(US & Canada) | Have a Data Quality Metric While Implementing AI Initiatives — Mitsubishi Electric Trane US Director of Data Governance

Sarang Bapat, Director of Data Governance, Mitsubishi Electric Trane US, speaks with Sue Pittacora, Chief Strategy Officer of Wavicle Data Solutions, in a video interview, about compliance in financial services, having a solid data governance framework, the collaborative roles of AI and data governance, and the importance of data quality in AI initiatives.

Mitsubishi Electric Trane HVAC US (METUS) is a leading provider of ductless and VRF systems in the U.S. and Latin America.

Shedding light on the aspect of compliance in financial services, Bapat discusses anti-money laundering compliance. He adds that regulators are interested in understanding how an organization assesses its risks and what drives the risk appetite.

Elaborating, Bapat states that in the anti-money laundering domain, factors like customer geography (domestic vs. international) matter. It is considered risky if the business is worldwide.

Next, the types of products and services offered (e.g., wire transfers, stored value cards, money orders) aid money laundering. With a larger organizational portfolio, one has to examine the risk from the type of customer, whether it is personal or commercial, or whether there are shell companies involved.

Further, Bapat continues that regulators also inquire about the data that drives the risk assessment engine. From where the data comes from to where it goes, who owns it, and the quality of the data.

To address the concerns, he suggests having solid data stewardship in place, which would help to demonstrate to regulators who is responsible for what.

In addition, regulators will want to understand the organizational controls in effect, says Bapat. Having a well-designed data governance framework can make the processes manageable, he notes.

While there might be some tweaking necessary in the existing framework, having a framework comes in handy to ensure compliance, says Bapat.

When asked about the collaborative roles of AI and data governance in METUS, Bapat shares that the organization is at the beginning of its journey. People are still confused about how AI can help or what the use cases could be.

Bapat then explains the situation by using a mutual fund analogy. He says that when it comes to investing in mutual funds, many people will say that they do not invest, but they have a 401(k) plan. This means people are investing in mutual funds without knowing.

Similarly, Bapat continues, data governance tools today already have AI components incorporated, which learn from the data and make observations. For example, a tool might notice different pin code formats or a field with numerous duplicates after multiple runs. This indicates that the machine has learned about the data well enough to present its observations, he adds.

These observations add value for data stewards because, with the massive growth in data, it becomes challenging for humans. He urges data stewards to consider both the rules they create and the insight AI provides.

Additionally, Bapat states that while AI learns and makes connections, it is not always accurate or actionable. Nevertheless, it is worth exploring and considering.

Emphasizing the importance of data quality, he affirms that the quality of the underlying data is crucial when it comes to LLMs and AI-powered applications. Taking the instance of the retrieval-augmented generation (RAG) algorithm, he states that if the retrieval data is not accurate, the inaccuracies will compound. Bapat suggests AI leaders and vendors identify the data elements and have a data quality metric while implementing AI initiatives.

Moving forward, he maintains that the quality of the data used must be continuously monitored, especially when incorporating third-party data sources. Bapat opines that while working with the legal team, the data glossary must be up-to-date and the data must be filled in the right fields. It is critical because digital products and intelligent algorithms leveraging AI are built on top of this ecosystem.

Thereafter, Bapat mentions that organizations need to delve into the details and identify the crucial data elements that the RAG algorithm will utilize. For example, during data augmentation, if relevant information is deactivated and the algorithm fails to detect that because the flag was not set, it becomes a data quality issue.

Circling back, Bapat reiterates that algorithms rely on the assumption that data is accurate; thus, assessing data quality in AI initiatives is critical.

CDO Magazine appreciates Sarang Bapat for sharing his insights with our global community.