Data Underpins the E, S, and G Strategy

In 2024, we are bearing witness to the launch of a plethora of new regulatory requirements in sustainability topics around the world. Several countries are bringing into law a variety of requirements that obligate companies to demonstrate robust due diligence and management of human rights issues within their international supply chains, as well as in other areas such as meeting environmental targets.

The demands are unlikely to wane, and companies will be held to account by a community of their rights holders from employees to customers to shareholders and regulatory bodies. We find ourselves navigating a business environment ill-suited for the faint of heart!

Navigating the road to success requires organizations to prioritize and invest their limited resources wisely, but where should they start?

The growing discourse on artificial intelligence (AI) highlights how business and industry leaders are evaluating ways in which technology might solve a plethora of challenges within supply chains and decarbonization initiatives for sustainable finance, for example.

The topic begs two important questions:

1. Are we overlooking the most obvious gap — data?

2. Does technology, AI in particular, offer a genuine case for catalytic change or is it over-hyped?

Data, a beginning, or an ending?

Before embarking on any initiative, identifying the fundamental issue to be solved is the critical first step. Once the issue is identified, knowing what information is needed to fully understand and address it quickly follows.

This is the data story, underlying every issue in life. Be it human rights, world peace, health, happiness, or financial stability; all are critically dependent on knowing the factors that influence, support, and/or hinder these initiatives.

What must follow then, is an understanding of data availability and readiness and an understanding of how data, and the analysis of that data, will aid problem resolution.

A grasp of these opening areas will ignite additional questions around rights holders (aka stakeholders), decision usefulness, organizational readiness, and outcomes.

A very human-based public interest project, iEarths tackled these specific issues shared here to illustrate an application use case.

The human rights backstory

The International Labour Organization (ILO), International Organization for Migration (IOM), and international human rights group Walk Free published a report (2022) indicating about 50 million people live in modern slavery, of which 28 million work under forced labor conditions. 

Finance Against Slavery & Trafficking (FAST) a United Nations University Centre for Policy Research initiative indicates that to “bring this figure close to zero by 2030 – to meet the UN Sustainable Development Goals Target 8.7 – we would need to reduce the number of people affected by around 10,000 individuals per day.” The weight of this statistical fact became the catalyst for the creation of iEARTHS. The vision is simple: Act to support NGOs, human rights groups, survivor organizations and enterprises in eradicating oppressive forced labor practices.

A glimpse into the technology

The consortium chose the development of Causal Artificial Intelligence Models, ideally suited for solving complex problems such as identifying, resolving, and eradicating forced labor practices (aligned with the CCLA initiative).

In simple terms, causal AI is designed to understand, and illustrate, the cause-and-effect elements behind an outcome (figure 1), the insights of which subsequently support human-led resolution.

Causal AI develops its learning from domain knowledge and specific knowledge, and extracts insights from historical data. In effect, Causal Modeling Capabilities offer comprehensive scientific modeling and analytical tooling with coarse-level services such as causal hypothesis, root cause analysis, and what-if scenario simulations. It is this human-like approach to complex problem-solving that makes this technology approach interesting in this use case.

Before AI can learn, it needs data – specific data

Because of its scientific approach, the data need to be contextualized before being “taught” to Causal AI. This contextualization was made possible through behavioral specialists and data scientists. The team adopted a hypothesis development approach that embedded a multi-stakeholder approach to yield unique and meaningful outcomes which depend on a rigorous methodology and approach (see Figure 2).

The methodology begins with developing hypotheses based on existing research on forced labor and behavioral science more broadly. These initial hypotheses are then further expanded, refined, and corrected for potential biases and blind spots by consulting with survivors of forced labor.

Hypothesis development in iEARTHS is effectively an exercise in identifying an ecosystem of all possible reasons (i.e.: causes) relating to an issue of interest (i.e.: an effect) within a defined setting (i.e.: context). In the final step of the initial hypothesis generation process, the general hypotheses created through the collaboration between behavioral scientists and survivors are formatted to translate the specific training data that are available for input as hypotheses to the Causal AI. Two example hypotheses (out of hundreds created) that were developed for model input as a part of the POC are depicted in figure 3.

To better understand the hypotheses challenge, imagine that you are trying to identify the possible causes (motivators) for a person perpetuating a forced labor act when you lack a contextual understanding or ability to recognize the environmental conditions within which it occurs. It is the quintessential example of “you don’t know what you don’t know.” the modern expression for Socrates’ “I only know that I know nothing.”

Filling this knowledge gap requires subject matter expertise to ensure the AI learns correctly, meaning it will not develop erroneous conclusions (aka hallucinations). The expertise emerges in different forms such as survivor voices, labor or human rights organizations, or organizations specializing in ethical trade and human rights.

Including a human perspective within AI learning is critical to eliminate biases, assumptions that arise from a privileged life, or the application of values and norms that lack universality. In other words, for an AI model to provide trusted responses, the data from which it learns must be consistent, accurate, complete and unbiased.

Bringing data to life through the survivors’ voice provided unique insights into the financial and non-financial behaviors and indicators of modern slavery and human trafficking in a few specific contexts (e.g. sex work in Europe or human trafficking from certain African countries). From these discussions the human behavior specialists were able to generate a behavioral framework contextualizing data inputs for the Causal AI model as reflected in Figure 4. 

Because Causal AI embeds expert-specific knowledge within the modeling process, insights from experts highlight specific and generalized relationships that the technology must respect. In technical terms, for example, this might be expressed as a survivor group expert knowing that economic benefits have a linear positive relationship with forced labor in vulnerable communities. This knowledge is embedded into the model to ensure it always respects this cause-and-effect relationship.

Importantly, the insights that emerge from iEARTHS will have a wide range of critical uses. For example, a strong, evidence-based analysis showing where supply chain forced labor (or child labor) risks lie, specifically what those risks are, and then what steps the organization needs to take to prevent these practices within its value chain.

To better understand these critical ‘triggers’ (or key dependencies), data is systematically structured in a way that understands the complexities of this data. The practice of data architecture that is best suited to address this challenge is referred to as semantic modeling – or knowledge graph.

Insights come from real data!

Real data means that data must be available and be of sufficient quality from which to gain insights once the Causal AI technology processes it against its algorithms. This journey, as basic as it sounds, starts with a robust data management and governance process. Four data-related elements are required (D. Wray, F&G 2023):

1. Data governance & controls – the checks and balances that ensure data lineage, authenticity, and integrity throughout its lifecycle.

2. Data management framework – the mechanism by which a company ingests, processes, secures, and stores its data.

3. Data collection mechanisms – processes through which companies acquire data, either internally or externally generated.

4. Data dissemination – the vehicles through which information is disseminated internally (for management decision purposes) and externally (for user usage).

Data management framework models are useful for establishing an internal data management system that stands up to the rigor of commonly used internal control process frameworks, such as COSO (ICSR or ICFR), or SOX (Sarbanes-Oxley). 

The EDM Council, a not-for-profit association, developed an industry-supported 8-core component data management capability model (DCAM, figure 5) to aid the implementation of a data management strategy for organizations of all sizes and types. To data experts, ESG reporting is a data challenge. ESG data must be high quality and above all, be trusted. To achieve trust in the data, DCAM defines all the capabilities needed to establish and maintain a robust data program.

A collaborative approach to solving highly complex issues

Supporting the vision of developing a meaningful solution (figure 6) that supports all community actors in eradicating modern slavery requires a collaborative approach that includes data, data science, technology, and cross-functional, cross-discipline and cross-organization input.

The iEarths project adopted a structured methodology and iterative development cycle which included human input throughout to validate the assumptions, inputs, and outcomes at each stage of data input, learning, testing, iteration, and outcome validation.

In addition to the human considerations, are the data privacy concerns and the ethical considerations of both the data usage and the technology application. On the data privacy element, using anonymized data from individuals and organizations was foundational for iEARTHS – nothing could be used that may lead to the identification of a specific person, or organization, to protect both rights to privacy and security.

In terms of technology usage, the iEARTHS team continuously ensures safeguards to prevent the technology from being used for social profiling or by perpetrators intending to develop new concealment practices.

Reality sets in

Beyond technology and data, the consortia reflected on how the solution could be further developed to be a decision useful for industry in identifying, fixing, and preventing modern slavery within international supply chains. Exactly what types of business challenges could the solution help solve?

Common challenges cited include the need to:

• Improve an organization’s supplier due diligence verifications (lowering legal and reputational risks), 

• Develop tailored plans to enable a company to improve its business-critical supplier labor practices (lowering legal, regulatory, and reputational risks, and increasing long-term cost benefits)

• Optimize procurement decisions that balance a set of complex legislation, environmental and social considerations, company values, cost, time and quality (optimizing cost-benefit trade-offs). 

The idea being that every decision-maker becomes empowered to make decisions for which they are more fully informed and therefore more accountable.

The ultimate objective of iEARTHS is to both support the public interest by empowering IGOs and NGOs with evidence-based information to support and inform their work, and to support companies in managing their legal, reputational, and business risks.

Stay tuned, this forced labor in international supply chains collaboration is actively underway. You can get updates through iEarths (iearths.org),[3]  EDM Council (edmcouncil.org), BEworks (beworks.com), Parabole (parabole.ai), or DW Group (dwgroup.consulting). A fairer future is being developed right now; one where you too can contribute your data or domain expertise towards public interest causes. 

The choice of what you do next is yours.

About the Authors:

Sandip Bhaumik, Co-founder & CTO, Parabole Inc: Sandip is responsible for the design, development, and rollout of Parabole’s technology and products. Sandip holds an MS in Signal Processing from the Indian Institute of Science, Bangalore, and a BE in Electrical Engineering from Jadavpur University.

John Bottega, President, EDM Council: John Bottega, President of the EDM Council, is a senior strategy and data management executive with more than 40 years of experience in the industry. Over his career, John has served as Chief Data Officer in both the private and public sectors, serving as CDO for Citi and Bank of America, as well as for the Federal Reserve Bank of New York, and head of data management for the US Department of the Treasury’s Office of Financial Research (OFR).

Jennifer Weeks, VP Strategy, BEworks Inc: Jennifer is a Vice President of Strategy at BEworks, where she leads work on the psychology of physical space design as well as technology use, with a focus on artificial intelligence. Jennifer holds a PhD in cognitive neuroscience from the University of Toronto, where she used neuroimaging and machine-learning tools to study how environmental distraction affects cognition across the lifespan.

David Wray, President, DW Group LLC: For over twenty-five years, David has been a finance, digital transformation and change management executive. He is co-founder of iEARTHS.org, a public interest initiative to eradicate modern slavery. He has worked with Fortune 500 companies and non-profit organizations throughout the Americas, Europe, Africa, and Asia.