Humans Can No Longer Manage Data, They Need AI — Head of Data Management and Governance

(US & Canada) Marla Dans, Head of Data Management and Governance at Chief Data Office, speaks with Martin Couturier, Director of AI Innovation at Saige, in a video interview about her journey, the importance of data governance, the three core components of data governance, the need for executive support, making the data governance function successful, adopting AI responsibly, and the top must-haves for responsible AI.

A 10-year veteran in the Chief Data Office, Dans began her career at JPMorgan Chase. Then she pivoted into a FinTech company, TradeWeb, and went on to Chicago Trading Company from there. In addition to that, Dans has also consulted for multiple financial services firms before beginning her career at Morgan Stanley and working in various leadership roles.

When asked to communicate the importance of data governance, she states that with the exponential growth of data and everyone wanting insights, data democratization is on steroids now. In this scenario, humans can no longer manage all the data, and they need AI.

Through data governance, Dans defines rules and best practices that ensure data is reliable, accurate, and available and establishes accountability and ownership for data. To get the most out of data, she outlines three core components of data governance, which include knowing, owning, and honing the data.

Knowing the data envelopes the aspect of data transparency; owning refers to data accountability; and honing is about improving the quality of the data, says Dans. Moreover, data must be wrangled, cleansed, normalized, enriched, and transformed into information usable for business insights.

Adding on, Dans says that data can only be considered to be AI-ready when the critical data sets are governed and certified.

Moving forward, she affirms that executive support is critical for any chief data office or data governance program to succeed. With C-Suite support, everyone gets involved, and then business priorities are aligned to identify quick wins.

Emphasizing quick wins, Dans focuses on short-term business problems, identifying the primary business drivers, and working accordingly. Next, she recommends laying out a three-year vision.

According to Dans, getting too caught up in tooling must be avoided, as tooling alone does not solve business problems. She suggests stressing initiatives that scale.

To make a data governance function successful, it will require transparency and ownership, reiterates Dans. This includes data catalogs and top-down support for data stewards across the organization for end-to-end data quality.

In addition, there are policies for regulatory requirements, data protection, and access control, which require partnering with the information sector. Dans notes that data products are also included in a data governance program, which must be aligned with business objectives.

She also mentions the usage of data products in master data management; whether it is client data, reference data, or external data, it all falls under the data governance umbrella. She suggests reading her publication on data analytics and digital transformation, which covers more on this.

Thereafter, Dans states that to adopt AI responsibly, a robust data management and governance program is a must. AI governance leverages data governance to enable expansion into managing models, model validation, ethics bias, policies, and controls.

However, it is imperative for organizations to first decide on the data that should be utilized to train the models. The top must-haves for responsible AI, in her opinion, are data quality, privacy, and security.

Furthermore, a data quality program has two core pieces, says Dans:

• Identifying and learning data quality issues

• Having a robust remediation process

Emphasizing data privacy and security, she notes that it is crucial to adhere to legal and regulatory compliance and prevent PII misuse. It also includes implementing data protection measures such as encryption, masking authorization, and access control.

Dans reflects that data provenance and lineage have transitioned into a must-have for responsible AI. With the advent of LLMs, it is fundamental to know the data origin, its movement and transformation through the organization, and how the model generates the result to be able to trust it.

From an operational perspective, it is necessary to debug, audit, and proof the AI models. She recalls that even ChatGPT cites source websites in some cases, providing a level of comfort.

Concluding, Dans states that organizations must consider ethical guidelines and potential biases while building policies, as the guidelines will promote the development of fair and unbiased models.

CDO Magazine appreciates Marla Dans for sharing her insights with our global community.