Enterprise Architecture Lessons From 25 Years in Data: Vida Williams
At COLLIDE 2023 in Atlanta, Vida Williams of the Virginia Alcoholic Beverage Control Authority looked back on 25 years in data and enterprise architecture. She drew lessons on data modeling, executive attrition, data-driven culture and the slow ROI of data, then cautioned against tech for tech's sake in generative AI.
By The AI Enterprise Magazine Team4 min read
Key takeaways
- Model the data first. Teams that understand data structure and meaning up front outperform teams that rush to the end.
- Start with the business problem. The dot-com bust ended tech for tech's sake, Williams said.
- Data health signals infrastructure health. Check data quality and gaps to gauge how sound the technology is.
- Executive attrition stalls innovation. Champions who leave leave problems unsolved.
- Data ROI has a long tail. Data makes the business clearer and more efficient but is rarely a profit center by itself.
At COLLIDE 2023 in Atlanta, Vida Williams, Chief Digital and Branding Officer at the Virginia Alcoholic Beverage Control Authority, used a keynote-style talk to look back on 25 years of data and technology work. Her session, "Considerations in Enterprise Architecture: An Executive Retrospective," traced a career from technical writer to engineer, enterprise architect and executive, and ended with a warning about the direction of generative AI at the time.
Williams said she was taking a step back because "if we don't really think about where we've come from" the industry is likely to repeat its mistakes.
What did early big data projects teach about data modeling?
Williams said her first big data project, a transformation of EPA data from a mainframe into relational models, would not be done that way today. She said that with what the industry now knows, the data should never have been forced into a fourth normal form relational structure.
Still, she credited the discipline. Fourth normal form, she said, was a methodology to make data accessible, and today's options include metadata, reference data and graph databases. In her experience leading large teams, the engineers who stop and sit down with the data "with the rigor that we had" succeed more often than teams racing to the end.
That up-front understanding of structure, intent and meaning is a modeling art she did not see the industry committing to anymore. Without it, she said, teams discover too late that they lack the data they need to make decisions.
What did open source and the dot-com bust change?
Williams described working on open-source projects, including Drupal, with contributors from around the world. She said the experience shaped how she approaches technology as an executive, and she praised the pace of open-source collaboration compared with large companies, where innovation can stall when return on investment is unclear.
The dot-com bust, she said, ended the era of "tech for tech's sake." Her first question to startup founders who pitch her is about the business model, not the app: "tell me what the business problem is" that technology makes more efficient. The bust also marked the point where the industry split between infrastructure and data. She compared it to a plumber versus someone who handles water quality, and said she chose water quality.
What should executives take from this?
Williams urged technologists to pursue executive and even legislative leadership roles. If practitioners do not refine the language around their work, she said, executives and regulators will lack the understanding to oversee it or to keep customers safe.
She named executive attrition as the number one killer of innovation, because champions leave for the next opportunity before problems are solved. She asked executives to consider how long they are willing to commit to solving a problem for one organization.
She also said the health of an organization's data indicates the health of its technology infrastructure. Companies that moved to new platforms and then to the cloud did not always bring their data along, she said. And the success of a strategic plan depends on checking people, processes and technology together.
What does a data-driven culture really require?
Williams said she has never met an enterprise that did not want a data-driven culture, and asked what it means in practice. Using an example a fellow attendee shared of automated speech-to-text in customer service, she described how automation shifts roles from data entry to analysis and then toward further automation. That is what communities fear, she said, so leaders must explain that new roles will be needed.
She said the designers who will build human experiences that make customers feel heard have not yet been trained.
On return on investment, she said data has "the longest tail" of ROI. Data refinement makes a business better and gives leaders clarity, she said, but it is rarely the payoff by itself. She also noted that warehouse, mesh, fabric and lake projects rarely deliver what they were sold, joking that she had met "a swamp project."
Why was she concerned about generative AI?
Williams closed with what she called her hot-button topic. She said she saw tech for tech's sake returning in generative AI and creating public panic. She asked the audience to agree that language models then in use were too restrictive, reflecting a narrow range of how people speak, and that the industry had not settled what it wants the future of AI to be.
If the industry repeats the bubble approach, she warned, the cost could be worse than lost money, because people and cultures could be left behind. She framed it as an ethical question, an evolution of machine learning bias, and a corporate pipeline issue, since restricting who can engineer these systems would worsen labor shortages. Her ask was for a more rigorous engineering conversation and a roadmap, rather than "waiting for the bubble to burst."
Where to continue the conversation
Discussions like this continue at COLLIDE 2027, the largest AI conference in the Southern U.S. with more than 1,000 enterprise data and AI leaders, on Wednesday, October 13, 2027, at the Sandy Springs Performing Arts Center in Atlanta. See /events/collide/. Registration is coming soon. Free live sessions are also on the 2027 webinar schedule.
Frequently asked questions
What did Vida Williams say about data modeling?
She said rigorous up-front modeling, like the discipline of fourth normal form, helps teams understand data structure and meaning and succeed more often than teams that rush.
Why is data ROI slow to appear?
Williams said data has the longest tail of return because refining it improves efficiency and clarity, but it is rarely a product on its own.
What is the biggest killer of innovation, according to Williams?
Executive attrition, because champions leave before the problem is solved.
What concerns did she raise about generative AI in 2023?
She warned of tech for tech's sake, restrictive language models and people being left behind, and called for a more rigorous roadmap.
This article is based on remarks by Vida Williams, Chief Digital and Branding Officer, Virginia Alcoholic Beverage Control Authority at COLLIDE Data + AI Conference 2023 on October 3, 2023. More coverage: COLLIDE AI Conference.
Filed under: Data & AI Infrastructure, People, Culture & Leadership, AI Governance & Trust
