Replacing a patchwork of tools with one system for a solo legal practitioner

The client

An independent criminal and motoring-law advocate. Most of his work comes from one referring solicitor, mostly over WhatsApp.

The problem

Every case ran across five or six tools: messaging threads, paper files, word-processor invoices, a spreadsheet for payments, cloud folders and a calendar. In practice this meant:

• Re-typing. Details from a message or scan were copied by hand into invoices, notes and calendar entries.
• Lost paperwork. Court papers and recordings were hard to find when needed.
• Missed follow-ups. Chasing papers, clients and payments depended on memory.
• Slow reporting. The weekly report to the referrer was compiled by hand.
• No audit trail. There was no reliable record of what changed, when or why.

We built a private web app around one sequence: Instruction → Conference → Proceeding → Hearing → Outcome → Payment. It is deliberately not a generic CRM. Every screen supports one of those stages.

• Instructions arrive ready to review. WhatsApp messages, emailed papers and scans are read automatically. The practitioner sees the original beside the extracted details and confirms with one click.
• Each case has one home. Documents, notes, hearings, invoices, payments and a full timeline sit in one place and are searchable in seconds.
• Admin follows the case. Invoices, reminders, calendar entries and attendance notes are generated from the case data instead of retyped.
• Payments reconcile themselves. A bank statement is matched to invoices. Confident matches are applied, and uncertain ones are queued for review.
• Reporting is a review step. The weekly report is drafted automatically, and the practitioner edits and sends it.

A deliberate decision: no AI

Because the data is sensitive and a misread court outcome could do real harm, we used transparent, rule-based extraction instead of a language model. Every automatically derived value shows a confidence level and waits for confirmation. Nothing is overwritten silently, and an unclear outcome is left blank rather than guessed. The client could trust the system, and it avoids sending client data to third-party AI.

Security

Client and legal data is protected by:

• mandatory multi-factor sign-in
• private storage with expiring file links
• role-based access
• an append-only audit log
• archiving instead of deleting
• full matter export

How it was delivered

• Built with Next.js, TypeScript and Supabase, and hosted on Vercel.
• Integrations (WhatsApp, email, calendar, transcription) were built behind mock providers, so the whole workflow was usable before any third-party account existed.
• The client used it while it was being built. About 90 iterations in seven weeks turned his feedback into working changes.

A single source of truth for every case, a daily view of what needs doing, and a workflow where his work comes down to reviewing, deciding and attending court.