Luca Micheli: The AI Pivot That Took Customerly From $100K to $1M ARR
Six years after his first appearance on Founder Views, Luca is back with the real story of how AI forced a full business model and go-to-market shift.Customerly went from a seat-based, product-led support platform for small SaaS teams to an AI-first customer service engine selling into mid-market and enterprise, where volume and ROI are obvious.In this episode we get into:The AI pivot: why they refused to build “old-school chatbots,” and how ChatGPT changed what was possibleQuality metrics that matter: error rate, confidence thresholds, escalation triggers, and why AI CSAT can be higher than humansWhat actually trains a good AI agent: knowledge base structure, what not to upload, and how hallucinations happen in the real worldAutomation outcomes: average ticket closure rates, what drives 80%+ vs 40–50%, and how teams improve over timeEnterprise GTM shift: moving from product-led to sales-led, filtering signups, longer cycles, bigger ACVOutbound reality: why the agency failed, what changed when they built outbound internally, and the tooling stack (Clay, Apollo, Lemlist, Pipedrive)Founder sales lessons: Challenger Sale thinking and why founders still need to own sales earlyThe Arena The Arena is a private Skool community for SaaS founders who are actively building and selling. I share real-time decisions, experiments, and assets as I use them while growing a bootstrapped SaaS.No theory. No polish. Just execution.Learn more at: https://www.skool.com/the-arena/Chapters / Timestamps00:00 – Reunion after 6 years and what changed (COVID + AI era)01:27 – Luca intro: what Customersly does today02:30 – From $100K ARR to near $1M and why pricing changed05:08 – “Chatbots are shit”: how they built AI without the bad UX07:10 – Under 1% error rate and reducing hallucinations09:52 – Grounded AI, intents, and automating beyond FAQs11:09 – Closure rate benchmarks and what “good” looks like16:41 – How to pick an AI support tool that actually works18:20 – Training mistakes: transcripts, clutter, and marketing banners causing hallucinations20:46 – Confidence thresholds and escalation as a feedback loop22:48 – How long it takes to move from 45% to 70–80% automation24:34 – Should AI learn from your inbox? Pros, risks, and why they avoid it29:41 – Implementation timelines: small teams vs enterprise rollouts31:38 – Why AI CSAT can beat humans (speed wins)35:46 – Escalation rules: human request, sentiment, low confidence, missing info37:21 – Going enterprise: ARPU jump and sales-led reality41:02 – Outbound experiment: agency failure and building it internally43:32 – LinkedIn ads + Clay targeting + the masterclass lead magnet49:25 – Challenger Sale and shifting the conversation53:20 – Founder lesson: why you can’t outsource what you haven’t done58:54 – Outbound stack: Clay, Apollo, Lemlist, Sales Nav, Pipedrive01:05:12 – 2026 vision and wrap