innovation, supports profes-

sional training, and invests in emerging technologies

to build effective governance ecosystems (BRASIL,

2021).

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

Despite various initiatives to promote digital de-

velopment, Brazil’s efforts need coherence and align-

ment with its social reality (Filgueiras and Junquilho,

2023). Key challenges hindering EBIA’s implementa-

tion include a clear timeline for strategic goals (Anna,

2023) and unclear governance structures (Gaspar and

Mendonc¸a, 2021). As emphasized in the public con-

sultation, regulatory bodies, speciﬁc authorities, and

existing structures are needed. The lack of policy inte-

gration leads to non-cooperation among actors, mak-

ing EBIA more of an ineffective document than a col-

laborative strategy for problem-solving (Gaspar and

Mendonc¸a, 2021).

Several AI initiatives have been identiﬁed in

Brazil’s public agencies. The Conecte SUS project

aims to modernize public health management (Lemes

and Lemos, 2020), while the EB S@

ude Project uses

AI to optimize military health databases (Lemes and

Lemos, 2020). Regulatory sandboxes are advanced

in the ﬁnancial sector, with institutions like the Min-

istry of Economy and the Central Bank of Brazil

implementing models since 2019 (Kubota and Rosa,

2024). The Brazilian Federal Revenue Service uses

SISAM for real-time processing of over 2.5 million

import declarations (Guedes and J

unior, 2024). Vari-

ous agencies, including the Federal Court of Accounts

and the Federal Police, employ AI robots to boost pro-

ductivity (Desordi and Della Bona, 2020). Additional

AI use cases are documented in an OECD.AI inven-

tory (OECD.AI, 2022).

Although broad social participation contributed to

the national AI strategy, it has yet to be fully imple-

mented nationwide. To address this, the Federal Gov-

ernment established the AI Hub (NIA), tasked with

creating a list of Priority AI Projects, deﬁning projects

eligible for funding, and promoting academic stud-

ies to solve public sector challenges. The project dis-

cussed in this paper is an initiative of the NIA SubHub

for Prospecting Strategic Projects, executed by the

University of Bras

ılia (UnB) through LAMFO under

a Decentralized Execution Agreement (DEA). This

SubHub comprises professionals from various ﬁelds,

as described in Figure 1.

3 EMPOWERING SERVICES: AI

FRAMEWORK

A standardized methodology is necessary for imple-

menting AI in public services. To meet the goals of

the National Artiﬁcial Intelligence Strategy, this pa-

per proposes a comprehensive process for deploying

AI solutions. The following sections outline steps for

identifying demands, selecting projects, designing so-

lutions, prototyping, and validation. This methodol-

ogy was developed through a literature review, high-

lighting key ﬁndings and limitations of AI in pub-

lic administration. By exploring best practices and

emerging technologies, we applied the concept of

Meta AI—learning to learn about AI—through thor-

ough applied research to achieve the project’s objec-

tives.

To address the challenge, the proposed methodol-

ogy aimed to answer the following question: What

is the best method for selecting AI projects in gov-

ernment with subjective issues to be judged and

more than ten alternative options? Initially, a lit-

erature review was conducted on methods for se-

lecting AI projects in government with subjective

judgment criteria, highlighting Multicriteria Decision

Analysis (MCDA) methods, particularly the Analytic

Hierarchy Process (AHP) (Saaty, 1977), due to its

widespread use for project selection. AHP Express

was selected for cases with many alternatives (Leal,

2020).

In addition to the method for selecting AI projects,

after the evaluation was completed, Lean Ofﬁce prac-

tices were incorporated to improve public administra-

tion services and reduce bureaucratic overhead. Two

literature reviews were conducted to advance this re-

search: one focused on digitalization and the other on

Lean practices.

This resulted in the development of the AI so-

lution selection and prototyping process for public

services, which was designed based on the proposed

methods in the literature and enriched through discus-

sions among the AI scientists, who are the authors of

this paper, and members of the SGD, the institution

responsible for promoting AI project development in

the Brazilian federal government. Consequently, a

methodology was proposed that reﬂects cutting-edge

knowledge and addresses the speciﬁc and complex

peculiarities of public administration in a country as

large as Brazil. The steps of the proposed method-

ology are illustrated in Figure 2, with each phase’s

objective described below.

Step 1: Survey. Conduct a comprehensive survey

of federal public administration agencies’ AI needs.

This step investigates departmental challenges and

identiﬁes areas where AI can enhance public service

efﬁciency and effectiveness. The research should be

thorough, incorporating interviews, questionnaires,

and analysis of existing data.

Step 2: Elimination. Following the application of

the Survey (Step 1), various project proposals for ap-

plying AI are identiﬁed. At this step, an initial ﬁlter

is used to ensure that the proposed projects can in-

deed be addressed through AI. This ﬁlter eliminates

Implementing AI for Enhanced Public Services Gov.br: A Methodology for the Brazilian Federal Government

Figure 2: Steps of the Project Selection Process for Prototyping.

proposals failing to meet the established criteria, en-

suring that only viable projects proceed to the next

phase.

Step 3: AHP - Classiﬁer. The proposals ﬁltered

in the elimination step proceed to prioritization. In

this step, the proposals are ranked in order of prior-

ity and preference, considering the subjective aspects

of stakeholders, such as expected impact, urgency,

and feasibility. The AHP is utilized here to structure

and facilitate decision-making, ensuring that the most

promising and strategic projects are selected.

Step 4: Problem Framing. In the proposed ap-

proach presented in this paper, Problem Framing was

used - for (Bardwell, 1991), Problem Framing helps

structure and provide information organization to un-

derstand the problem, which people do not do daily.

Hence, it helps create a data management and analy-

sis environment that aims to solve problems. A lack

of understanding of the problem can lead to solu-

tions that do not bring the desired effects (Robertson,

2016).

Step 5: Design Sprint. Afterward, an adapted ver-

sion of the Design Sprint was used (Knapp et al.,

2016). A study (Mendonc¸a de S

a Ara

ujo et al., 2019)

mentions that the computer systems industry has

widely adopted Design Sprint to minimize the prob-

lems in deﬁning requirements in software projects.

Another study (Larusdottir et al., 2021) highlights the

focus on adapting the design of the product, usually

a computational application, to the users’ needs and

adds that, in the Design Sprint proposal, a relatively

small team, which initially has a superﬁcial idea of

the users’ requirements, intends to, at the end of the

process, have a prototype solution.

Step 6: Prototyping the Solution - Type 1. Aiming

at the prototyping of the solution selected in Step 5,

a theoretical and summarized prototype of the solu-

tion is conceptualized. This conceptualization occurs

through collaborative brainstorming sessions between

the project researchers and the team from the SGD.

In this step, the existing computational and scientiﬁc

methods in the literature that are most suitable for im-

plementing the solution are identiﬁed and discussed.

The deﬁnition of requirements and the initial system

architecture are outlined.

Step 7: Prototyping the Solution - Type 2. The

researchers, specialists in artiﬁcial intelligence, im-

plement the computational framework conceptualized

in Step 6 as a Minimum Viable Product (MVP). The

MVP is a functional version of the solution with ba-

sic functionalities sufﬁcient for initial tests and vali-

dations. This step includes selecting and identifying

databases, developing AI algorithms, system integra-

tion, and preliminary testing.

4 DETAILED EXECUTION:

PHASE GUIDE

A detailed guide on the applicability of the methodol-

ogy deﬁned in Section 3 is presented below.

4.1 Survey

The development of the survey was based on the aca-

demic experience of the project team, as no speciﬁc

survey was identiﬁed in the literature. At the grantor’s

request, a preamble to the survey was prepared, with

guidelines for the application and use of AI mod-

els in public administration. The preamble includes

content on the premises for preparing guidelines, the

deﬁnition of generative AI, applications of generative

AI in government, and the limitations of generative

AI and large language models (LLMs). This survey

was carefully designed to accommodate varying lev-

els of AI maturity in agencies, aiming to collect spe-

ciﬁc knowledge about public sector processes related

to problems and needs that can be framed as strategic

projects and addressed with AI-based solutions.

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

4.2 AHP Express

The Analytic Hierarchy Process (AHP) is a multicri-

teria decision-making method that uses hierarchical

or network structures to represent a decision problem

and develops priorities for the alternatives based on

the decision maker’s judgments (Saaty, 1987). AHP

should be understood as a facilitator, a thought struc-

turing process, rather than an algorithm that solves

problems (Colin, 2007). According to (Colin, 2007),

the application of the AHP comprises four phases:

1. Hierarchy representation: involves the develop-

ment of the decision hierarchy that relates the various

levels of interrelated elements. 2. Pairwise compar-

isons: includes evaluating preferences for each deci-

sion element at a given hierarchy level. 3. Eigen-

vector method: encompasses using the eigenvector

method to estimate the relative weights of decision el-

ements at a particular level and assess the consistency

of preferences established in pairwise comparisons.

4. Aggregation of priorities: concerns the aggre-

gation of relative priorities to evaluate the outcome

related to the objective.

In addition to the formula for calculating the in-

consistency of judgments and the maximum accept-

able level, (Saaty, 1977) indicates that when there are

comparisons involving more than nine elements, the

expectation is that consistency will be low. In line

with the proposal of (Saaty, 2006), Step 2 - Elimina-

tion was devised by the authors of this paper for cases

involving more than nine alternatives.

The step-by-step application of the AHP Express

method is detailed in (Leal, 2020). It follows the prin-

ciples of traditional AHP, whose framework can be

seen in (Saaty, 1987; Colin, 2007). The main dif-

ference is that, as the AHP Express method assumes

judgment consistency, there is no need to calculate in-

consistency. From the judgment of the preferred al-

ternative, obtaining the values of the other pairwise

comparisons through the relationship established in

the initial judgment line is possible. The AHP Ex-

press method was chosen for being suitable for the

agile project management model.

4.2.1 Elimination

This phase was developed to address two main objec-

tives: ﬁrstly, to eliminate AI project proposals identi-

ﬁed during the Survey phase that are deemed infeasi-

ble, and secondly, to classify the remaining proposals

according to their maturity levels in descending order.

A set of criteria and sub-criteria were established to

achieve these goals, comprising a suitable model for

an AI solution. AI specialists, including the authors

of this study, initially formulated these criteria and

subsequently reﬁned them through input from stake-

holders within the NIA. Notable contributors included

leaders from SGD/MGI, MCTI, Enap, Finep, Serpro,

Dataprev and other entities.

Table 1 (Appendix) outlines the speciﬁc criteria

and sub-criteria utilized to assess the feasibility of

each project proposal for AI implementation. These

criteria encompass technical feasibility, data require-

ments, proposed solutions, societal impact, and strate-

gic alignment. Possible answers to the questions are:

YES NO I DON’T KNOW

• If any proposal receives a “NO” response for any

sub-criteria, it is automatically eliminated from

the process.

• “YES” responses are scored as “1”.

• “I DON’T KNOW” responses are scored as “0”.

• All proposals must pass through this elimination

phase.

• At the end, the scores of the proposals will be

summed-up and placed in descending order.

Initially, the nine highest-scoring projects will be

considered for the AHP-Classiﬁer step, as these indi-

cate higher maturity levels. If there is a tie among the

top nine, the criteria weights can be adjusted based

on stakeholder input. It is recommended that weights

be adjusted only in the case of a tie, with the most

relevant criterion assigned a weight of 2.

4.2.2 AHP Express Classiﬁer

The criteria used in the AHP-Express method to clas-

sify/select AI projects from each Government area

are described below. Although only the criteria were

scored, since the methodology needed to be adapted

for agile project management, they were described

through sub-criteria presented to stakeholders during

the AHP application interviews.

1. Urgency of Solution: Time to solve the problem

without AI: How long will it take without AI? Impact

of delay: What are the consequences of not address-

ing the problem? Future urgency: Will the problem

become more urgent?

2. Availability of Data Sources: Internal data qual-

ity: Are internal data reliable? Cost of external data:

How much will external data acquisition cost? Time

to acquire external data: How long will it take?

3. Broader Population Reach: People impacted:

How many people will be affected? Geographical

reach: Will the AI solution be available nationwide?

Accessibility: Will it be accessible across income lev-

els and digital skills?

Implementing AI for Enhanced Public Services Gov.br: A Methodology for the Brazilian Federal Government

4. Greater User Value: User beneﬁts: What advan-

tages does the AI provide? User satisfaction: Is it

user-friendly and does it meet needs? Time saved:

How much time will users save?

5. Projected Access Volume - Cost Estimate: Es-

timate per period: What is the estimated access vol-

ume? Growth curve: How is access expected to grow?

Seasonality: Is demand seasonal?

6. Lower Technical Effort: Problem complex-

ity: Does the problem need advanced AI techniques?

Tools and libraries: Are there tools available for de-

velopment? Expertise: Is there AI expertise? Repli-

cability: Can the solution be used for other problems?

Maintenance: Is maintenance challenging?

7. Ease of Implementation: System integration:

Does the solution integrate with existing systems?

Regulation: Is it compliant with laws?

8. Risks, Ethics, and LGPD: Are risks mapped? Are

there ethical concerns? How does LGPD impact the

solution?

Meetings with process stakeholders are essential

for applying AHP Express. It is recommended to hold

three meetings for each segment (Area, Department,

Ministry): one with expert researchers, one with seg-

ment members, and one with the SGD group, each

consisting of two to four members.

The process starts with presenting up to nine

project proposals from Step 2 - Elimination, the eight

criteria, the Saaty Scoring Table, and a brief expla-

nation of the dynamics. Participants ﬁrst assess the

importance of each criterion in response to the ques-

tion “Which AI project should be selected?” and then

evaluate each project based on these criteria. AHP

Express uses the Saaty scale (1-9), where one indi-

cates equal preference, and nine indicates extreme

preference.

4.3 Problem Framing

To carry out this step, a meeting was held with the

individuals most directly involved with the identiﬁed

and prioritized problem, including people from vari-

ous areas since the problem is cross-functional. These

participants were instructed beforehand to identify the

causes and characteristics of the problem, enabling

a fruitful conversation and ensuring everyone under-

stood its context and relevant aspects. Considering

the project’s objective, the availability of data is es-

sential. Databases with relevant data were analyzed to

illustrate the problem, and the currently implemented

solution was examined to identify strengths and areas

for improvement.

4.4 Design Sprint

Three of the ﬁve traditional Design Sprint steps were

carried out in two days, with a multidisciplinary team

(the Squad) - made up of people involved in the

problem: Understand/Map, Sketch, and Decide. In

“Understand/Map”, the long-term objectives were de-

ﬁned, the questions established for the Sprint were an-

swered, the user map was created, interviews with ex-

perts were carried out, and the target user was deﬁned.

In “Sketch”, there was a division into two groups,

which analyzed all available information and came

up with proposals – the entire group analyzed this in

“Decide”, arriving at a solution deﬁnition.

4.5 Prototyping

This step pertains to the mapping and development of

prototypes considering the agile approach in a demon-

stration environment and the documentation of AI

solutions obtained in Step 5. The experimentation

through prototyping aims to evaluate the feasibility

of an MVP solution. The requirements and speciﬁca-

tions will be developed closely with the requesting en-

tity during the design sprint phase. The Type 2 proto-

type, conceived in Step 6, is delivered through Python

code, developed in a controlled environment that sim-

ulates the user’s real-world problem. This prototype

accompanies a detailed technical report and a presen-

tation for the main stakeholders and interested parties.

4.6 Validation

Validating AI solutions in the public sector demands

a rigorous approach to ensure effectiveness and ef-

ﬁciency. This paper describes the validation pro-

cess of an MVP prototype for an interactive chatbot

providing information and virtual addresses for pub-

lic agency services. MVP validation for AI applica-

tions, like chatbots, requires a systematic process that

combines functionality tests, user evaluations in con-

trolled settings, and real-world deployments. Robust

development, testing, and monitoring tools are es-

sential for successful validation and ﬁnal implemen-

tation, ensuring the solution effectively meets user

needs.

The process begins by deﬁning the chatbot’s ob-

jectives and success criteria, measured by metrics

such as correct response rate, interaction time, and

user feedback. Internal tests validate functionality,

including conversation ﬂow and database integration.

This is followed by user testing in controlled envi-

ronments, gathering qualitative and quantitative feed-

back for improvements. Adjustments are made be-

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

fore large-scale deployment. The chatbot is then beta-

tested in a real environment, where performance is

monitored through continuous user feedback. After

validation, a crucial monitoring plan is established to

ensure the chatbot adapts to new demands and up-

dates, maintaining its effectiveness.

Based on its speciﬁcations, the prototype’s user

interface is also designed for evaluation. This struc-

tured approach ensures AI solutions meet technical

and operational standards and are validated under

near-real conditions, providing a foundation for large-

scale implementations.

5 RESULTS

This study proposes an innovative methodology for

improving Brazilian public services using AI, as out-

lined earlier. Developed in collaboration with SGD

members, AI researchers, the authors, and NIA col-

laborators, this methodology follows the steps de-

tailed in Section 3. Below are the results from each

step of its application.

Initially, discussions between project executors

and stakeholders determined that the Executive Sec-

retaries of the Ministries would select the agencies for

the survey. To ensure the survey was comprehensive

and precise, we meticulously crafted 16 open-ended

questions, focusing on the Service Context, Problem

Characterization, AI Needs, and AI Solution.

The survey was conducted in the Department of

Digital Public Service Platforms (DESER) of SGD,

and three project proposals were raised. Subse-

quently, four people from the General Coordination

of Automation and Service (CGATE), two AI profes-

sors, a facilitator professor, and SGD staff carried out

the elimination step, in which all three projects passed

the ﬁlter. After the ﬁlter, the process proceeded to

the AHP Express. Three groups of respondents were

interviewed for the application of AHP Express: 1)

professors specialized in AI, 2) SGD public servants,

and 3) public servants from the originating area of the

proposal/demand. The respondent teams consisted of

2 to 4 participants. Each stakeholder from each group

had their response recorded during an interview, in

which the questions contained in Table 1 were asked.

In the AHP Classiﬁcation, the three projects selected

in Step 2 were analyzed:

• Automated Recovery of the Gov.br Account ac-

cess

• CGATE citizen support channels

• Citizen support channel for public services on the

Gov.BR Portal

These projects focus on the need for high citizen

interaction through digital platforms, particularly the

Gov.br Portal. Consequently, the classiﬁcation pro-

cess results highlight the importance of impacting a

large portion of the population. The three groups

evaluated the projects based on the criteria in Sec-

tion 4.2.2. The ﬁnal AHP Express results, detailed in

Figure 3, represent the geometric means of the three

scores for each criterion. The project “Citizen service

channel for public services on the Gov.br Portal” was

selected for the prototyping phase.

Figure 3: AHP Express Result.

The Problem Framing and Design Sprint steps

were conducted as described in Sections 4.3 and 4.4.

As a result, after the Problem Framing of Problem 3,

the following answers were reached for Problem 3:

• Where do we want to be in 1 year? Increase

public service engagement rate by 100% through

Gov.Br.

• What is the product? Generative AI chatbot.

• What does the product do? Assists and answers

citizens’ questions about the public service they

need and directs them to the service page.

• What problem is the team trying to solve with

the product? Better direct citizens to the service

they are looking for.

• Who currently uses the product, and who

would we like to use it? Citizens (Gov.br users).

• What is the best phrase to communicate the

long-term goal? Reduce requests to support

channels: Gov account support and ombudsman

by 25%.

The prototyping will involve developing an Ad-

vanced Generative Agent to interact with citizens on

the Gov.br Portal, assisting, clarifying doubts, and di-

recting them to the appropriate service pages. Proto-

typing will occur in two steps, as outlined in Section

4.5. Below are the results of the strategy and archi-

tecture developed by AI experts and stakeholders for

this generative agent:

Data Collection and Structuring: Identify and or-

ganize data communication. Attach and restructure

additional information. Summarize documents. Pro-

pose a FAQ document.

Document Processing: Extract text, topics, key-

words, and structure using document intelligence

tools (e.g., Azure).

Implementing AI for Enhanced Public Services Gov.br: A Methodology for the Brazilian Federal Government

Vector Database: Generate semantic chunking with

layout. Create embeddings. Organize the database by

themes and problems. Deﬁne indexes in the vector

database.

Prompt Embedding: Deﬁne settings for the genera-

tive agent to communicate inclusively with polite, ac-

cessible language tailored to different user proﬁles.

Specify topics the agent won’t respond to, such as po-

litical matters or prejudices. Provide examples for In-

context Learning, like step-by-step guides for public

services, responses about losing Gov.br access, pro-

cesses needing human intervention, and complaint in-

structions. Enable human overﬂow by offering sup-

port links, operating hours, and contact info. Ensure

system status is informed to users before starting a

conversation.

Advanced Semantic Search: To enhance the gen-

erative agent’s precision, a speciﬁc LLM, LLaMA,

trained on a Portuguese-language database, will be

used. This model will optimize Retrieval-Augmented

Generation with Fine-Tuning (RAFT) (Zhang and

et al., 2024), combining supervised ﬁne-tuning with

RAG (Lewis and et al., 2020). RAFT adjusts train-

ing data for better question responses using Chain-of-

Thought reasoning from speciﬁc documents, enhanc-

ing RAG by integrating external database retrieval

with pre-trained models.

The proposed RAFT solution, which combines su-

pervised ﬁne-tuning with RAG, introduces an inno-

vative approach to improving the accuracy of inter-

actions between generative AI and citizens. Validat-

ing a chat intelligent agent with an LLM includes

testing its functionality, performance, user experi-

ence, domain-speciﬁc knowledge, ethical considera-

tions, benchmark comparisons, and ongoing monitor-

ing to ensure it meets requirements and provides sat-

isfactory interactions (Xi et al., 2023). At the time of

writing, Step 8 is under development, as described in

Section 4.6.

6 DISCUSSION

This paper presents a comprehensive methodology for

deploying AI solutions in public services, incorpo-

rating Lean Ofﬁce waste elimination methods, pro-

cess optimization, and standardization before evaluat-

ing workﬂow automation. Once automated, a contin-

uous improvement (kaizen) loop optimizes the NLP

chatbot’s workﬂow. Digitization enables the automa-

tion of standardized processes, eliminating waste and

embedding intelligence through smarter control algo-

rithms. Key questions addressed include the link be-

tween operational excellence in public administration

and AI in workﬂows and the driving forces behind im-

proved performance and efﬁciency in public services

through AI.

Beyond portal access beneﬁts, the Brazilian gov-

ernment can gain additional advantages. Using Webo-

metrics (Saeidnia et al., 2024), interactions between

citizens and the generative agent will produce valu-

able data. Authorities can leverage this to:

• Identify patterns in citizen behavior, needs, and

preferences.

• Understand popular services and common com-

plaints.

• Create data-driven public policies.

• Personalize services and improve efﬁciency.

• Monitor service trends and conversion rates for

successful demand resolutions.

• Enhance metrics for “Gov.br” portal usage.

Data privacy and security are crucial when apply-

ing AI in public administration. The information col-

lected covers healthcare, smart cities, and public ser-

vice forecasting at varying levels of maturity. These

challenges emphasize the need for robust safeguards

to protect sensitive data and ensure adherence to data

protection regulations.

Other projects have been selected using the pro-

posed methodology across different agencies and sec-

retariats. To harmonize the selection process at a na-

tional level, one option is to apply uniform weights

to all criteria for each project. This would standard-

ize the selection process, ensuring that each crite-

rion is equally evaluated from a national perspective

rather than from an individual secretariat’s viewpoint.

Consequently, the process would better align with

the broader objective: identifying which AI project

should be prioritized for Brazil as a whole.

7 CONCLUSION AND FUTURE

WORK

This paper presented an innovative methodology for

implementing AI solutions in Brazilian public ser-

vices, aimed at enhancing efﬁciency, accessibility,

and service quality. Integrating Lean Ofﬁce, Design

Sprint, and advanced AI techniques, the methodology

is both ﬂexible and adaptable to diverse governmen-

tal needs, ensuring inclusivity. Collaboration among

AI experts, public servants, and stakeholders was key

to ensuring technical feasibility and alignment with

public policies.

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

The methodology addresses a speciﬁc demand

from the SGD, part of the National Digital Gov-

ernment Strategy. Given Brazil’s numerous federal

agencies, it includes the selection and prototyping of

projects within a complex government structure. Its

ﬂexibility allows application in other countries with

extensive governmental frameworks, promoting digi-

tal transformation in varied contexts.

The development of a generative agent for the

Gov.br Portal meets the demand and transforms cit-

izen interactions into valuable data. This data pro-

vides insights for personalizing services. We propose

using LLaMA X, trained on Portuguese documents,

with RAFT for improved accuracy. Future work in-

cludes developing a ChatGPT-based AI for compar-

ison and exploring integrations with programs like

Bolsa Fam

ılia and systems such as CPF (Social Secu-

rity Number), CNPJ (National Registry of Legal En-

tities), SEI (Electronic System of Information), and

institutions payroll.

ACKNOWLEDGEMENTS

Acknowledgments and recognition are extended to

the Ministry of Management and Innovation in Public

Services, especially the Secretaria de Governo Digi-

tal (SGD), for ﬁnancially supporting this project and

enabling the practical feasibility of solution identiﬁ-

cation and prototyping.

REFERENCES

Anna, U. (2023). Artiﬁcial intelligence landscape in south

america. AARMS–Academic and Applied Research in

Military and Public Management Science, 22(1):101–

114.

Atabekov, A. (2023). Artiﬁcial intelligence in contempo-

rary societies: legal status and deﬁnition, implemen-

tation in public sector across various countries. Social

Sciences, 12(3):178.

Bardwell, L. V. (1991). Problem-framing: A perspective on

environmental problem-solving. Environmental Man-

agement, 15(5):603–612.

Birhane, A., Kalluri, P., Card, D., Agnew, W., Dotan, R.,

and Bao, M. (2022). The values encoded in machine

learning research. In Proceedings of the 2022 ACM

Conference on Fairness, Accountability, and Trans-

parency, pages 173–184.

Blount, K. (2024). Using artiﬁcial intelligence to prevent

crime: implications for due process and criminal jus-

tice. AI & SOCIETY, 39(1):359–368.

BRASIL (2023). Decreto Nº 11.437, de 17 de marc¸o

de 2023. https://www.planalto.gov.br/ccivil 03/

ato2023-2026/2023/decreto/D11437.htm#art5. [On-

line; accessed 27-June-2024].

BRASIL, M. d. C. T. e. I. (2018). Estrat

egia Brasileira

para a Transformac¸

ao Digital. https://www.gov.br/

mcti/pt-br/centrais-de-conteudo/comunicados-mcti/

estrategia-digital-brasileira/estrategiadigital.pdf.

BRASIL, M. d. G. e. I. e. S. P. (2024). Servic¸os e

informac¸

oes do brasil. https://www.gov.br/pt-br. Ac-

cessed: 27 June 2024.

BRASIL, Minist

erio da Ci

encia, T. e. I. (2021). Estrat

egia

brasileira de intelig

encia artiﬁcial: Ebia. https://bit.

ly/3C6LgY4.

Chen, T., Gasc

o-Hernandez, M., and Esteve, M. (2024).

The adoption and implementation of artiﬁcial intelli-

gence chatbots in public organizations: Evidence from

us state governments. The American Review of Public

Administration, 54(3):255–270.

Chen, Y.-C., Ahn, M. J., and Wang, Y.-F. (2023). Arti-

ﬁcial intelligence and public values: value impacts

and governance in the public sector. Sustainability,

15(6):4796.

Cohen, S., Manes Rossi, F., and Brusca, I. (2023). Debate:

Public sector accounting education and artiﬁcial intel-

ligence. Public Money & Management, 43(7):725–

726.

Colin, E. C. (2007). Pesquisa operacional: 170 aplicac¸

oes

em estrat

egia, ﬁnanc¸as, log

ıstica, produc¸

ao, market-

ing e vendas. Livros T

ecnicos e Cient

ıﬁcos.

Desordi, D. and Della Bona, C. (2020). A intelig

encia arti-

ﬁcial e a eﬁci

encia na administrac¸

ao p

ublica. Revista

de Direito, 12(02):01–22.

Dwivedi, Y. K., Sharma, A., Rana, N. P., Giannakis, M.,

Goel, P., and Dutot, V. (2023). Evolution of artiﬁ-

cial intelligence research in technological forecasting

and social change: Research topics, trends, and fu-

ture directions. Technological Forecasting and Social

Change, 192:122579.

Engstrom, D. F., Ho, D. E., Sharkey, C. M., and Cu

ellar,

M.-F. (2020). Government by algorithm: Artiﬁcial

intelligence in federal administrative agencies. NYU

School of Law, Public Law Research Paper, (20-54).

Fan, Z., Yan, Z., and Wen, S. (2023). Deep learning and ar-

tiﬁcial intelligence in sustainability: a review of sdgs,

renewable energy, and environmental health. Sustain-

ability, 15(18):13493.

Faria, A. V. A., de Melo, M. K., de Oliveira, F. A. R.,

Weigang, L., and Celestino, V. R. R. (2022). Auto-

mated slr with a few labeled papers and a fair work-

load metric. In International Conference on Web

Information Systems and Technologies, pages 1–23.

Springer.

Filgueiras, F. and Junquilho, T. A. (2023). The brazilian

(non) perspective on national strategy for artiﬁcial in-

telligence. Discover Artiﬁcial Intelligence, 3(1):7.

Gaspar, W. B. and Mendonc¸a, Y. C. d. (2021). A in-

telig

encia artiﬁcial no brasil ainda precisa de uma es-

trat

egia.

Gov, A. (2024). Gov.br

e a p

agina de

governo mais acessada do mundo.

Implementing AI for Enhanced Public Services Gov.br: A Methodology for the Brazilian Federal Government

https://agenciagov.ebc.com.br/noticias/202403/gov-

br-e-a-pagina-de-governo-mais-acessada-do-mundo.

Accessed: 27 June 2024.

Guedes, L. and J

unior, M. O. (2024). Artiﬁcial intelligence

adoption in public organizations: a case study. Fu-

ture Studies Research Journal: Trends and Strategies,

16(1):e860–e860.

Jevinger,

A., Zhao, C., Persson, J. A., and Davidsson,

P. (2024). Artiﬁcial intelligence for improving pub-

lic transport: a mapping study. Public Transport,

16(1):99–158.

Kitsios, F., Kamariotou, M., and Mavromatis, A. (2023).

Drivers and outcomes of digital transformation: The

case of public sector services. Information, 14(1):43.

Knapp, J., Zeratsky, J., and Kowitz, B. (2016). Sprint: o

etodo usado no Google para testar e aplicar novas

ideias em apenas cinco dias. Editora Intr

ınseca Ltda.

Kubota, L. C. and Rosa, M. B. (2024). Intelig

encia

artiﬁcial no brasil: Adoc¸

ao, produc¸

ao cient

ıﬁca

e regulamentac¸

ao. digitalizac¸

ao e tecnologias da

informac¸

ao e comunicac¸

ao. Digitalizac¸

ao e tecnolo-

gias da informac¸

ao e comunicac¸

ao: oportunidades e

desaﬁos para o Brasil. Rio de Janeiro: Ipea.

Kutyauripo, I., Rushambwa, M., and Chiwazi, L. (2023).

Artiﬁcial intelligence applications in the agrifood sec-

tors. Journal of Agriculture and Food Research,

11:100502.

Larusdottir, M., Roto, V., and Cajander, A. (2021). Intro-

duction to user-centred design sprint. In Ardito, C.

e. a., editor, Human-Computer Interaction – INTER-

ACT 2021, volume 12936 of Lecture Notes in Com-

puter Science, pages 283–293. Springer, Cham.

Leal, J. E. (2020). Ahp-express: A simpliﬁed version of the

analytical hierarchy process method. MethodsX, 7.

Lemes, M. M. and Lemos, A. N. L. E. (2020). O uso

da intelig

encia artiﬁcial na sa

ude pela administrac¸

ublica brasileira. Cadernos Ibero-Americanos de Di-

reito Sanit

ario, 9(3):166–182.

Lewis, P. and et al. (2020). Retrieval-augmented genera-

tion for knowledge-intensive nlp tasks. In Advances in

Neural Information Processing Systems, volume 33,

pages 9459–9474.

Maragno, G., Tangi, L., Gastaldi, L., and Benedetti, M.

(2023). Exploring the factors, affordances and con-

straints outlining the implementation of artiﬁcial in-

telligence in public sector organizations. International

Journal of Information Management, 73:102686.

Mendonc¸a de S

a Ara

ujo, C. M., Miranda Santos, I.,

Dias Canedo, E., and Favacho de Ara

ujo, A. P. (2019).

Design thinking versus design sprint: A comparative

study. In Marcus, A. and Wang, W., editors, Design,

User Experience, and Usability. Design Philosophy

and Theory, volume 11583 of Lecture Notes in Com-

puter Science, pages 335–344. Springer, Cham.

Neumann, O., Guirguis, K., and Steiner, R. (2024). Explor-

ing artiﬁcial intelligence adoption in public organiza-

tions: a comparative case study. Public Management

Review, 26(1):114–141.

Nonato, S. D. O., de Souza Figueiredo, S. d. S.,

de Melo Alves, C. A., and Dias, C. N. (2024).

Cooperac¸

ao interorganizacional em planos es-

trat

egicos de intelig

encia artiﬁcial: uma an

alise

comparativa. Future Studies Research Journal:

Trends and Strategies, 16(1):e838–e838.

OECD.AI (2022). Ai use case in the public sector. Re-

trieved on 29-Jun-2024.

Robertson, S. I. (2016). Problem Solving: Perspectives from

Cognition and Neuroscience. Psychology Press, 2nd

edition.

Rocha, C. A. A., Weigang, L., Dib, M. V. P., Faria, A. V. A.,

Cajueiro, D. O., de Melo, M. K., and Celestino, V.

R. R. (2022). Leveraging transfer learning for long

text classiﬁcation with limited data. In International

Conference on Web Information Systems and Tech-

nologies, pages 98–120. Springer.

Saaty, R. W. (1987). The analytic hierarchy process—what

it is and how it is used. Mathematical Modelling,

9(5):161–176.

Saaty, T. L. (1977). A scaling method for priorities in hier-

archical structures. Journal of Mathematical Psychol-

ogy, 15.

Saaty, T. L. (2006). Rank from comparisons and from rat-

ings in the analytic hierarchy/network processes. Eu-

ropean Journal of Operational Research, 168(2):557–

570.

Saeidnia, H. R., Hosseini, E., Abdoli, S., and Ausloos, M.

(2024). Unleashing the power of ai: a systematic re-

view of cutting-edge techniques in ai-enhanced scien-

tometrics, webometrics and bibliometrics. Library Hi

Tech.

Samoili, S., Cobo, M. L., G

omez, E., De Prato, G.,

Mart

ınez-Plumed, F., and Delipetrev, B. (2020). Ai

watch. deﬁning artiﬁcial intelligence. towards an op-

erational deﬁnition and taxonomy of artiﬁcial intelli-

gence.

Scutella, M., Plewa, C., and Reaiche, C. (2024). Vir-

tual agents in the public service: examining citizens’

value-in-use. Public Management Review, 26(1):73–

88.

Straub, V. J., Morgan, D., Bright, J., and Margetts, H.

(2023). Artiﬁcial intelligence in government: Con-

cepts, standards, and a uniﬁed framework. Govern-

ment Information Quarterly, 40(4):101881.

Wang, C., Han, B., Patel, B., and Rudin, C. (2023). In pur-

suit of interpretable, fair and accurate machine learn-

ing for criminal recidivism prediction. Journal of

Quantitative Criminology, 39(2):519–581.

Xi, Z., Chen, W., Guo, X., He, W., Ding, Y., Hong, B.,

Zhang, M., Wang, J., Jin, S., Zhou, E., et al. (2023).

The rise and potential of large language model based

agents: A survey. arXiv preprint arXiv:2309.07864.

Yu, K.-H., Beam, A. L., and Kohane, I. S. (2018). Artiﬁ-

cial intelligence in healthcare. Nature biomedical en-

gineering, 2(10):719–731.

Zhang, T. and et al. (2024). Raft: Adapting lan-

guage model to domain speciﬁc rag. arXiv preprint

arXiv:2403.10131.

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

100

APPENDIX

Table 1: Criteria and Subcriteria for Feasibility of AI Implementation in Public Services.

Criteria Subcriteria

1- Is There Technical Feasibility?

a. Cost: The AI solution’s development, implementation, and mainte-

nance are economically viable.

b. Time: The solution can be implemented within the PBIA timeline;

c. Maintenance: The AI solution is feasible to maintain and update;

d. Operationalization: Necessary resources for implementation (human,

material, infrastructure) are available;

2- Will the data not be a limiting

factor for the model?

a. Data Quality: Some data patterns can be learned to help solve the

problem;

b. Data Quality: The available data tend to be sufﬁciently accu-

rate, complete, and reliable once preprocessed so as not to impede the

progress of the project;

c. Data Quantity: There is sufﬁcient data to train and test the AI solu-

tion.

d. Data Diversity: The data signiﬁcantly represent the target popula-

tion’s characteristics.

3- Can AI improve the solution?

a. Effort: The current solution to the problem is unsatisfactory, and the

AI-based solution will increase process efﬁciency.

b. Effort: The solution cannot be found using simple rules (e.g., through

spreadsheets, If..else) or basic statistical analysis.

c. Effort: There is a need for prediction, classiﬁcation, or the use of

language models (like ChatGPT) based on input data.

d. Measurability: The outcome of the process and/or service is measur-

able.

4- Will there be a good Impact on

society?

a. Beneﬁts: The AI solution will bring real beneﬁts to society, either

directly or indirectly through improved management;

b. Threats: The AI solution considers and mitigates negative impacts

on society;

c. Acceptance: The AI solution will be accepted by society;

d. Regulation: The AI solution complies with existing laws and regula-

tions.

5- Is it in line with Strategic Align-

ment?

a. Alignment with the organisation’s mission: The AI solution aligns

with the organisation’s mission;

b. Strategic priorities: The AI solution contributes to the organisation’s

strategic priorities;

c. Reputation risk: The AI solution will not harm the organisation’s

reputation;

d. Governance: The solution contributes to advancing speciﬁc public

policies and government-established goals.

Implementing AI for Enhanced Public Services Gov.br: A Methodology for the Brazilian Federal Government

101