PROPERTY INVESTMENT · AGENTIC AI
AI-orchestrated feasibility for a property investment firm: conversational input, a tested Python calculation engine and custom interfaces.
- CLIENT
- Property investment firm
- SECTOR
- Property investment
- BUILT
- Agentic feasibility platform
- RULE
- The AI never does the maths

01 / THE BRIEF
Setting up a first-pass property feasibility takes substantial time. Analysts translate deal details into assumptions, financial calculations and scenario comparisons, while keeping every input transparent enough to challenge.
BEFORE
- ×Deal briefs translated into spreadsheet assumptions by hand
- ×Each feasibility set up manually
- ×Hard to test the investment case quickly
AFTER
- ✓A written brief becomes editable assumptions
- ✓A tested Python engine produces every number
- ✓Scenarios compared side by side
- ✓Investment judgement stays with the team
02 / HOW IT WORKS
- STEP 1Analyst describes the deal
- STEP 2AI captures assumptions
- STEP 3Python engine calculates
- STEP 4Scenarios compared
- STEP 5Team decides
Agent orchestration
LangGraph and LangChain capture the brief, call the right calculation tools and explain outputs in plain English.
Deterministic maths
A separate Python engine for costs, financing, tax assumptions, cashflow and returns, with repeatable results.
Four deal pathways
Development or improvement, followed by sale or hold, with scenarios, sensitivities and exports.
Guardrails
LangSmith tracing and evaluations, plus tests that stop the assistant from giving financial advice.
03 / RESULTS
0
calculations performed by the language model; all maths runs in the Python engine
2,076
automated checks in the test suite
26
deal scenarios tested
< 20 s
to generate a feasibility in one test
The sub-20-second result is a single test example, not an average across live deals. The checks and scenarios run as test gates, with LangSmith used for tracing and evaluation.
04 / BUILT WITH
05 / SECURITY, PRIVACY & GOVERNANCE
- ACCESS
- Hosted on AWS with authenticated access, encryption, and row-level security so each user sees only their own deals.
- PRIVACY
- Deal data treated as commercially confidential, personal information minimised, and client data never used to train models.
- COMPLIANCE
- Tested so the assistant explains outputs but does not give financial product advice or recommendations.
- GOVERNANCE
- AI governance measures informed by ISO/IEC 42001: every assumption visible and editable, runs traceable, judgement with people.
NEXT CASE →
Supplier invoices to Xero bills
No re-keying: coded draft bills in Xero, every one approved by staff.

thirty minutes, free, with the people who built it
Book a free call→