How Mejuri and Spanx are implementing AI in eCommerce: Lessons from NRF 2026
NRF 2026 in New York marked a clear inflection point — AI in eCommerce shifted from "what's possible" to "what's working." Following our time on the ground at NRF, we partnered with NORA to host a webinar with senior leaders from Spanx, Mejuri, and Shopify, exploring their real-world AI practice and extracting practical AI implementation strategies for Australian and New Zealand retailers.
Our panel, moderated by co-founder Kelly Brown, featured Kyla Robinson, SVP of Digital at Spanx, Rohit Nathany, Chief Product and Technology Officer at Mejuri, and Eduardo Frias, Field CTO at Shopify, who shared what's actually working in their businesses right now.
AI delivers measurable outcomes today
Spanx now manages over 50% of customer service inbounds through AI agents, with CSAT scores higher than human agents. During Black Friday, this allowed the team to operate more leanly whilst handling increased volume, delivering material cost savings in engineering.
"The quality response we get when customers tell us their interaction after is actually higher with our AI agents than our human agents," Kyla explained.
Industry benchmarks suggest 50-60% automation potential for customer service, whilst Mejuri uses AI for email personalisation and content generation. These aren't experiments anymore — they're operational standards.
Agent-led commerce requires quality product data
The Shopify-Google commerce protocol announcement enables agent-to-agent commerce, but success depends on data enrichment. "LLMs do not understand PDPs," Eduardo noted. "If LLMs do not understand your data, your products won't be surfaced."
Spanx has partnered with specialists to optimise product data for LLMs, seeing traffic lifts. Mejuri achieved a high share of voice in AI search, but Rohit observes conversion patterns vary by category: jewellery customers still want to browse multiple options despite AI discovery.
How to prepare your product data for AI-led discovery:
- Audit product data structure for LLM compatibility
- Assess whether your category suits agent-led conversion or primarily discovery
- Partner with specialists for data enrichment
- Build measurement frameworks for traffic and conversion validation
Change management matters more than technology
"It's not just changing how you work, but potentially impacting your job itself. There's a real fear there," Rohit observed.
Spanx addresses this head-on through "human in the loop" — two team members solely focused on training LLMs on the brand's "girlfriend to girlfriend" tone. This proves critical for handling sensitive questions around weight and postpartum with appropriate care.
Demand planning teams often resist moving from Excel despite higher accuracy promises. A Spanx-Palantir pilot for inventory forecasting showed success in specific use cases but required significant change management investment.
Premium human experiences remain valuable. Spanx directs VIP customers to a named human via text and curates in-person experiences for loyalty members — because transactional purchases and emotional ones demand distinct strategies.
Rethink build vs. buy economics
Here's where it gets interesting: Spanx's wholesale team reduced a two-week manual task to one hour using Claude. Mejuri's finance team cut bank coding workload by 50% through semantic matching.
Kyla's framework cuts through the noise: can your team reach 80% of the use case at 5-10% of the cost? If yes, build. If not, buy. She tells her team: "If they built it from scratch, I say I don't want to see it."
Eduardo reinforces the pragmatism: don't build your own LLM. Leverage OpenAI or Anthropic and customise for brand voice.
Implementation steps:
- Identify manual tasks taking 2+ hours weekly
- Run two-hour experiments with AI before hiring
- Conduct exploratory spikes before committing to SaaS
- Ask vendors: "Why can't I build this myself?"
- Reward experimentation, even without transformative outcomes
What this means for ANZ retailers
AI has moved from experimentation to operational reality. The brands seeing results start with proven use cases, invest in data foundations, address change management proactively, and foster experimentation cultures.
For Australian and New Zealand retailers, the question isn't whether to adopt AI, but where to start and how quickly to scale.
Want deeper insights? Download the full webinar recording on AI implementation, agent-led commerce and preparing your team for 2026.
Download the full webinar
Listen to the conversations in detail including what to expect from AI in the near future:
