Your executives are asking for AI-powered customer experiences, but your data teams are still playing catch-up with basic self-service requests—and that gap is costing you competitive advantage every day it persists.
While most companies are stuck debating whether their data is "ready enough" for AI, leaders at Sephora, Hyatt, WEX, and Macquarie Bank figured out how to flip the script: they built intelligence-first cultures where business users drive personalization, fraud detection, and customer insights without waiting in line for analyst support.
Here's exactly how they did it, and why their approach is producing measurable results that matter to the C-suite.
1. Build a Semantic Layer Foundation for AI-Powered Customer Experiences
Struggling with Inconsistent Data or Untrusted AI Experiences? Steal Sephora's Model
If your customers don’t trust your AI—or your analysts don’t trust the data powering it—you’re not alone. Sephora tackled this head-on with a two-part strategy: investing in a semantic foundation, and creating a federated data stewardship model that made trust a team sport.
They empowered business users to actively shape and improve the data definitions behind AI, while simultaneously scaling explainability and reliability in customer-facing applications.
The result? Stronger data culture, fewer hallucinations, and a faster path to personalized experiences customers actually believe in.
Implementation Steps:
Invest in semantic/metadata layers using tools like Atlan or ThoughtSpot's semantic modeling to create consistent data definitions across teams
Appoint business SMEs as data stewards to define terms and enrich data within their own domains
Design AI solutions with explainability baked in, so customers can see the “why” behind each recommendation
Connect business users directly to the data through BI tools where they can update and improve semantic models in real time
Celebrate data stewardship with visibility and recognition programs to strengthen adoption across orgs
“Our investment in the BI space with ThoughtSpot was primarily because of the enablement of meaningfulness to the data... being able to get to our clients where they're interacting with data and being able to enhance our semantic layer.”
Manbir Paul, VP of Engineering, Sephora
🎧 Want to hear how Sephora’s semantic strategy powers real customer impact? Tune into Sephora’s story on The Data Chief.
2. Deploy Machine Learning for Fraud Detection and Risk Management
Want to Cut False Positives by 50%? Here's How Macquarie Did It
Your fraud operations team is probably drowning in false positives while real threats slip through. Macquarie Bank tackled this head-on with ML models that freed up their team to focus on cases where customers actually need help.
Implementation Steps:
Deploy ML models for pattern recognition in fraud detection, focusing on reducing false positive rates rather than just catching more fraud
Implement continuous monitoring with your client protection team to tune models based on real-world performance
Measure operational impact by tracking how much analyst time you're freeing up for high-value customer support cases
Scale gradually from pilot programs to full deployment, monitoring customer satisfaction alongside fraud metrics
"Machine learning has allowed us to significantly reduce somewhere between 50 to 60% with respect to false positive detection, which then has freed up lot of the capacity in the fraud operations team to really focus on the cases where our customers require help."
Ashwin Sinha, Chief Data Officer, Macquarie Bank
🎧 Want your data analysts to stop building dashboards and start fighting fraud with AI? See how Macquarie made the change on the Data Chief podcast.
3. Build Superuser Networks to Scale Data-Driven Personalization
Need to Scale Analytics Across 1,400+ Properties? Here's Hyatt's Playbook
If you're trying to scale data insights across multiple locations or business units, Hyatt's approach to building superuser networks might be exactly what you need.
Hyatt didn’t just scale from 700 to 1,400+ hotels: they built a thriving analytics culture that keeps up with business velocity and drives loyalty growth.
Their secret? Empowering a global network of superusers and running their most strategic data initiatives through agile 90-day outcome cycles.
Implementation Steps:
Identify and train superusers in every business unit to share insights, support local teams, and drive adoption
Build digital communities (like Teams channels) to share learnings and collaborate across properties
Standardize KPIs across regions, while allowing local flexibility
Use 90-day agile pods (small, cross-functional teams with clear ownership) to deliver measurable CX outcomes
Set upfront success metrics tied to customer experience, and review progress regularly with executives
"We want to make sure that we're building super users around the world so that they can help people grow in a self-service way... all the business people that want to lead and run fast never have to talk to the data organization."
Ray Boyle, VP Data and Analytics, Hyatt
🎧 How do you scale personalized CX across 1,400+ properties without bottlenecks? Hear Hyatt’s story on the Data Chief podcast.
4. Design AI for Trust and Explainability in Customer-Facing Applications
Building Customer Confidence in AI? Learn from WEX's Virtual Assistant Strategy
Your customers might get the right answer from your AI, but do they trust it enough to act on it? WEX learned this the hard way and rebuilt their approach around customer confidence, not just accuracy.
Implementation Steps:
Add explainability features that show customers the data sources and reasoning behind AI recommendations
Tailor responses for different customer segments based on their technical literacy and information needs
Iterate based on user feedback rather than just accuracy metrics—trust and adoption matter more than perfect answers
Provide documentation paths so customers can validate AI responses with additional context when needed
"Answering the question correctly is not the same as building confidence that it's the right answer... what information or documentation, what data do you need to validate that answer and go, yeah, that makes sense. I trust it."
Karen Stroup, Chief Digital Officer, WEX
🎧How do you turn AI from a black box into a customer confidence booster? Learn more about WEX’s approach on the Data Chief.
Deliver Better Customer Experiences with AI
AI can’t fix your customer experience if your foundation is broken. But as these leaders show, the real breakthroughs aren’t coming from pure technology: they’re coming from smart, scalable strategies rooted in data trust, business alignment, and AI that’s actually explainable.
Whether you're just starting to modernize your CX strategy or trying to drive more value from existing tools, these best practices give you a roadmap without waiting on perfect data or a massive reorg.
Now’s the time to stop debating if your data is ready for AI—and start using it to create experiences your customers trust and remember. Start your free trial today!
Frequently Asked Questions
How do I actually get business users to become data stewards without it feeling like extra work?
Start by identifying your most engaged business users who already ask great data questions—they're your natural stewards. Sephora made this work by connecting stewardship directly to tools people already use (like ThoughtSpot), where merchandisers could enrich data definitions while doing their regular analysis.
Manbir Paul's team at Sephora found that when stewards could see their contributions improving everyone's data experience, adoption accelerated naturally.
What's the fastest way to prove ROI on self-service analytics tools like ThoughtSpot?
Focus on analyst capacity freed up for higher-value work rather than just user adoption numbers. Macquarie Bank measured this by tracking how many analysts moved from dashboard creation to AI/GenAI projects after rolling out ThoughtSpot—that's your clearest ROI story.
Start with a pilot group of business users who have the most repetitive data requests, then measure how much time your analytics team saves on those requests.
How do I convince skeptical executives to invest in semantic layers when they just want quick wins?
Show them the fraud detection angle—it's the fastest path to measurable business impact. Macquarie Bank reduced false positives by 50-60% specifically because they had clean, semantic data feeding their ML models. Frame semantic layers as "AI insurance" that prevents embarrassing hallucinations in customer-facing applications.
Start with one high-stakes use case where bad data could hurt customers, then expand from there. WEX learned this lesson when their early chatbots failed because customers didn't trust the answers—now they design every AI solution with explainability features.
What's the best way to structure cross-functional data teams without creating more meetings and bureaucracy?
Copy Hyatt's 90-day outcome cycle approach—it forces focus and prevents endless planning. Form small pods (3-4 people max) with clear decision-making authority, combining business, tech, and data team members for each specific outcome.
Ray Boyle's team found that having executives review outcomes every 90 days created urgency without micromanagement. The trick is giving pods real autonomy to execute while maintaining regular check-ins. Set up measurement plans upfront so everyone knows what success looks like, then let the pods figure out how to get there.
