Getting ready:

Assembling your AI roundtable

Getting ready:
Assembling your AI roundtable

Don’t tackle AI governance alone. Before initiating your AI journey, it’s essential to assemble the right mix of stakeholders for an “AI roundtable.” This is a pivotal step towards success due to the myriad opportunities and challenges AI presents. A diverse range of perspectives is crucial for making balanced decisions. While the composition of your AI roundtable will vary based on your organization’s unique structure, certain key roles are indispensable:

Chief AI Officer and Data Science teams: Their inclusion is essential for leveraging overall AI strategy, model development, and data analysis expertise as well as ensuring AI’s technical viability and alignment with business objectives

Chief Data Officer/Data Office: Their participation is vital for managing all data-related aspects, particularly ensuring the accessibility of trusted data

Legal, Compliance, and Privacy teams: Their involvement is crucial for navigating legal risks, understanding regulations related to AI and data, and ensuring compliant usage

Business Unit Leaders: Inclusion of these leaders is important, particularly if AI implementation could significantly impact their teams

Candidates for the AI roundtable
What are the desired characteristics of business and technical stakeholders?

Entrepreneurial or trailblazer spirit
Well-respected cross functionally
Subject matter expertise
Proactive
Responsive
Uses time efficiently

Step 1: Define the
use case

How do you know what to do — and when? Defining your AI use cases Crafting a use case is an essential step in ensuring the effectiveness and alignment of your AI model with organizational needs. To achieve this, consider framing your approach in the form of key questions that guide your planning and implementation process.

Business
How does the AI model align with the organization’s strategic goals and objectives?Who are the key business stakeholders impacted by the AI model, and what are their insights?What does success look like for this use case, and how will it drive business value?Legal, ethics and compliance
What are the legal constraints or considerations the AI model must adhere to?How does the AI model comply with the ethical standards of the organization and industry?What compliance requirements are relevant to the data and functionalities of the AI model?Data
What specific data is required, including sources, types, and formats?How will data be managed, accessed, and protected throughout the AI model’s lifecycle?How will you ensure data quality and integrity so it’s suitable for the AI model?

Make sure your AI use cases
map to your business goals

Make note of all the scenarios that apply from the following lists. Prioritize them based on your unique company needs.

Brainstorm business initiatives where an AI project will have a positive impact

Sample initiatives and outcomes:

Reduce operating expenses

Increase customer retention

Diversify and grow revenue streams

Improve cross-functional productivity

Digital transformation

Data monetization

Identify and prioritize use cases that align with strategic initiatives

Examples:

Customer experience enhancement

Operational efficiency

Sales and marketing optimization

Risk management

Fraud detection

Identify any potential risks an AI project could introduct to a strategic initiative

Sample scenarios:

Compatibility with current systems: Will the AI solution integrate seamlessly with existing architecture or are there potential incompatibilities?

Skills and expertise: Is there a risk of lacking or losing essential skills and subject matter expertise needed for the AI project?

Data management and integrity: Could there be risks related to data loss or corruption during the AI project implementation?

Financial management: Are there potential budget overruns associated with the AI project and how can they be mitigated?

Data security: What are the risks to data security when implementing AI and how will sensitive information be protected?

Data governance and compliance: How will the Ai project adhere to existing data governance frameworks and compliance regulations? Are there any new compliance challenges introduced by AI?

Step 2: Identify and understand data

AI starts with trusted data. It's why the old adage—“garbage in, garbage out”—is especially true when it comes to AI. It explains why once you’ve defined your AI use cases, you need to take a close look at your data.

Can you identify all the data
sources you need for the use case?
Where does the data reside?Is the data compatible with your AI platform?Are there existing metadata connectors
between your data governance platform
and the source and destination platforms?Is sensitive or critical business data stores
at the source?Are the data stakeholders and the benefits to each stakeholder known for the use case?
Who are the data owners?Who is needed to validate and certify the data?Who benefits from the use case and how?Can the use case benefits be measured?
What are the hard benefits that can be measured?Are there soft benefits?How will you measure and report on them?Are the data policies needed for the
data assets known or easily defined?
Do the existing policies need to be refined?Are there regulations or industry best practices that need to be assessed?

Is there sensitive data involved?Are there data residency restrictions that need to be taken into account?Who should be allowed access to the data?What are the data retention policies?

Learn how to become an AI governance leader.

Watch our webinar for best practices

Step 3: Document models and results

As you progress to Step 3 of the AI Governance framework, the focus turns to meticulously documenting the development and outcomes of your AI model. This phase is critical for ensuring transparency, compliance and effectiveness of the AI implementation.

Model development documentation
How are you documenting the AI model’s development process, including algorithm choices, parameter settings, and version control?What methods are being used to track and record the performance and adjustments of the model over time?Data product and usage tracking
How is the associated data product being documented and tracked?What systems are in place to ensure accurate and comprehensive usage logs of the AI model?Data lineage and transparency
Can you clearly trace the origin, transformations, and applications of the data used in the AI model?How is data lineage being maintained and documented, especially in industries with stringent regulatory requirements?Model analysis and reporting
What processes are in place for continuous analysis and reporting on the AI model’s performance?How are challenges, anomalies, or biases in the model being documented and addressed?Preparation for production
What criteria are being used to determine when the AI model is ready to move into production?How are the initial results from the model being evaluated and validated against the defined use case and objectives?

Step 4: Verify and monitor

The final set isn’t really a final step. Step 4 of the AI Governance framework emphasizes the ongoing nature of AI governance. Once your model transitions into production, continuous monitoring and verification are crucial to ensure its effectiveness, compliance, and adaptability to evolving regulations.

Model performance verification
How doy ou verify that the Ai model performs as intended before its full-scale deployment?What measures are in place to ensure the. model meets both technical specifications and business objectives?Transition to production environment
What is your process for integrating the AI model into the operational environment?How do you document and trace the data flow within the AI system to understand its transformation and role in decision-making?Ongoing data quality and compliance monitoring
What mechanisms are in place for monitoring the model's performance, data drifts or unexpected behaviors?How do you ensure ongoing compliance with regulatory and ethical standards, especially when the model interacts with new data sets?Periodic Model retraining
What triggers the retraining of the AI model and how frequently is this done?How do you integrate new data, regulations and technological advancements into the retraining process?