Aurelia Pollet, VP of Customer Experience at CarParts.com, led a small team that built a fully custom AI chatbot and ticketing system from scratch in three months.

She shares how they did it, what went wrong when their AI agent Harper made promises it couldn't keep, and why the real value of AI isn't cost-cutting — it's creativity that was never possible before.

KEY TAKEAWAYS

• Build vs. buy is a real question: Off-the-shelf tools were bloated with features they didn't need and missing the ones they did — building from scratch gave them full control at lower cost.
• Start with one small thing: The chatbot led to the ticketing system, which led to email automation. One use case compounds into the next.
• Guardrails are the hardest part: Their agent Harper promised customers she could cancel orders — she couldn't. Too few guardrails and AI goes rogue; too many and you've just built a robot.
• AI enables proactive CX: With 50,000 orders/week, proactive outreach was impossible before. Now they're building toward monitoring every order and surfacing problems before customers call.
• Insight without action is noise: Aurelia doesn't want to be the "Chief Complaint Officer." Her job is to tie customer insights to company strategy and act — not just report.
• Maintaining AI is a full-time job: Agents need ongoing updates just like human employees — new policies, new products, new questions. The build is the easy part.

ABOUT OUR GUEST

Aurelia Pollet is the VP of Customer Experience at CarParts.com, a publicly traded e-commerce retailer shipping ~50,000 orders/week. A mechanical engineer by training, she spent 15 years in luxury goods before moving through B2B and nonprofit roles. She led CarParts.com's AI-native rebuild of its entire support infrastructure. Connect with Aurelia on LinkedIn

CHAPTERS

• 03:46 - Welcome and introduction
• 05:42 - Aurelia's background: from mechanical engineer to VP of CX
• 07:05 - What "end-to-end CX" means at CarParts.com
• 10:09 - Adding value at scale: the win-win framework
• 12:26 - Building an AI chatbot from scratch in 3 months
• 14:34 - From chatbot to full AI ticketing system
• 16:51 - How AI frees humans for empathy and complex cases
• 20:08 - Gathering customer insight across the journey
• 24:50 - Insight as action (avoiding the "Chief Complaint Officer" trap)
• 26:50 - The build vs. buy decision
• 32:57 - Hard lessons: Harper's promises and the guardrail problem
• 38:54 - What's next: stabilizing and scaling the AI stack
• 43:23 - Advice for leaders who haven't started yet

About the Show: Chief Customer Officer Podcast is a show about real strategies for customer-led growth in the AI era—from leaders actually executing, not just talking about it.

Your Hosts:

• Jay Nathan – CEO of Balboa Solutions and Co-Founder of ChiefCustomerOfficer.io
• Jeff Breunsbach – Head of Customer Success at Junction and Co-Founder of ChiefCustomerOfficer.io

Jay Nathan and Jeff Breunsbach break down what 50+ CS leaders are actually doing with AI — and what most teams are still getting wrong.

Jeff shares fresh takeaways from a Planhat-organized event in Boston, and Jay goes deep on the Claude skills his team has built to scale coaching, standardize deliverables, and rethink how prep calls even work. No hype, just real strategies in the field.

KEY TAKEAWAYS

• We're still in early adopter territory: Jeff attended an event with 60 CS leaders and walked away realizing that even among leaders who talk about AI daily, widespread team adoption is far behind what the tech echo chamber suggests. The majority of companies are still evaluating — not executing.
• You don't know your processes as well as you think: Before you can automate or augment a workflow with AI, you have to actually know what that workflow is. Jay and Jeff agree: most teams haven't documented or visualized their core processes, which is the real bottleneck to AI leverage.
• Top-down mandates don't drive AI adoption: "Just use AI" from the CEO doesn't work. Real adoption requires demonstrating behaviors at every layer of the org, giving teams specific problems to focus on, and creating regular show-and-tell moments where people can see and iterate on each other's work.
• Call transcripts are your most underutilized asset: Jay calls Fathom call recordings "the most valuable thing we have in our company."
• Claude skills as a team force multiplier: Jay built an executive readout coaching skill in Claude that replicates his own feedback patterns, so every team member starts their deck review at a higher baseline.
• The future CSM is a technical account manager: Multiple CS leaders at Jeff's event flagged this shift — the CSM role is evolving toward one that requires real technical curiosity.
• Forward deployed engineering is closer than you think: Jay makes the case that giving CSMs a local build of your product — and the ability to generate a feature-based pull request from a customer conversation — is entirely possible today.
• Build a center of excellence, not just a mandate: Sustainable AI adoption at the team level needs a few dedicated people vetting standard tools, scheduling engagement touchpoints, and measuring outcomes — not just an org-wide Slack message to "go use AI."

CHAPTERS

• 00:00 - Intro & Jay's flight back from Seattle
• 01:29 - Jeff's takeaways from a Planhat AI event in Boston
• 03:15 - Crossing the chasm: where is AI adoption actually at?
• 06:59 - We don't know our processes as well as we think
• 09:21 - Driving team-wide AI adoption: what actually works
• 11:43 - Measuring AI impact: KPIs and the multi-attribution problem
• 17:47 - Individual vs. team AI adoption — why standardization matters
• 18:39 - Jay's executive readout coaching skill in Claude
• 22:52 - Jeff's product marketing skill: Notion to branded one-pager
• 26:28 - Call transcripts as the ultimate knowledge source
• 27:15 - Automating product feedback with Fathom + Linear PRDs
• 29:42 - Forward deployed engineering and voice of customer at scale
• 33:28 - The future CSM is technically fluent
• 35:07 - What to look for when hiring CSMs right now
• 39:17 - Building a center of excellence for AI adoption
• 42:00 - Jay's website glossary and partner portal (show & tell)
• 46:08 - Wrap up

Jay Nathan and Jeff Breunsbach get into the foundational data problem most SaaS teams are ignoring — and why it's about to become a crisis. If your data is siloed, your AI agents will be siloed too.

Plus: Jay introduces "Company as Code," a markdown-in-GitHub system for building personalized outreach and landing pages at scale — and why the same playbook applies directly to customer success.

KEY TAKEAWAYS

• Start with a customer master record: Before any AI agent can help you, you need a deduplicated view of your customer outside your CRM. Jay spoke with companies from 150 to 3,500 people this week — they all have the same problem.
• Company as Code: Jay is documenting personas, ICP definitions, and outreach styles as markdown files in a GitHub repo — one canonical version that every AI tool pulls from, updated by anyone on the team via pull request.
• Siloed AI = siloed data with a new wrapper: If your sales AI only sees sales data and CS's AI only sees CS data, you've rebuilt your silos with an AI layer on top. Fix the data first.
• Personalization is now just token costs: Custom landing pages, personalized emails, interactive surveys — what used to require massive investment is now a few tokens.
• Existing customers need outbound too: Just because someone's a customer doesn't mean they understand your product. Use the same personalization playbook to re-engage and educate your base.
• ABM belongs in Customer Success: Jay shares the "Spreading the FLU" story — a field-level understanding program that educated every influencer at top accounts. CS teams should run the same play.
• AI-powered upsell prioritization: Use Fathom recordings, emails, and Slack signals to surface the top 10–20 customers most ready for a new product — before sending a rep in cold.
• The CSP question is getting louder: "It's 2026. Do you need a customer success platform anymore?" Jay is hosting a webinar with two practitioners who built in opposite directions.

CHAPTERS

• 00:01 - Weekend updates
• 03:45 - Unifying customer data at Junction with BigQuery and Metabase
• 07:00 - Building a customer master record outside your CRM
• 10:10 - Why siloed AI is just siloed data with a new wrapper
• 11:48 - Company as Code: markdown files in GitHub as a single source of truth
• 14:45 - Personalized landing pages for prospects — and customers
• 19:00 - The risk side: when custom-built tools fall short
• 23:45 - Using Claude Cowork to research private companies via public comparables
• 27:00 - Applying outbound personalization to your existing customer base
• 29:00 - AI-powered upsell prioritization using call recordings and signals
• 33:00 - Spreading the FLU: account-based marketing applied to CS
• 35:00 - QPR as an interactive landing page for top accounts
• 36:45 - Webinar preview: Do you still need a CSP in 2026?

About the Show: Chief Customer Officer Podcast is a show about real strategies for customer-led growth in the AI era—from leaders actually executing, not just talking about it.

Your Hosts:

• Jay Nathan – CEO of Balboa Solutions and Co-Founder of ChiefCustomerOfficer.io
• Jeff Breunsbach – Head of Customer Success at Junction and Co-Founder of ChiefCustomerOfficer.io