Data Science Skills and Generative AI: Best Buy, Equifax, Google
At COLLIDE 2023 in Atlanta, data science leaders from Best Buy, Equifax and Google discussed how generative AI and other emerging technologies would reshape roles and skills. They covered hard and soft skills, production trade-offs, regulatory risk and what hiring managers look for.
By The AI Enterprise Magazine Team4 min read
Key takeaways
- Both skill sets count. Yasko said data scientists need strong programming and cost awareness, plus listening skills at the start of a project.
- Roles are converging with software engineering. Javed advised learning ML infrastructure, MLOps or data engineering.
- Generative AI has limits. Heavy models can be too slow and costly for real-time, high-scale use, and regulated firms worried about code provenance.
- Humans stay in the loop. Best Buy expected early customer-facing uses to be reviewed by employees.
- Hiring is about fit. Yasko screens for intelligence and attitude and builds university relationships early.
At COLLIDE 2023 in Atlanta, three data science practitioners from Best Buy, Equifax and Google discussed how generative AI and other emerging technologies would change the work of data scientists, and what job seekers and hiring managers should focus on. The panel, "Data Science's Next Act," was moderated by Allen Terleto, Field Chief Technology Officer at Redis. The speakers were Warren Hearnes, Vice President, Head of Data Science at Best Buy; Chris Yasko, SVP and Chief Data Scientist at Equifax; and Rafiya Javed, Machine Learning Engineer at Google.
Terleto noted that the panelists spoke for themselves and not for their employers.
Which skills matter most for data scientists: hard or soft?
Yasko said the answer is both. The hard skills, in his view, are less about any one language than about writing algorithms, debugging code and understanding runtime cost. He pointed to the growing field of FinOps and the need to know how much CPU and memory a model uses.
On soft skills, he said storytelling at the end of a project matters, but listening at the start matters more. Data scientists must hear "what they have not said" and probe the stakeholder's real problem, he said, to avoid presenting months of work that misses the point.
Hearnes described the same lesson from early in his career. He had assumed the math would tell the story, then learned that executives needed coaching on what optimization or machine learning was doing for them.
How are data science roles changing?
Javed said titles such as data scientist and machine learning engineer vary by company and region, and advised job seekers to read role descriptions closely. At Google, her team of machine learning software engineers builds and serves models and the data pipelines behind them, supported by research and infrastructure teams.
She said the field had moved toward the software engineering end of the spectrum over the past decade. For people early in their careers, she suggested learning about ML infrastructure, MLOps or data engineering, because showing willingness to cross into those teams makes candidates easier to hire.
Which emerging technologies will matter?
Hearnes named generative AI as the easy answer, then pointed to quantum computing and fully homomorphic encryption as areas further down the road. He gave the most attention to reinforcement learning, noting that reinforcement learning with human feedback helped make pre-trained transformers far more useful.
He also predicted that large language models could let a domain expert describe constraints in plain language, such as the doors, hours and staffing at a distribution center, and have the system write the optimization model.
What are the risks and limits of generative AI in production?
Javed said a rich model is not always worth its cost. In real-time ad and feed ranking, latency and scale requirements can make heavy models impractical, and "you have a hammer but you can't actually throw it at everything," she said. Generating results for billions of search queries would be slow and very expensive, she added.
Yasko, whose industry is regulated, said data leakage into prompts is a known worry, but he was more concerned about the reverse. Code pulled from a generative tool could carry unknown provenance, including open source, proprietary or copyrighted material. He also said regulators "don't know what to do" yet and will need help understanding what is possible.
At the time, Yasko said Equifax saw generative AI fitting creative and productivity tasks such as call center letters, code assistance and marketing, and was apprehensive about putting it into products where it must answer questions on data it was not trained on.
Hearnes said Best Buy's main risks were brand and reputation, and expected early uses to keep a human in the loop, such as triaging customer questions with retrieval-augmented systems. He said trust in algorithms can build over time, comparing it to relying on GPS.
What do hiring managers look for?
Yasko said he looks for "the gift and the choice": intelligence, tested through problem solving and math, and attitude, which he checks through personal reference calls. For mid-career hires he brings candidates to Atlanta for a day of interviews with five to seven people, and he said his groups have low turnover. He also said Equifax works with university labs, including at Georgia Tech and Kennesaw State, to meet students before graduation.
Hearnes suggested balancing exploring new skills with exploiting what you already know, borrowing the idea from reinforcement learning. He added that "not every decision is the final decision."
Where to continue the conversation
Conversations 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 skills do data scientists need?
Chris Yasko said both hard and soft skills: programming, debugging and runtime cost awareness, plus listening to stakeholders to understand the real problem.
Is data scientist the same as machine learning engineer?
Rafiya Javed said titles vary by company and region, and that the field has moved toward software engineering, so candidates should read role descriptions closely.
What are the risks of generative AI for regulated companies?
Yasko cited code of unknown origin entering products and regulators who were still working out how to respond.
What do hiring managers look for in data scientists?
Yasko described testing intelligence and checking attitude through references and extended interviews with many team members.
This article is based on remarks by Warren Hearnes, Vice President, Head of Data Science, Best Buy; Chris Yasko, SVP and Chief Data Scientist, Equifax; Rafiya Javed, Machine Learning Engineer, Google; moderator Allen Terleto, Field CTO, Redis at COLLIDE Data + AI Conference 2023 on October 3, 2023. More coverage: COLLIDE AI Conference.
Filed under: People, Culture & Leadership, AI Use Cases & Applications, AI Governance & Trust
