Results
The challenge
Their inbound came from five places — Shopify contact forms, Gmail, Instagram DMs, WhatsApp, and two landing-page forms. A single founder was triaging everything manually, often hours after the lead had cooled.
Two problems: speed (leads were going cold) and consistency (the same buyer was getting different answers depending on who replied).
What we built
We built a single agent that:
- Ingests every inbound channel through their existing stack (no new tools).
- Pulls the customer's order history, segment, and last interaction from Shopify.
- Drafts a personalized reply grounded in the brand's voice guide and FAQs.
- Sends automatically if confidence is high; escalates to the founder with a pre-drafted reply otherwise.
All actions are logged in Google Sheets with the agent's reasoning, so the team has a full audit trail.
How it shipped
The whole project went live in 3 weeks. Week 1 was discovery and stack mapping; weeks 2–3 were build, eval, and gradual rollout (first read-only, then send-with-approval, then fully autonomous for the top 3 inquiry types).
Post-launch, we kept a 2-week optimization window where we tuned the draft prompts based on what the founder was editing most. That cut the manual-edit rate from ~30% of drafts to under 5%.