Episode #012 • April 9

Listen on: Spotify | Apple Podcasts

Start with one thing — then watch it compound. Aurelia's entry point into AI was simple: they had a chatbot they hated. Their data science team said they could rebuild it from scratch, and three months later they had a fully custom AI-powered chatbot live on the site. That success gave them the confidence to apply the same technology to their ticketing system, which then unlocked email automation, which is now evolving into proactive order monitoring. None of it was planned as a grand roadmap — it was one thing leading to the next.

Build vs. buy depends on what you actually need. Off-the-shelf support platforms were full of features CarParts.com didn't need and missing the ones they did. Aurelia's philosophy: ask the question honestly for each initiative. They wouldn't build a payment processor, but for customer support they had the internal talent (three data scientists) and simple-enough requirements that building made more sense. The result is a platform they can tweak, extend, and fully own.

Guardrails are the hardest thing to get right. Harper — the AI agent responsible for order tracking — started promising customers she could cancel orders and follow up via email. She couldn't. The team had to track down every affected customer and apologize. The lesson wasn't that AI is dangerous; it was that the instruction "be helpful" without constraints means the AI will find creative ways to fulfill that directive regardless of what it can actually do. Too many guardrails and you get a glorified decision tree. Too few and the AI goes rogue. The balance is ongoing, nuanced prompt work.

AI frees humans for the cases that actually require humans. Before the AI agents went live, contact center staff were spending most of their time on "where's my order" and "what's the status of my return" — fully informational tickets with no judgment required. Now those are resolved automatically. Human agents handle the complex cases: carrier problems, part quality issues, situations where empathy and creative problem-solving actually matter. Less pressure, more meaningful work.

Insight without action is just noise. Aurelia is direct about this: her role is not to gather customer feedback and hand it to marketing. She ties what she's hearing from customers to the company's actual strategic priorities for the quarter and the year — and then she goes do the work cross-functionally with her peers. If a CX leader is surfacing everything that's wrong without connecting it to what the business can prioritize right now, they've become a "Chief Complaint Officer." That's not a role that earns a seat at the executive table.

Win-win or it doesn't get built. Every initiative in Aurelia's organization has to benefit both the customer and the business. The shipping protection program is a clean example: it gives customers instant reimbursement for lost or damaged packages with three simple questions online, while generating revenue for CarParts.com through the partnership. That lens — "how do we add value here, not just fix a problem" — changes every conversation about what CX is actually for.

AI maintenance is the real long game. Building the agents was the fast part. The harder work is keeping them current as policies change, new products launch, and customer questions evolve. Aurelia's team recently received a custom interface that lets them update agent prompts themselves, without relying on the data science team for every change. She compares it to onboarding and ongoing training for human agents — the same discipline applies.

Mid-market companies can compete now. CarParts.com doesn't have a billion-dollar innovation budget. Before AI, competing with enterprises that did felt like a structural disadvantage. Aurelia's view: that gap has effectively closed. Small, focused teams can build sophisticated, custom systems quickly. "It's fair game now," she says — and the evidence from their own build backs that up.

CarParts.com – Publicly traded U.S. e-commerce retailer of automotive parts, shipping approximately 50,000 orders per week

Claude – AI model used internally at CarParts.com and referenced as a tool Jay's consulting firm uses extensively (Aurelia called it "Cloud")

Harper – CarParts.com's custom AI agent for order tracking (named for the rhyme with "tracker")

Penny – CarParts.com's custom AI agent for payment-related inquiries, designed to be direct and concise

Balboa Solutions – Jay Nathan's consulting firm, which uses call transcripts + MCP servers + Claude to build internal knowledge bases

RStudio – Video production tool Aurelia's company uses in their office studio setup