Blog
Perspectives on AI customer service.
How we think about agent architecture, retrieval grounding, and building support automation that support teams can actually trust in production.
Perspective · Nov 3, 2026 · 5 min read
Resolved or just gone quiet? Use re-contact rate to check any AI support numberA deflection or resolution figure can hide customers who gave up or came back. How to check re-contact and reopen rates on your own helpdesk data, and what to ask a vendor.
Perspective · Oct 27, 2026 · 5 min read
Who decides what counts as resolved? Questions to ask any AI support vendorA resolution rate is only as trustworthy as its definition and whoever grades it. Six questions to ask before you sign, and how to test the answers on your own tickets.
Strategy · Aug 20, 2026 · 7 min read
AI customer support for compliance and regtech: what works, what doesn'tKYB/KYC, AML, fraud investigation, debt collection compliance — regtech support questions are highly repetitive and legally consequential. Here's where AI works, and where a wrong answer isn't just unhelpful.
Strategy · Aug 19, 2026 · 7 min read
AI customer support for immigration tech: what works, what doesn'tCase status, document checklists, USCIS processing times, form instructions — immigration tech support is high-volume, high-stakes, and almost entirely documentable. Here's how to get it right.
Strategy · Aug 17, 2026 · 8 min read
AI customer support for healthcare tech: what works, what doesn'tMedical billing, prior auth, credentialing, EMS claims — healthcare tech support questions are high-stakes and highly repetitive. Here's where AI support works, and where it must stay hands-off.
Strategy · Jul 7, 2026 · 7 min read
AI support for staffing and recruiting platformsStaffing platforms have three distinct customer types asking questions constantly — candidates, recruiters, and employers. Here's how AI support works across all three without confusing them.
Strategy · Jul 7, 2026 · 7 min read
AI customer support for HR and payroll platformsPayroll questions are time-sensitive, compliance-adjacent, and highly repetitive. Here's why AI support is a natural fit — and where it needs to stay hands-off.
Strategy · Jul 7, 2026 · 8 min read
AI customer support for fintech: what works, what doesn'tFintech companies face a specific support challenge — high-anxiety questions from customers for whom money is never abstract. Here's how to build AI support that works in that context.
Perspective · Apr 28, 2026 · 6 min read
Resolution rate is the only AI support metric that mattersDeflection rate tells you how many customers the bot touched. Resolution rate tells you how many customers got their problem solved. Only one of these is useful.
Product · Apr 24, 2026 · 5 min read
Why AI support needs guardrails on destructive actionsCancellations, refunds, and address changes on shipped orders can't be undone. Here's how to build confirmation flows that keep humans in the loop on irreversible decisions.
Engineering · Apr 21, 2026 · 7 min read
How to build a knowledge base for AI customer serviceThe quality of your knowledge base determines the quality of your AI agent's answers. Here's what to include, what to leave out, and how to maintain it.
Operations · May 5, 2026 · 6 min read
Escalation design: when the AI should hand off and howEscalation is not failure. A well-designed AI agent hands off more gracefully than most human routing systems. Here's how to set the criteria that make it work.
Operations · May 8, 2026 · 8 min read
What makes a knowledge base good enough for an AI agentMost AI support failures trace back to the knowledge base, not the agent. Here's how to audit your content before you go live — and what to fix when the agent keeps getting things wrong.
Strategy · May 10, 2026 · 7 min read
Five metrics that actually tell you if your AI support agent is workingDeflection rate is easy to fake. Here are the five metrics that separate AI agents doing real work from AI agents that look good on a dashboard.
Engineering · May 13, 2026 · 5 min read
How to scope AI agent actions so they don't do more than intendedThe same agent that looks up an order can also cancel it — unless you scope your actions carefully. Here's the design discipline that keeps AI agents safe to deploy.
Engineering · May 15, 2026 · 7 min read
How to evaluate an AI support agent before you trust it in productionRunning an eval on 500 historical tickets before go-live is the difference between shipping with confidence and discovering your resolution rate three weeks in. Here's how to do it right.
Product · May 17, 2026 · 6 min read
Voice vs chat: what changes when your AI agent picks up the phoneThe same AI agent that resolves your chat tickets will behave differently on a voice call. Here's what actually changes and how to design for it.
Want to talk about any of this?
We're thinking through these problems with our design partners right now. If you want in on that conversation, get on the calendar.