March DLT Monthly Meeting Summary - “Data Prep for AI"
Part 1/3
We began with a conversation about how AI has changed the landscape, including some points shared by the group:
Governance and Regulation - There is a clear need for oversight and governance amid an explosion of data, with a focus on meeting regulatory expectations.
Explosion of Toolsets - There has been an emergence of toolsets as well, and that may require equal evaluation.
Gradual rollout - One member shared how they are gradually increasing use cases across the organization - while focusing on data preparation and governance.
Unstructured Data Challenges - An issue of unstructured data arose, as well as the need for AI systems to understand user context and deliver appropriate data access controls.
Another member commented that “The LLM knowledge for RAG is within a lot of PDFs, SharePoints, PPTs that lack meta data and structure, while also containing a lot of IP/PII to try and contain as well.”
Leveraging What Works - In terms of using external vendors, one member shared that they are using AI to convert knowledge base articles into “smart snippets”. One member shared their strategy of automating unstructured data from scratch.
Vendor Lock-In - Another highlighted a major issue of vendor lock-in that has become pressing.
Moving data from one cloud vendor to another would cost 6 figures today but will blossom to 8 figures in the next couple years. Is there really an option or will the cloud vendor necessarily become a “strategic vendor”?
There are risks of having multiple vendors though too. So members are sorting out their options here, and we may want to have a deeper dive to discuss this issue more.
Part 2/3
We then delved more deeply into risk management and governance, with the following highlights:
Jack Berkowitz, Chief Data Officer from Securiti, shared key principles, including identifying existing models and associated risks and ensuring transparency and traceability in data flows.
He emphasized the importance of cybersecurity principles, including implementing firewalls and compliance with regulations, and the need for clear pipelines in data governance.
He highlighted the need for organizational change enabled by technology as evidenced by inconsistencies in customer service.
Part 3/3
We then broke into breakout groups to discuss governance practices among our own organizations with some takeaways:
New roles - Bringing in new roles like data librarians and architects to help manage.
Product managers - Product mindset maturity can yield good domains and quality. And wrap it with engineering data.
Education - Some work to be done educating the workforce - 1) share what Gen AI is and what not to do with it. And share some guardrails. 2) have sponsors and advocates within departments, and 3) train the trainer
An “AI Institute” - One member shared standing up an AI Institute where all ideas came through and sorted out vendors and tracked progress of what’s working and what’s not.
“Weighting quality of data” - data has accuracy and timeliness scoring that needs to be considered.
Design principles - There is a tension between governance, bureaucracy, and speed - as well as the need for executive buy-in.
Faster executive decision-making - There is a top down need for faster corporate decision-making.
So we closed the meeting realizing there is more work to be done and shared - as we each have our own successes and setbacks in this space.
