Co-design and stakeholder mapping are interconnected processes that support collaborative decision-making, stakeholder engagement, and the development of solutions that meet the diverse needs and expectations of stakeholders. Co-design is a participatory design approach that involves collaborating with those most proximal to the problem. It promotes participatory decision-making, mutual learning, and shared ownership, resulting in solutions that are more relevant, effective, and sustainable.  Stakeholder mapping provides a foundation for effective co-design by helping project teams identify the relevant stakeholders to involve in the process.² By understanding the interests, concerns, and potential impacts of different stakeholders, project teams can tailor development to be inclusive of the needs of specific stakeholder groups.

‍

Within the context of AI development, stakeholder mapping typically involves the following steps: ³

1. Identification of Stakeholders: This step involves listing all parties that might be affected by or have influence over the AI system. Stakeholders can include owners, input providers, end-users, developers/builders, investors, regulatory bodies, impacted communities, advocacy groups, and others.
2. Analysis of Stakeholder Interests and Influence: After identifying stakeholders, the next step is to understand their concerns, interests, and the level of influence they have over the project. For instance, regulatory bodies might have high influence and specific concerns about privacy and fairness.
3. Assessment of Impacts on Stakeholders: This step involves analyzing how the AI system will affect different stakeholders. You should consider how the technology may benefit or pose challenges to various stakeholders. For example, it might improve efficiency for end-users but could raise ethical concerns for advocacy groups. 
4. Acknowledge Positionality: Acknowledging positionality means understanding the influence of one's own social, cultural, economic, and political background on interactions and impacts. While similar to assessing impacts, this specifically considers the broader context in which the stakeholder operates and how they may engage with the solution. Stakeholder mapping for AI solutions involves evaluating how different groups are affected by the technology, identifying power dynamics, anticipating disproportionate impacts on specific subpopulations, and recognizing systemic inequities. This awareness promotes inclusive decision-making and equitable outcomes by considering diverse perspectives and concerns.
5. Prioritization of Stakeholders: Not all stakeholders will have the same importance or influence on the AI project. This step involves prioritizing stakeholder needs and concerns based on factors like their level of impact and influence.
6. Engagement Strategy: Based on the stakeholder mapping, the next step consists of developing an engagement strategy to address the concerns and needs of various stakeholders. This approach might involve regular communication, involvement in decision-making processes, or addressing specific ethical concerns they might have.
7. Ongoing Reassessment: Stakeholder mapping is not a one-time activity. As the AI project progresses, stakeholders' needs and the landscape might change, necessitating regular reassessment and adjustment of strategies.

‍

Within your stakeholder engagement plan, the organization should identify relevant stakeholders by their roles, including:⁴

• End-Users: Their primary interest is in the AI system's functionality, usability, and benefits. They are concerned about how AI will affect their tasks or lives and the lives of those they influence.
• Input or Data Providers: Their primary interest is in the system equitably and accurately improving their lives. Note that, especially in higher education, the end users may not be the most affected parties; for example, advisors and professors may use predictive analytics that affect students, even though the students are not using the tools.
• Developers and Technical Teams: These stakeholders are focused on the technical feasibility, innovation, and implementation of AI. They are interested in resources, tools, and support needed for development and creating innovative solutions using AI.
• Business Leaders and Managers: Concerned with the strategic alignment, return on investment and market competitiveness, these stakeholders focus on how AI can drive business growth and efficiency.
• Ethical and Legal Advisors: Their interest lies in ensuring the AI system adheres to ethical standards and legal requirements. They are concerned with issues like bias, privacy, and regulatory compliance.
• Investors and Funding Bodies: They are Interested in the AI project's financial viability and potential returns. They seek assurance that the project is financially sound and sustainable.
• Regulatory Bodies: Their interest is ensuring the AI system complies with laws and industry standards. They are concerned with public safety, privacy, and ethical use of AI.

Involving stakeholders from each group supports the successful development and deployment of AI systems that reflect the distinct set of interests and concerns. 

‍

Once identified and their needs analyzed, organizations can employ various methods to prioritize stakeholders as part of their engagement plan. No matter the method chosen, the key is to assess how significantly the AI system affects each stakeholder and how much power they have to influence the project's outcome. Your engagement strategy should focus on stakeholders who have high levels of both impact and influence, ensuring that the most critical voices and concerns are addressed in the development process, especially for those who are highly impacted but may have limited power. 

‍

Reference this resource we created, Stakeholder Mapping Process Guide, to guide your discussion at this phase.

‍

For more guidance and support with stakeholder mapping, explore this helpful resource: Stakeholder Mapping Templates and Worksheets.

‍

Defining your target audience and crafting detailed user personas form integral components of stakeholder mapping, setting the stage for further user research. Identifying your target audience involves thorough market research and analysis to pinpoint those who will most benefit from your product or service. During this process, you will be engaged in activities such as data collection on potential user needs, demographic analysis, psychographic examination for behavioral insights, interaction evaluation with similar offerings, and the identification of specific challenges and needs your product seeks to meet. This meticulous approach ensures your product is tailored effectively to meet the needs of your intended users. 

‍

For more guidance and support with defining your target audiences, explore this helpful resource: Examples on Defining Target Audience.

‍

Following the initial identification of your target user, you will create user personas, which are fictional representations of your audience derived from your market analysis. The personas help in understanding the motivations, goals, and pain points of your potential customers. User persona creation process involves:⁵

• Categorize Your Users: Group your audience into unique segments by identifying shared characteristics, behaviors, or objectives, then disaggregate data within these segments to identify each group's  prevalent traits, specific needs, preferences, and challenges. This analysis helps in forming distinct personas.
• Draft Persona Details: Construct comprehensive profiles for each identified persona. They should include specifics like their name, age, profession, aspirations, obstacles, pastimes, and a short story describing their usual engagement with your product or service.
• Rank Personas: When dealing with several personas, organize them by their relevance and impact on your business. Determine which personas are key to your product's success and focus your efforts accordingly.
• Confirm with Real Data: Cross-check your persona outlines against actual user data to ensure accuracy. The personas you've developed should accurately reflect your real-world audience.

‍

For more guidance and support with user personas, explore these helpful resources:

Guiding Questions to Understand Your Users

Templates for Defining User Personas

‍

Once you have identified stakeholders and their motivations and mapped levels of influence and impact, the next step is to create a plan of engagement to ensure diverse perspectives are considered in the development of the system. An efficient stakeholder engagement strategy includes mapping out what kind of feedback to request from whom and when to engage with the various stakeholders. Effective engagement helps understand and address potential ethical, legal, and social implications, fosters transparency and trust, and aligns the AI project with broader societal values and norms. This inclusive approach is key to mitigating risks and maximizing the benefits of AI technologies.

‍

For more guidance and support with stakeholder engagement, explore this helpful resource:

‍Stakeholder Engagement Throughout The Development Lifecycle

‍

‍

After identifying stakeholders and devising a plan for their effective engagement, the team must now develop a strategy for selecting and preparing data that precisely mirrors the use cases, objectives, and the wide array of stakeholders involved. Representative training data is the cornerstone of developing effective and fair AI models. 

‍

There are two main aspects to consider regarding data representation: technical considerations and data equity. 

‍

The technical process of converting raw data into a format that the algorithm can work with effectively can affect data representation. The technical considerations include feature selection and transformation, where the intrinsic characteristics of the data are identified and translated into a form that highlights the essential patterns and relationships that affect the performance of ML models and determine how well the model can learn from and make predictions or decisions based on the data.

‍

For more information on representation in Data, visit the Data Representation Scheme section of this guide.

‍

Beyond the technical aspects of data representation lies a big challenge: the diversity and equity of data sources, which ensure equitable representation of the target population to mitigate bias. Data inequality leads to moral risk, which includes biases related to demographic groups. For example, due to data underrepresentation or misrepresentation, facial-recognition algorithms have a difficult time identifying people of color; skin-lesion-classification systems appear to have unequal accuracy across races; recidivism-prediction instruments give Blacks and Hispanics falsely high ratings and credit-scoring systems give them unjustly low ones.⁶ Data equity ensures that the developed system is technically proficient, socially responsible, and aligned with the diverse needs and expectations of its users and impacted parties. One of the keys to equitable data is a data set representative of the population. 

‍

By implementing justice, equity, and inclusivity principles and practices to guide anyone who works with data (especially data related to people) through every step of a data project, data equity can ameliorate biased outputs from ML systems.⁷ Firstly, data equity ensures that the AI model can accurately understand and respond to the diverse scenarios it will encounter in real-world applications. This practice helps reduce biases, which can occur if the data does not reflect the diversity in the target population. Secondly, representative data contributes to AI systems' fairness and ethical integrity, making them more equitable in their applications. Finally, it enhances the reliability and performance of AI models, as they are trained on data that truly represents the varied conditions and demographics they will serve. Without equitable data representation, AI systems risk being biased, ineffective, or even harmful in practical applications.

‍

A machine learning model should have representative data that meets the conditions of data equity to reduce bias. Suppose we have an ML model that predicts student retention rate in a college or university, and the resulting model is passed through a bias identification tool like Aequitas, which identifies that the variables for student race and gender are the main factors causing the model to be biased. Such a case is a classic example of data underrepresentation. Developers can rectify this problem by generating synthetic data to increase the representation of racial or gender subgroups underrepresented in the data, retraining the model, and then rerunning it through the bias identification tool again to see if those bias-causing factors have been eliminated.  Remember, the quality of insights derived from data is directly proportional to the quality of data inputs, and biased training data can lead to biased models.

‍

‍For more guidance and support with identifying bias in data, explore this helpful resource: Aequitas Bias and Fairness Audit Toolkit.

‍

‍