Requests for the

Application of AI in
OCB Data Management
Beyond ChatGPT: Embracing a Spectrum of AI
Innovations for OCB's Operations

Intro
This document serves as a comprehensive overview of the innovative artificial intelligence (AI)
solutions proposed or requested by various departments within OCB. These solutions extend beyond
well-known applications like ChatGPT and Copilot, delving into a diverse spectrum of AI technologies
that hold the potential to revolutionize our operational landscape.

Our goal is to systematically address these needs, recognizing their critical impact on both
immediate and future operational efficiency. Each section of this document is dedicated to exploring
different AI technologies—from machine learning models and natural language processing tools to
predictive analytics and automation systems. This approach aims to ensure holistic enhancement of
our capabilities across all levels of operation.

By the conclusion of this document, I hope we can set a clear path for the implementation of these
diverse AI technologies. Our stakeholders need to understand the breadth and depth of AI's
potential beyond the familiar realms of ChatGPT and Copilot. We need to pave the way for
transformative improvements in our processes and services with AI.

AI represents the biggest revolution we have yet to fully encounter. It presents as many
opportunities as it does challenges and risks. Various groups, including Intersectional teams, IKM,
PMO, and operational teams, are trying to address these within their scopes. The purpose of this
document is to ensure that the following six requests do not get lost amid these efforts.
AI and GDPR at MSF
The integration of AI within our operations attracts considerable attention due to the General Data
Protection Regulation (GDPR). We recognize that many of the proposed AI solutions will require
access to data across the entire organization, necessitating stringent management practices to
ensure compliance with GDPR.

As we venture deeper into the realm of artificial intelligence, it is imperative that we manage these
data interactions in an intersectional manner. This means coordinating across various departments
and teams to ensure that all AI implementations adhere to GDPR standards and protect the privacy
and integrity of our data.

The proposals listed below have the appealing aspect of not being subject to the constraints
highlighted above. These AI solutions have been designed with a focus on minimizing GDPR
compliance risks by limiting the scope of data they access or by employing data processing
techniques that enhance data anonymity and security.
Deleted: <#>¶
Disclaimer
We are not experts at AI and don’t pretend to be. We aim to call for help from experts on these
topics.

We may make some wrong assumption or supposition in the following statements.

We used ChatGPT to reduce this report's size by half.

Request 9: Using AI to Facilitate the

Adop?on of Data Science Techniques by
Analysts
Background and Request
Our business analysts and data analysts, although proficient in their domains, are not specialized in
data science. There is an increasing need to enable them to utilize advanced data science techniques
such as predictive modeling, regression, and forecasting to improve decision-making and business
strategy.

Requestors:

- Miguel Cuenca - SCO Coordinator

- Tim Van Roeyen - Supply Data Analyst
- Mina Kanashiro - Head of Supply Technical and Information Systems

Responsible: ??? Commented [MK1]: What would be the role of the

responsable? If we are talking about sponsor to whom the
ICT referents should address to, it could be either myself of
Miguel.
Proposed AI Solution
Commented [YL2R1]: Hi Mina, I think this is about the
Implement an accessible artificial intelligence platform that acts as an interface between advanced person to be considered the sponsor if this request becomes
an official project (with a SteerCo etc) or, if this level of
data science techniques and our analysts. This platform can include automated machine learning complexity is not required, the sponsor/responsible pushing
tools that simplify the use of complex models without requiring deep expertise in programming or this forward
statistics.

Key Features of the Platform

1. User-Friendly Interface: A simple graphical interface allowing analysts to enter data, select
models, and interpret results without coding.
2. Model Automation: Use of AutoML to automatically select the best models and parameters
based on the provided data.
3. Tutorials and Support: Provision of integrated educational resources to help users
understand the models and their applications.
AI Expertise and Technologies Used
The platform relies on AutoML technologies that reduce the entry barrier for non-specialists by
enabling them to generate high-quality models. Systems like Google Cloud AutoML or Microsoft
Azure Automated Machine Learning can be used to implement this solution.

Expected Impact
By enabling our analysts to use advanced data science tools, we anticipate a significant improvement
in data analysis, quicker and more informed decision-making, and an increase in added value for our
projects and clients.
Request 2: Development of AI for
Multilingual Querying of Internal and
Confidential Documents
Commented [MK3]: After discussion with Mélanie and
feedback from Walesa (Front Office), we agreed that this
request should be broader. It will not only cover the
translation aspect, but also solution search. Please refer to
Request 2.1 as a proposition, including the original email
from Walesa P I sent to Nicolas.
Background and Request
Given the dispersed nature of our document repositories and the critical importance of accessible
and usable information for decision-making and innovation, it is crucial to leverage AI to support
advancements in information and knowledge management (IKM) within MSF. This entails tackling
two significant challenges: the translation of vast amounts of information and the interpretation of
resources across our diverse platforms. Hence, we require a comprehensive AI solution that can
seamlessly translate documents into multiple languages and interpret information to assist users in
navigating our platforms effectively.
Requestors:
- Melanie Vandevoir - knowledge management and internal communication officer
- Mina Kanashiro - Head of Supply Technical and Information Systems
Responsible:
- Melanie will synchronize with IKM Steerco.

- Proposed AI Solution
- To address these challenges, we propose the development of an advanced AI-driven system
and search engine. This system will integrate state-of-the-art natural language processing
(NLP) algorithms and machine learning models to automate the translation of documents
into various languages and interpret complex data from disparate sources. Additionally, it
will provide personalized guidance to users, offering relevant documents, contextualized
information, and directing them to technical experts or community spaces for further
assistance.

Key Features of the System

- Multilingual translation: the AI system will utilize advanced NLP techniques to accurately
translate documents into multiple languages, ensuring accessibility for all employees
regardless of their work language.
- Information interpretation and guidance: through sophisticated algorithms, the system will
analyze and interpret information across platforms, providing users with tailored
recommendations, relevant documents, and guidance.
- Integration with existing platforms: the AI solution will seamlessly integrate with MSF's
existing platforms, such as the OCB Intranet, to enhance accessibility and usability without
disrupting established workflows.
 - Continuous learning and improvement: leveraging machine learning capabilities, the system
will continuously learn from user interactions and feedback to refine its recommendations
and enhance overall performance over time.
Contextualization and personalization: by understanding user preferences and contextual
factors, the system will deliver personalized recommendations and insights, optimizing the
relevance and usefulness of the information provided.
AI Expertise and Technologies Used
- To be defined

Expected Impact
- Enhanced knowledge exchange: by breaking down language barriers and providing tailored
guidance, the system will facilitate the exchange of critical knowledge and expertise.
- Improved decision-making: access to timely and relevant information, coupled with
personalized insights, will empower MSF staff to make more informed decisions, ultimately
improving the effectiveness and efficiency of operations.
- Optimized resource utilization: by streamlining information retrieval and interpretation
processes, the AI system will enable MSF personnel to utilize their time and expertise more
efficiently, leading to greater productivity and resource optimization.
- Fostered innovation: by democratizing access to information and expertise, the system will
stimulate innovation within MSF, fostering the development of solutions to challenges and
driving continuous improvement across the organization.

Request H: Data Analysis for EveryOne

Background and Request
The supply dept aims to democratize the analysis of its data using AI.
Currently, our users access supply chain data mainly through PowerBI reports. There is a growing
demand for a more intuitive and direct method of querying this data, allowing users to make natural
language queries to get real-time information such as stock valuation assessments.

Requestors:

- Miguel Cuenca - SCO Coordinator

- Tim Van Roeyen - Supply Data Analyst
- Mina Kanashiro - Head of Supply Technical and Information Systems

Responsible: ???

Proposed AI Solution
Develop and integrate a natural language processing (NLP) functionality that allows users to ask
questions in natural language directly within the PowerBI interface or through a dedicated
application. This AI will interpret the questions, access relevant data, and provide real-time answers.
Key Features of the System
1. Natural Language Query Interface: Implementation of a user interface where questions can
be asked in natural language.
2. Integration with PowerBI: Connection of the NLP AI with data already processed and
visualized in PowerBI to extract and present responses.
3. Real-Time Responses: The system's ability to process queries and provide answers based on
the latest available data.

AI Expertise and Technologies Used

Use of NLP models such as GPT (OpenAI) or BERT (Google) to understand and generate responses
from complex data. These models can be integrated into PowerBI through specific APIs or plugins
that facilitate the interaction between users and backend databases.

Expected Impact
- Implementing this technology will increase the responsiveness of decisions based on supply
chain data by allowing dynamic and natural language queries. This will enhance the user
experience by simplifying access to critical information and speeding up the decision-making
process.
- Supply Data are currently pre-processed and available through the DART platform (Data
Analytic Report Tool). Data are qualitative. The finance and the HR are on their way to follow
the same way that the Supply Department, we will be able to leverage on supply experience
for the other departments.
Request K: AI Analysis of Documents
Shared on the Public Websites
Background and Request
The Stockholm Unit [SEU] would like to be able to gather data from all our public reports i.e. the
ones SEU have managed and that are published on our website www.evaluation.msf.org. SEU wants
then to be able to ask AI to conduct specific analytics such as “tell me everything the reports say
about human resource management” (summarizing) or “what have been the main inhibitors to
projects reaching vulnerable groups” (more analytical). The reports are already public and therefore
there are no data protection concerns.

It is essential to be able to analyze the content of these documents to summarize thematic issues
from the reports, as well as to be able to extract more analytical and/or strategic insights, but this
task is time-consuming if performed manually, and we would like it to be more accessible, more
broadly.

Requestor:

- Linda Öhman - Head of the Stockholm Evaluation Unit

Responsible:

- How can we move forward???

Proposed AI Solution
Implement an AI system capable of reading and analyzing all documents shared on our website. This
system would use natural language processing techniques to understand the content of the
documents and allow users to ask specific questions about these data using natural language.

Key Features of the System

1. Automatic Data Extraction: The AI would scan and extract text from all publicly available
documents on the website.
2. Content Analysis: Use of NLP to analyze the text, identify main themes, expressed
sentiments, and other relevant metrics.
3. Natural Language Query Interface: Allowing users to formulate questions in human language
to query the analyses made, such as "What are the main topics addressed in the latest
annual report?"

AI Expertise and Technologies Used

The use of advanced NLP models based on Transformer architectures (such as BERT or GPT) allows
for efficient processing of large amounts of text. These technologies can be directly integrated into
our web infrastructure for continuous and real-time processing of new documents published.
Similar initiatives
- We are not aware of solution developed at MSF that fits 100% for this concern.
- While waiting for a fit-for-purpose solution, OCA and MSF USA did enable a protected
ChatGPT solution https://ai.msf.org/.
o Today it is not available to all for cost management concerns.
o Can it learn from SEU data? How to make it available for the SEU team [7 people]?

Expected Impact
- This deep analysis capability would not only improve internal understanding of trends and
perceptions but also quickly respond to specific questions, thereby aiding in access to
evaluation findings, providing input for learning to support in formulating strategies based
on up-to-date and precise data. Operational efficiency would also be enhanced by reducing
the time required to manually analyze content. The value of evaluation could increase since
access to findings would be augmented.
Request M: Analysis of Social Networks and
Media to Understand Contexts of Ac?on
and Analyze Sen?ments
Background and Request
To better understand the impact of our organization in various contexts and to monitor public
opinion about us, it is essential to set up tools for analyzing social networks and media. This analysis
will allow us to adjust our strategies based on observed trends and sentiments expressed by the
public and officials. It can also allow us to detect any artificial campaigns which could affect our
operations (e.g. Disinformation campaigns against vaccinations).

Requestor:

- Sandrine Tiller - Coordo of Humanitarian access & Negotiation Team

Interested:

- Natacha Loisse - Digital operational communications officer

- Ailin Martinez – Digital Strategist

Proposed AI Solution
Develop an artificial intelligence system dedicated to real-time analysis of posts on social networks
and in the media. This system will use natural language processing to extract and analyze
sentiments, bad intentioned campaigns (incl. artifical ones) and discussion topics (ex. Malaria and
Vaccines) related to our operations. Commented [ST4]: Can we also use it to detect artificial
campaigns (eg. Disinformation) on topics (eg. Malaria,
AI to update our context analysis. Daily / weekly emails with key evolutions in the humanitarian vaccines)
landscape, particularly useful for emergencies (eg. People being displaced, military attacks). This is
more straightforward, a bit like a news service on vitamins.

Key Features of the System

1. Continuous Monitoring: The system continuously scans social media platforms and media
publications to gather mentions of our organization and relevant discussions around civil
society or humanitarian issues.
2. Sentiment Analysis: Use of NLP techniques to assess the sentiments expressed in the
mentions and publications, classifying these sentiments into categories such as positive,
negative, or neutral.
3. Trend and Topic Detection: Identification of emerging topics and key discussions prevailing
in the media and on social platforms, enabling strategic responsiveness.
4. Sending Monthly to Head Of Mission and Cell with resume on discussion on the country (and
the region) – in ‘hot’ contexts, where we are seeing more anti-NGO sentiment, this may be
more frequent.
AI Expertise and Technologies Used
The use of NLP models such as BERT or GPT-3 for sentiment analysis and named entity recognition
enhances the accuracy of information extraction. Integrating these models into media monitoring
systems allows for processing large volumes of textual data and identifying insights in real-time.

Similar initiatives
A such tool has been developed as a personal initiative by a Data Engineer in OCA but not yet
industrialize, we would like to move forward with him on this initiative

Expected Impact
- This technology will enable us to stay at the forefront of media and social discussions
concerning our sector, offering a deeper understanding of public perceptions. It will also
facilitate proactive decision-making based on an analysis of current sentiments and trends,
contributing to improving our public image and communication strategy.
- We did not yet receive any formal request nor from the ops nor from the com but it looks
obvious that such tools could benefits them by having more immediate access to relevant
information.
Request S: Integra?on of MicrosoT Copilot
in PowerBI for Advanced Data Analysis
Background and Request
In our constant quest to improve the efficiency and intuitiveness of our data analyses, it is essential
to enable our users to gain deeper and more accessible insights from our complex data sets. The
increased use of artificial intelligence to assist our users in interpreting and visualizing data is a
priority.

Requestor: 17 OCB Data Analysts – advanced Power Bi Users

Responsible: Nicolas Degroodt
Status: Request sent to OCT ICT Coordination

Proposed AI Solution
Integrate Microsoft Copilot into our PowerBI environment. Copilot is an AI-based solution that
allows users to make natural language queries and receive data visualizations and in-depth analyses
automatically. This technology transforms how users interact with data, making analysis both
simpler and more powerful.

Key Features of the System

1. Natural Language Queries: Allows users to ask questions directly in natural language, such as
"What was the sales trend last quarter?" and receive answers in the form of automatically
generated visualizations and reports.
2. Proactive Suggestions: Copilot can suggest analyses and visualizations based on trends it
identifies in the data, thus offering insights that are not solicited but potentially valuable.
3. Continuous Improvement: The system learns from user interactions to enhance its
suggestions and responses over time.

AI Expertise and Technologies Used

Microsoft Copilot is powered by cutting-edge natural language processing and machine learning
technologies, enabling complex understanding and interaction with data. Integration with PowerBI
leverages existing BI capabilities while enriching them with advanced AI capabilities.

Technical Challenges
Although we are already using Microsoft Azure, our current single-tenant setup for the entire
organization presents specific challenges for enabling Microsoft Copilot in PowerBI. This
configuration limits our flexibility to seamlessly integrate new AI solutions without impacting the
entire organizational environment. The international office mandates a group to find out global
solution.
This Could takes a while. In the meantime we could investigate around the creation of sub-tenant or
alternative tenant.

Expected Impact
The adoption of Microsoft Copilot in PowerBI is expected to revolutionize access to information and
decision-making within our organization, making data analysis more intuitive and less technical. This
will allow a larger number of employees to leverage data for strategic insights without the need for
deep data expertise.

Request V: Development of AI for following

Procurement Tender Processes
Background and Request
The procurement process is very sensitive and sometimes underestimated when sending
RFQs, as it generates a considerable amount of data that is usually lost. For example, the ratio
of RFQs sent to those received can give us an indication of the accuracy of the selection of
suppliers invited or the quality of the material sent. Being able to intercept and store this
information through AI can certainly mean the possibility of improving procurement activities
(quality, risk, savings and people skills) and attention to this improves the transparency of the
tendering process by reducing the possibility of giving different information between
suppliers. Formatted: English (US)

Requestors: Andrea Grilli - Regional Procurement Manager Middle East

Responsible: Aline de Andrade – Head of Procurement

Proposed AI Solution
Imagine AI integrated with Outlook, Power BI and Teams to improve the tendering (by email)
process. This AI would help formulate templates, track responses, and even generate
automated statistics on RFx processes, duration, and supplier responses. It could even
automatically generate a categorised and country-specific supplier list.

Key Features of the System

• Drafting RFQs: Copilot for Sales can help you draft RFQs by providing templates or suggesting
language based on past successful RFQs. It can also help ensure that all necessary information
is included in the RFQ. Formatted: English (US)
• Tracking Responses: While Copilot for Sales cannot directly track email responses, it can help
you stay organized by reminding you to follow up on sent RFQs after a certain period of time.
• Analyzing Supplier Responses: Copilot for Sales can help analyze responses to RFQs by
highlighting key information, such as pricing details or delivery timelines, which can assist in
the decision-making process.
AI Expertise and Technologies Used
The use of Copilot for sale, Copilot, Power BI, Outlook

Expected Impact
In the field. To be able to speed up the tendering process, to enable interested parties to be
more efficient, not to lose information, to organise work,
At head office, to receive "free of charge" all the information in real time on the tenders in
progress, those already carried out and the suppliers contacted, in order to create a single
database.