← Case Studies
AI Agent · Legal Services

Cutting probate inventory drafting from 6 hours to 11 minutes

An Azure AI Foundry agent that reads estate documents, extracts structured data, identifies beneficiaries and assets, and drafts a complete estate inventory. Built in 4 weeks. Evals included.

94%
Groundedness
Outputs traced to source documents
89%
Task completion
Full inventory drafted without escalation
11 min
Average latency
From document upload to draft inventory
30
Test cases
Across handwritten, typed, and edge cases
Demo

See it working

Demo video — coming soon

The Problem

Estate inventory drafting is expensive, slow, and error-prone

Probate solicitors in the UK spend an average of 5–6 hours drafting a single estate inventory from a will and supporting documents. The process is manual: read the will, identify assets and beneficiaries, cross-reference title deeds and financial schedules, classify each asset by type, and structure the inventory in the correct format. Handwritten wills, inconsistent document quality, and complex estates make it slower still.

For a firm handling 40–60 probate matters per month, this adds up to more than a full-time equivalent spent on a task that doesn't require legal judgment — it requires careful reading and structured data extraction. Every hour spent on drafting is an hour not spent on the parts of probate that actually need a solicitor. And errors in estate inventories create downstream delays, client complaints, and in some cases, regulatory exposure.

The Solution

A 5-step agent on Azure AI Foundry

The agent handles the entire drafting workflow autonomously, with a practitioner review step at the end. It doesn't replace the solicitor — it eliminates the parts that shouldn't require one.

Architecture
Document UploadDocument IntelligenceAI Foundry AgentAI SearchEstate Inventory Draft
Azure AI Foundry · Azure AI Search · Document Intelligence · Semantic Kernel · Azure AD B2C
01

Document intake

Will documents, supporting schedules, and title deeds uploaded via the practitioner interface. Document Intelligence extracts text from PDFs, scanned images, and handwritten originals.

02

Entity extraction

The agent identifies and normalises key entities: testator name, date of death, executor names, property addresses, financial institution names, and account references.

03

Beneficiary identification

Reads the will to identify beneficiaries, their relationships, and the applicable inheritance rules. Flags ambiguous or conditional bequests for practitioner review.

04

Asset classification

Classifies each identified asset by type (property, bank accounts, investments, personal effects) and applies the correct valuation methodology for each category.

05

Estate inventory draft

Generates a complete, structured estate inventory in the firm's standard format — ready for practitioner review, not a first draft requiring major editing.

Evals

30 test cases before deployment

We ran 30 test cases across three document types: clean typed wills, handwritten wills, and edge cases (documents with ambiguous bequests, multiple property addresses, or missing schedules).

MetricScoreWhat it measures
Groundedness94%Outputs traceable to source documents
Task completion89%Full inventory drafted without human escalation
Entity accuracy96%Names, dates, and references correctly extracted
Asset classification91%Correct asset type assigned
Average latency11 minUpload to completed draft inventory
Honest Limitations

What it doesn't handle well

The agent struggles with three edge cases: wills drafted before 1970 that use archaic legal language and non-standard asset terminology; estates with more than eight separate property addresses (the context window becomes a constraint); and documents with significant water damage or degraded scanning quality below a legibility threshold. In these cases the agent escalates to the practitioner rather than producing a draft — the 11% of test cases that didn't complete without intervention were all in these categories.

Tech Stack

What it's built on

Azure AI Foundry — agent orchestration and reasoning
Azure AI Search — semantic retrieval across document corpus
Azure Document Intelligence — OCR and structured extraction from PDFs and handwritten wills
Semantic Kernel — agent planning and tool calling
React — practitioner-facing web interface
Azure AD B2C — authentication and access control
Azure Monitor — logging, alerting, and audit trail
Want this for your firm?

We can build this for legal services, financial services, or any document-heavy workflow.

The same architecture applies to any workflow that involves reading documents, extracting structured data, and producing a structured output. Book a 25-minute call to talk through your use case.

Flux Assistant

Online

This assistant can make mistakes. Any pricing, costings, or financial figures mentioned are illustrative only — a Flux expert will provide accurate numbers for your project.

Hi, I'm the Flux assistant. Ask me anything about our services, pricing, or how we can help your business.