By Teresa Torres · producttalk.org · @ttorres on X · LinkedIn
This episode of a product podcast walks through how Xelix built an AI Helpdesk for accounts payable teams, which often receive over a thousand vendor emails daily. The product turns that inbox noise into structured tickets, enriches them with ERP data, and pre-drafts replies with confidence scores for human review. The team describes scoping the problem narrowly, prototyping in thin vertical slices, and prioritizing retrieval and vendor matching over raw model power. They also discuss shifting the UX from an inbox clone to a ticket-first view and measuring impact through handling time and auto-resolution. The discussion matters for anyone building AI features into operational workflows where trust and accuracy hinge on good data grounding.
01Key takeaways
- Start with a narrow, high-volume, high-cost request type such as invoice status or payment reminders.
- Invest in retrieval and entity matching; accurate enrichment often beats simply swapping in a bigger model.
- Use thin vertical slices and beta users to validate each increment before expanding scope.
- Keep a familiar interface for onboarding, but move to a ticket-first design to unlock AI-driven features.
- Show match quality and response confidence so humans know when to review and edit drafts.
- Measure outcomes such as handling time, messages sent from the tool, and percent auto-resolved rather than relying on impressions.
02Key sections
- Problem and product overview
- The team explains the pain of accounts payable inboxes flooded with vendor messages and introduces the Helpdesk as a way to structure that work. The product aims to make responses faster and more consistent.
- Carpaccio-style prototyping
- They describe building in daily thin slices rather than large batches, which let them test with real users early. Beta feedback shaped what got built next.
- Retrieval and vendor matching
- Accurate replies depended on correctly identifying the vendor and linking related invoices and ERP records. Enrichment quality mattered more than model size.
- UX pivot to ticket-first design
- A familiar inbox view eased onboarding, but a ticket-centric interface was needed to support the AI features. The shift balanced familiarity with new capabilities.
- Measuring impact and next steps
- The team tracks messages sent from the Helpdesk and the share of requests resolved automatically. Future work includes targeted generation, multiple specialized responders, and more agentic routing.
03From the post
“Listen to this episode on: Spotify | Apple Podcasts Accounts payable inboxes can see 1,000+ vendor emails a day. Xelix’s new Helpdesk turns that chaos into structured tickets, enriched with ERP data, and pre-drafted replies—complete with confidence scores. In this episode, Claire Smid (AI Engineer), Emilija Gransaull”
04Frameworks mentioned
Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.