A Manchester law firm was spending 14 hours a week on intake. The agent now books the consults.

NorthBridge Legal had a working website. Forms came in. That was the problem.

Someone on the intake team had to open every submission, decide if it was a real matter, chase the details that weren’t on the form, then play calendar tennis until a consult landed. Fourteen-plus hours a week. Partners were still getting pulled into the admin when volume spiked.

That’s not a marketing problem. It’s an intake problem. We built the agent around that, not around a chatbot demo.

What was actually eating the week

Three jobs, all manual, all mixed together:

1. Triage. Read the form. Is this a client, a tyre-kicker, or someone who needs a different firm?
2. Chase. Half the submissions were missing a document, a date, or a phone number. That meant a polite email, a wait, a follow-up.
3. Book. Once the file was complete enough, find a slot that worked for a solicitor and the prospect. Calendly existed. The human still had to get the person onto it.

Fourteen hours sounds like “a process.” It was three people context-switching through a shared inbox.

We’d already written about the outcome in passing: that legal client went from about three hours of inbox triage a day to twenty minutes reviewing only the high-priority threads. This is the build behind that line.

What we shipped

A GPT-4 intake agent sitting on top of the existing contact form. No new website. No “please describe your issue to our virtual assistant” widget on the homepage.

When a form lands:

• The agent reads the submission in full, not a keyword match.
• Clearbit enriches the sender (company, role, domain) so scoring isn’t just “how nicely they wrote the message.”
• It scores the lead against NorthBridge’s ideal-matter profile.
• It drafts a personalised reply: next step, missing items, or a polite redirect.
• If the lead is a fit, it books the consult straight into the firm’s Calendly.
• A solicitor still sees the thread. They just don’t have to write the first three emails.

Zapier is the plumbing. OpenAI is the brain. Calendly is the calendar. Clearbit is the extra context. The firm already had most of those tools. We wired them into one path.

What we refused to automate

Anything that looks like legal advice.

The agent can ask for a missing date of incident. It can say “this sounds like employment, here’s who you’ll speak to.” It cannot tell someone they have a claim. That stays with the solicitor, on the booked call.

If the model isn’t sure, it doesn’t guess. It flags the thread and a human takes it. Twenty minutes a day is that review queue, not a rubber stamp.

Why this one converted instead of sitting in Zapier hell

Most “AI for law firms” pitches start with a chatbot. Clients don’t want to chat. They already filled in a form. The form is the intake.

So we started at the form, not the homepage. The prospect’s experience barely changed. The firm’s week did.

That’s also why it was a Zapier build, not a six-month custom platform. The firm needed intake to stop leaking partner time this quarter. Custom later, if the volume justifies it. They own the workflows. No lock-in to a legal-tech SaaS that holds their leads hostage.

If your intake looks like this

This pattern isn’t only for law. It’s any service business where a form is doing three jobs badly: qualify, complete, book.

Trades with quote requests. Accountants in January. Consultancies drowning in “quick questions.” Same shape.

If you want the long version of “what even is this,” we already wrote how small businesses are getting 20+ hours back with AI automation. This post is the legal-firm proof, not another list of tools.

The short version

NorthBridge didn’t need another website. They needed the form to finish the job.

We scoped it, priced it fixed, and shipped an agent that enriches, scores, replies, and books. Partners went back to matters. Intake went back to being a queue, not a second job.

Full project notes are on the case study. If your inbox looks like theirs, get a quote — I’ll look at the actual workflow and send a fixed price. First reply within four hours.