CASE STUDY / SMB AI / Aug 18, 2025
Every inquiry answered before the phone rings twice
AI intake for a small business: inquiries answered, qualified, and summarized before a human picks up.
- Client
- Zeelume
- Vertical
- SMB AI
- Stack
- Next.js · TypeScript · LLM APIs · Workflow Automation
Customer inquiries stopped going to voicemail. The assistant answers first, asks the qualifying questions, and hands the owner a summary — not a missed call.
Context
Zeelume is a small business where, like most small businesses, the same person doing the work is the person answering the phone. Every inquiry that arrived during a job either interrupted the work or went to voicemail — and voicemail is where new customers give up.
The problem
The owner's day held two jobs: doing the work and fielding the same dozen questions about pricing, availability, and scheduling. Neither job was getting full attention, and after-hours inquiries got no attention at all.
What Haven built
- An AI intake assistant grounded in Zeelume's actual pricing, policies, and calendar
- Qualification flow: the assistant asks what the customer needs before the owner spends a minute on it
- Written guardrails — what the assistant may quote, what it must never promise, when it must hand off to a human
- Summaries delivered where the owner already looks, with the full transcript attached
- Escalation for anything unusual, so edge cases reach a person instead of an improvised answer
Results
Inquiries get a useful answer at any hour, and the owner starts each conversation already knowing what the customer needs. The guardrails have held: the assistant qualifies and schedules, and humans still make the promises.
Build something like this
Ten minutes with the Brain turns your version of this problem into a structured brief — and a fixed-scope plan within 48 hours.
Start a projectMORE SMB AI WORK
SMB AI / HavenTech
Our own front door runs on the product we sell
Haven's own platform: AI intake, client portal, and delivery operations — the first customer of everything we build.
- Next.js
- TypeScript
- PostgreSQL
- LLM APIs
- Stripe