Case Study · AI
An AI model that turns a postcode into a home retrofit plan
Our client helps UK homeowners, landlords and tenants cut the carbon footprint of their homes. AMT built the platform and the AI model behind it: enter a postcode, and the system identifies the property and generates a tailored retrofit plan, then connects owners with assessors, contractors and lenders.
Schedule Technical DiscoveryYour first conversation will be with a senior technical partner, not a salesperson.
Published
At a glance
- Client
- Company offering carbon emission reduction plans for homes
- Market
- United Kingdom
- Users
- Homeowners, landlords, tenants, assessors, contractors, lenders
- What AMT built
- AI prediction model plus the platform around it, including a separate app for assessors and contractors
- Training data
- Data on 20 million properties from UK government sources, plus climate, weather, flood and solar irradiance data from 1950 to 2023, updated monthly
- Model
- Python, Random Forest, multi-output classifier
- Stack
- ReactJS, NestJS, Python, MongoDB
The short version
Homes waste energy through old heating, poor insulation and outdated appliances, and owners rarely know what to fix first. AMT built a system where a user types a postcode and gets a free retrofit plan for that property. The model weighs the building's age, construction, insulation, floors, materials and energy use against decades of local climate data. Premium users get access to assessors and contractors, and lenders can fund the work.
The problem
Homes are a large source of carbon emissions. Inefficient heating, ventilation and air conditioning, poor insulation and outdated appliances raise energy bills, damage the environment and affect health. Owners and tenants need to know which improvements make sense for their particular building, and generic advice does not tell them.
What AMT built
Postcode to property to plan
The user enters a postcode. The system identifies the property and generates a retrofit plan aimed at reducing that building's carbon emissions.
A model trained on 20 million properties
The model considers building age, construction type, wall insulation, floor data, building materials and energy use. This data comes from UK government sources and covers 20 million properties.
Decades of location data
The model also takes in climate, weather, flood and solar irradiance data from 1950 to 2023, updated monthly. That gives each plan an understanding of where the property sits and the natural conditions that affect its emissions.
A marketplace around the plan
- Free plans for homeowners, landlords and tenants
- Premium subscription giving access to assessors and contractors
- A dedicated app that assessors and contractors use for their work
- Lenders who handle loan requests and help onboard new users
- Bidding for subcontractors, based on the work paths each plan identifies
How the model works
| Input | Examples |
|---|---|
| Property data | Building age, construction type, wall insulation, floor data, materials, energy use |
| Location data | Climate, weather, flood risk, solar irradiance |
| Time span | 1950 to 2023, refreshed monthly |
| Method | Random Forest with a multi-output classifier, in Python |
| Output | A retrofit plan for the specific property |
A multi-output classifier fits this problem because one property needs several recommendations at once, not one answer.
Technology stack
| Layer | Technology |
|---|---|
| Front end | ReactJS |
| Back end | NestJS, Python |
| Database | MongoDB |
| AI model | Python, Random Forest, multi-output classifier |
Before you build something similar
- Is the data public, licensed or yours? This model draws on UK government property data. Check usage terms before you train on anything.
- How often does the data change? Climate and weather inputs here refresh monthly. Plan the retraining and data pipeline, not just the first model.
- Do you need one prediction or several? If each record needs a set of recommendations, a multi-output approach is simpler than chaining single models.
- Who acts on the prediction? The plan is only useful because assessors, contractors and lenders can act on it. Design that side too.