How to build tools on your firm's website with Claude

Introduction

95% of the visitors to a firm's website leave without a trace. Not because the offer is bad, but because a contact form is no longer enough to trigger action. The real break is giving them something concrete before they ever call.

That is exactly what AI tools make possible: a settlement agreement calculator, a lease scanner, a diagnostic for a disputed situation. The visitor interacts, gets a useful first analysis, and leaves their details in the same breath. On your side, a structured case lands straight in the CRM, with the context, the level of urgency and the qualification of the matter.

Here is how it gets built, brick by brick.

The mechanics behind every tool

The working pattern is always the same, whatever the firm's specialism.

The visitor uses the tool on a dedicated page of the site. The experience feels like a guided conversation, with no heavy form, and hands back immediate value: an estimate, an analysis, a list of points to watch. Meanwhile the Claude API analyses the situation and produces a structured answer in JSON carrying the key information about the matter.

That JSON goes by webhook to Make (formerly Integromat), which handles the rest: creating or updating the contact in the CRM, attaching a deal, launching the right marketing sequence. A hot matter triggers a partner alert within the minute. A warm matter enters a seven-day email sequence. A cold prospect joins the newsletter.

All of it without human intervention at every step.

The recommended technical stack

Four tools are enough to put together a solid set-up, with no exotic dependency and no disproportionate infrastructure cost.

WordPress or Webflow for the front end. The tool embeds through a simple HTML block on an existing page. If the site is already on WordPress, stay on WordPress. For firms starting from scratch, Webflow gives more control over design without technical debt.

Make for the automation. Better suited than Zapier for handling complex JSON structures, with pricing by volume of operations that stays predictable even at scale.

HubSpot or Notion as the CRM. HubSpot for firms already structured who want scoring and email workflows. Notion for those looking for a simple first container before moving up.

The Claude Sonnet API for the intelligence. Claude produces the most careful and best structured legal analysis among the large models today. Sonnet offers the best value for high-volume uses.

On the technical side: how the API call is structured

The call to Claude is never made straight from the browser, so the API key is never exposed. The good practice is to create a serverless function (Netlify Functions, Vercel Functions, or a PHP endpoint on the WordPress side) that acts as the intermediary.

That function receives the visitor's message, adds the tool's system prompt, calls https://api.anthropic.com/v1/messages with the key held in an environment variable, and returns the answer to the front end. Count roughly 80 lines of JavaScript for the mini-chat on the front end, and around thirty lines for the serverless function.

The system prompt is the central piece. It is what gives Claude its role (lease analyser, settlement calculator, dispute diagnostician), the tone to adopt, the questions to ask step by step, and above all the output format expected at the end of the conversation: a JSON block with the fields contact, dossier, and qualification.

Wiring up Make in five steps

Once the conversation ends, the front end extracts the JSON from Claude's last answer and POSTs it to a Make webhook.

Step 1. In Make, create a scenario that starts with a Webhooks · Custom webhook module. Give it a sample payload so it infers the structure automatically.

Step 2. Follow it with a HubSpot · Create or update a contact module, with the email as the uniqueness key, filling the contact's standard properties.

Step 3. A module named Create a deal creates an attached deal and fills the custom properties created beforehand: type_dossier, score_solidite, montant_estime, chaud.

Step 4. A module named Enroll a contact in a workflow triggers the right sequence depending on the value of type_dossier.

Step 5. For firms without a CRM, replace the HubSpot modules with Notion · Create a database item, and add a conditional Slack · Send a message module that alerts the partner as soon as the field chaud is true.

Count half a day for the first installation, a few minutes for each variation after that.

What we see on the first firms equipped

The consolidated figures on the first deployments give a stable order of magnitude: between four and six times more qualified leads a month than a classic contact form, an acquisition cost divided by three per signed matter, and an average completion time of two minutes on the visitor's side. By the first meeting, the partner already has the whole structured context.

The variations depend on the volume of traffic upstream and on the rigour of the commercial follow-up. But the principle is stable: turning invisible traffic into qualified matters, without multiplying manual steps.