AgentryBook a call
← Writing
Case study

An AI agent that audits any business in 30 seconds

Drop a website, get back the three highest-impact automations for that company, with honest time-savings for each. A live agent that doubles as our best lead magnet.

TL;DR

We built a public agent that takes a single input, a company website, and returns a specific automation audit: what they do, the three workflows where an AI agent would save the most time, and where we would start. It researches the company from their live site and the open web, reasons about their actual operations, and writes a grounded report in about thirty seconds. It is genuinely useful on its own, and every run captures a qualified lead into our pipeline.

The problem

Cold outreach that opens with “can I help you with AI?” gets ignored, because it asks for the prospect's time before giving them anything. The thing that actually earns a reply is specific, useful insight about their business — but producing that by hand means an hour of research per prospect, which does not scale.

We wanted to give every prospect a real, tailored automation audit before asking for a minute of their time — and to do it in seconds, not an hour.

What we built

A single-input agent that runs a short research-and-reason loop:

website → research → reason → audit → capture lead

Engineering decisions worth calling out

What separates a useful audit from AI slop:

Grounded, or it does not ship

Every opportunity has to name something the business actually does, pulled from their real homepage copy and live web results. The prompt treats generic advice as a failure. If the inputs are thin, the agent says so and stays conservative rather than inventing a number.

Honest estimates beat impressive ones

Time-savings are ranges, complexity is labeled plainly, and every report ends with a caveat about what a real call would confirm. The goal is that a sharp operator reads it and thinks 'that is exactly right' — value first, even if they never hire us.

Open by design, bounded by cost

A lead magnet should be usable by anyone with no friction, so there is no login and no gate before the value. Abuse and spend are bounded instead by per-IP and daily caps plus a per-domain cache, so repeat hits are free and the worst case is capped.

Degrades gracefully

Some sites block scrapers. The agent pulls from the homepage and an independent web-search source, so when one is blocked the other carries the report. A site it genuinely cannot read gets an honest error, not a hallucinated audit.

Every run is a captured lead

The moment a report is generated it lands in our CRM pipeline as an inbound deal, with the prospect's email and the audit summary attached, and pings us to follow up while it is warm. The tool markets, qualifies, and routes in one step.

The stack

A Python agent on serverless compute, the Anthropic API for the research synthesis, a live web-scraping and search layer for grounding, Postgres for capture, and a Next.js front end. The same building blocks we use for client work, pointed at a marketing problem.

Why it works as a lead magnet

Interactive tools convert several times better than static downloads, because the prospect invests their own data and gets something tailored back. This one earns the reply three ways at once: it is a live agent anyone can watch work (the proof), it captures a qualified lead with full context (the inbound), and we can run it on a prospect and open with a finished audit instead of a cold pitch (the outbound). One build, three jobs.

See it work on your business

Drop your website and watch the agent build your audit. Free, about 30 seconds.