Artificial intelligence has moved beyond the novelty phase. Most organizations now have employees experimenting with ChatGPT, copilots, agents and automation tools, but while AI is certainly useful, the challenge is making sure isolated experiments are transformed into true business advantages.
That was the central theme of The Lead Summit panel discussion “Move from AI Experimentation to Mature AI Operating Models and Consumer Experiences.” There, Mimi Ruiz, VP, Ecommerce, Columbia Sportswear; David Cost, CDO, Rainbow Apparel; and Serge Moreau, VP, Technology, Tommy John, shared their experiences integrating AI into their businesses in a deliberate, measurable and responsible way.
Cost compared AI use today to how spreadsheets gained adoption some 30, 40 years ago. Then, “spreadsheet jockeys” learned how to use the innovation, gained a productivity advantage then built increasingly sophisticated models and workflows over time. But when organizations tried to scale those capabilities across departments, they realized the application just wasn’t built for enterprise level.
AI is following a similar trajectory. While employees today are building valuable prompts and automations with “$20 ChatGPT,” companies still struggle to operationalize them securely and consistently across teams.
AI Initiatives Must Begin with Process Clarity
A recurring frustration is that many organizations use AI for AI’s sake, or want to automate processes they do not fully understand.
Before introducing automation, organizations need to understand how work actually gets done, who is involved, what decisions are being made, and where bottlenecks exist. Once that is sorted out, AI can accelerate a process, but it cannot compensate for bad process design.
Next, companies must look within to make sure there are enough skilled employees to help departments with AI innovations. Not all will need a forward-deployed engineer to act as a liaison between tech creation and tech utilization, but “you need an advocate per department to provide business objectives that might not be in the purview,” said Moreau.
AI should also not be mistaken for a business strategy on its own; it is simply a tool to achieve strategic goals. For established brands, strategies remain the same as before AI: to acquire and serve customers, grow market share, increase efficiency and create value.
“One in five consumers will choose a brand just because AI told them that they should choose it. And for those under the age of 44, 7 out of 10 will use LLM as part of their shopping journey. So if I’m to shepherd a new generation and continue to build Columbia’s trajectory, I have to show up how consumers are shifting from the shopping experience,” said Ruiz. “But [AI is] an accelerator enabler, not a new strategy.”
Ruiz described how Columbia builds its organizational structure around three pillars for new AI adoption—a collective guiding principle, a cross-functional governance committee that meets monthly, and a formal approval process for AI use cases. “It can’t just be a technology strategy; it has to permeate throughout the whole organization.”
Rainbow operates entrepreneurially with more flexibility and less bureaucracy than Columbia, but it still has tight AI justification methods, said Cost. “We can’t just do something because we’re going to get more work out of the same money. We need to show what’s the impact on the business.”
For mature organizations using AI, governance is becoming just as important as innovation. AI introduces risks and organizations must consider customer privacy, proprietary information, compliance requirements, legal exposure and data security.
“Business leaders know the more they talk about AI, the higher their stock price goes,” said Cost. “But they may not appreciate that the more they use the word publicly, they’re creating a liability where the SEC can prosecute you for propping up the stock if you’re not using AI in a material way. We’re still very early days in this process, but if you don’t have proper oversight, you’ve created a higher liability you might not understand.”
AI Agents vs Human Ones
Companies should also remain skeptical of vendor promises. AI performs best when it enhances human capabilities rather than attempting to replace them entirely, and companies need to ascertain if their goal is to reduce head count, or to augment existing ability with existing staff.
Cost shared an example where a vendor claimed that its AI solution could handle approximately 40% of customer service interactions without human intervention. But once in place, customer service hours remained largely unchanged, as did case volumes and staffing needs.
“AI thought they had ‘closed out the call,’ but the customer would call back and ask for a human,” said Cost. “It was clear the automation had no impact on the human workload whatsoever.”
Moreau also stressed the importance of having humans to keep AI in check, especially when dealing with agents ordering items for customers. “You always need a human in the loop. If an agent orders something through ChatGPT on behalf of a customer, where is the onus? A lot of ambiguity and fraud could happen where the customer says they didn’t order it.”
At the end of the day, AI should enhance the existing team, not replace it.
“[A company] usually winds up trying to do more with less people,” Moreau added. “As an example, I built a QA framework [with AI] with a team who went from taking a full day to do a QA process to scaling it down to eight hours, and then to two hours. That was life changing as far as the development mindset.”
