At COLLIDE 2025, leaders from The Coca-Cola Company, UnitedHealth Group, The Clorox Company and Randstad Enterprise discussed the people side of AI. They covered building trust through transparency, measuring AI by business value instead of headcount, role-specific training and who should own AI fluency.
Key takeaways
- Trust follows a track record. Separating feasibility R&D from production protects credibility before a long-term commitment.
- Transparency beats black boxes. Explaining why an AI output is what it is turns stakeholders into co-owners.
- Measure value, not headcount. UnitedHealth Group tracks features delivered and speed to value, which eases job fears.
- Make training contextual. Role-specific paths, hands-on workshops and prompt competitions drove more adoption than generic courses.
- Create space to learn. Managers at every level must give people time to upskill alongside their day jobs.
Enterprises racing to put AI into production are finding that the hardest work is human, according to a panel at COLLIDE Data + AI Conference 2025 in Atlanta. "Technology gets the spotlight, but people determine whether AI initiatives succeed," said moderator Felipe Archila, lead for digital workplace analytics at The Coca-Cola Company.
He was joined by Katie McDonald of UnitedHealth Group's chief data office; Nathan Gadir, a vice president of business data and architecture with more than 15 years in healthcare; Arpit Gularti, associate director of analytics in marketing analytics at The Clorox Company; and Sean Sanders, head of data science at Randstad Enterprise. They discussed trust, job fears, upskilling and who should own AI fluency.
How do you build trust in AI with skeptical stakeholders?
Trust follows a track record, Sanders said, comparing it to hiring. "The biggest predictor of you getting that job is actually having that job," he said. Teams that have delivered before get a longer leash.
Until then, Sanders separates research from production. His team treats early work as an R&D project with two to three months to see whether data, explainability, security and scalability barriers can be overcome. Skipping that feasibility step and going straight to productizing sets a project up for failure, he said, and hitting a roadblock mid-project makes the data team look less competent.
Gularti said business stakeholders often see AI systems as black boxes, so transparency is essential. Clorox now generates thousands of marketing assets with generative AI, all of which need legal and compliance review. His team built a system that predicts whether an asset is compliant and explains why not, so brand marketers know which claims to change. Involving stakeholders at each step makes them feel like co-owners, he said.
Archila called this the IKEA effect. People who help build something are more likely to embrace it, so his team leans on a champions network of forward-thinking leaders in each function.
How should companies address fears that AI will replace jobs?
Gadir said organizations need a clear position that AI is meant to augment employees and make repetitive work easier, not replace their jobs, and they need to communicate it.
McDonald described an internal town hall where a comment that 90% of code should be AI-generated alarmed engineers. Her organization measures AI differently, she said: "We're not measuring the benefit of AI in headcount." Instead it tracks how many more features teams deliver and how quickly they deliver value. "It's not headcounts, it's business value," she said, and that framing has helped people trust the technology.
In healthcare, McDonald added, teams document risks before a proof of concept begins. The skeptics who ask how things could go wrong are valuable at that stage. "This is where the negative Nellies are so valuable," she said.
What does meaningful AI upskilling look like?
"Meaningful upskilling really comes from when it's contextual and not generic," Gularti said. Clorox runs quarterly AI workshops. A recent three-hour session included an internal prompt competition built around a framework covering a prompt's goal, context, sources and expectations, with mixed teams of AI power users and staff from marketing, finance and supply chain. Afterward, he said, people who had Copilot licenses but weren't using them began building agents and bringing forward use cases.
UnitedHealth Group's AI dojo program offers separate paths for executives, leaders, product managers and engineers, McDonald said, with the goal of putting everyone in the organization through it by year end. Engineers must complete and submit a graded project. Feedback that projects weren't relevant to people's roles is now reshaping the program. Executives take a 90-minute hands-on workshop in person, while mid-level leaders use online courses.
Sanders pointed to Bloom's taxonomy as a guide. Not everyone needs to create or apply AI, he said; some roles only need to understand or remember. At Coca-Cola, Archila said, marketing produces one- to two-minute videos called AI skill pills that show a prompt and challenge employees to try it.
Who should own AI fluency in the organization?
No one on the panel claimed a settled answer. Sanders said legal, sales and operations all need AI skills, but individual teams may lack the data expertise to start. He described a Gartner framework called a data franchise, in which a central team houses expertise and supports other teams based on their maturity, though his organization hasn't adopted it.
Gadir said AI fluency should extend across the whole organization, while deep AI expertise can sit in specific groups. Archila agreed, with a central function overseeing strategy and governance.
What is the most overlooked people factor in AI transformation?
McDonald said support at every level, including managers who give their teams time to train while still doing their day jobs. Sanders noted that transformation conflicts with a CEO's need to hit KPIs and protect the existing business. Gadir emphasized valuing each other's perspectives. Gularti said "AI is really here to amplify our expertise and not erase it." Archila urged leaders to sit with the people doing the work and listen before leading with technology.
See sessions like this at COLLIDE 2027
COLLIDE, the largest AI conference in the Southern U.S., brings more than 1,000 enterprise data and AI leaders to Atlanta for keynotes and panels on the business and human sides of AI. The 2027 conference takes place Wednesday, October 13, 2027, at the Sandy Springs Performing Arts Center. Registration is coming soon at COLLIDE.
Frequently asked questions
How do you build trust in AI within an organization?
Panelists recommended a track record of delivery, a separate feasibility phase before production, transparency about how AI outputs are produced and involving stakeholders as co-owners.
How should companies respond to fears that AI will take jobs?
Communicate that AI is meant to augment people and measure its benefit in business value, such as features delivered, rather than headcount reduction.
What makes AI upskilling effective?
Training that is contextual to people's roles, with different paths for executives, leaders and engineers and hands-on projects or workshops, according to the panel.
Should AI fluency be centralized or spread across functions?
Panelists said AI fluency should extend across the organization, while deeper AI expertise can sit in a central team that supports functions based on their maturity.
Watch: Powering the People Side of AI: Culture, Change Management, and Upskilling for the AI Enterprise on YouTube
Event: COLLIDE AI Conference · More coverage: COLLIDE AI Conference insights
