How Rent Report Works

Data sources

Rent Report draws on multiple complementary data sources to build each rental market assessment:

PropertyData API (primary comparables)

The primary source of comparable rental listings is the PropertyData API, which aggregates live and recently-let listings from major UK property portals including Rightmove, Zoopla, and OnTheMarket. For each assessment, we pull comparable properties within a search radius centred on the subject postcode, filtered by bedroom count and property type. Where available, we capture the listing URL so comparable cards in the report link directly to the source listing.

Realyse sandbox API (asking-rent comparables)

Additional asking-rent comparables are sourced from the Realyse API using a bounding-polygon search around the subject address. These results are merged into the comparable pool alongside PropertyData results, with deduplication by address and rent level.

OpenAI web search (AI-gathered comparables)

In parallel with the structured data sources, an AI model with live web search capability runs a targeted search for comparable properties on Rightmove, Zoopla, and OpenRent. These results supplement the structured comparable pool, particularly where direct API coverage is sparse. Per-comparable AI reasoning is checked against known portal URL patterns; non-direct URLs (search results pages) are stripped before the comparables are saved.

EPC Open Data API (floor area enrichment)

Where a subject property or comparable does not have an explicit floor area, the system attempts to enrich it using the Government's EPC Open Data API. Floor area estimates from matching EPC records are used to improve size-based rent adjustments. EPC lookups are cached in the database for 30 days to avoid redundant API calls. The system also captures the full energy profile (walls, roof, floor, windows, heating, lighting) from EPC records and renders it as a colour-coded table in the report.

ONS PRMS (government rental benchmarks)

Local authority-level median rent data from the ONS Private Rental Market Summary is integrated to provide a government benchmark reference point. This is displayed in reports alongside the property-specific estimate to give wider market context.

OpenStreetMap Overpass API (transport and amenity scoring)

Transport and local amenity data is sourced from OpenStreetMap via the Overpass API. The system scores each property by proximity to tube and rail stations, bus stops, supermarkets, schools, parks, and other key amenities. This transport accessibility score feeds into comparable weighting — a property with better transport links is expected to command a higher rent, and comparables are adjusted accordingly.

Ofsted schools data

School quality scoring uses proximity to Ofsted-rated schools. Properties near Outstanding or Good-rated schools receive a positive school score, which is factored into comparable weighting and property adjustments.

HM Land Registry SPARQL (sales context)

Sales price data from the HM Land Registry price paid dataset provides broader market context within the report. This is informational and does not directly affect the rent range calculation.

Comparable selection methodology

The core comparable selection process works as follows:

1. Radius search — comparable listings are pulled within a 1 km radius for London properties and 1.5 km for properties elsewhere in England. A wider radius (up to 3 km) is used as a fallback when the initial search returns insufficient results.
2. Adjacent bedroom search — where a tight comparable set would result in low confidence, comparables for ±1 bedroom are included and adjusted for the bedroom count difference.
3. Type matching — the system applies tolerant property type matching, grouping similar types (e.g. flat and apartment) together while keeping distinct types separate unless data is scarce.
4. IQR outlier rejection — the interquartile range (IQR) method is applied to remove statistical outliers from the comparable pool before computing rent estimates. Comparables with rents above Q3 + 1.5×IQR or below Q1 − 1.5×IQR are excluded from the calibrated range calculation.
5. Recency decay — comparables are weighted using exponential decay based on listing age. A comparable from three months ago carries significantly less weight than a current listing. This ensures the rent range reflects current market conditions rather than stale data.
6. Score-based ranking — remaining comparables are ranked by a composite score incorporating distance from the subject property, recency, type match quality, and EPC floor area availability.

Per-comparable adjustments

Each comparable is adjusted to reflect the specific differences between it and the subject property. Adjustments are applied in the following order, and each step's percentage and absolute change is recorded:

• Recency adjustment — accounts for market movement since the comparable was listed
• Size adjustment — adjusts for floor area differences between the comparable and the subject, using a square-metre rate derived from local market data
• Bedroom count adjustment — where adjacent-bedroom comparables are included, a per-bedroom premium is applied
• Furnished/unfurnished adjustment — a modest premium is applied where the subject is furnished and the comparable is not, or vice versa

Trivial adjustments (less than 1% impact) are suppressed from the displayed waterfall to keep the report concise. The adjusted rent is the figure used in the P25/P50/P75 rent range calculation.

Rent range calculation

After adjustments and outlier rejection, the system computes:

• P25 (lower bound) — the 25th percentile of adjusted comparable rents
• P50 (mid estimate) — the median of adjusted comparable rents
• P75 (upper bound) — the 75th percentile of adjusted comparable rents

Where a property has a service charge (leasehold flats), this is deducted from the gross rent comparables before the range is calculated, producing a net-of-service-charge estimate.

Confidence scoring

Every report includes a confidence score (0–100) that reflects how reliable the rent range estimate is. The score is computed from several factors:

• Comparable count — more comparables produce higher confidence. The system requires a minimum of three comparables to produce a reportable estimate; confidence increases up to around ten comparables.
• Recency of comparables — a pool dominated by recent listings scores higher than one with mostly older data.
• Type match quality — an exact property type match (e.g. flat for a flat) scores higher than an adjusted adjacent-type match.
• IQR spread — a tighter interquartile range (more consistent comparable rents) produces higher confidence than a wide spread.
• Floor area availability — having verified floor area data (from EPC or declared) scores higher than using a bedroom-count proxy.
• Radius tightness — comparables within 500 m score higher than those pulled from the extended radius fallback.

The resulting confidence score is labelled in plain English: High (75+), Medium (50–74), or Low (<50). A low confidence score indicates that the rent range should be treated as indicative, and additional professional advice may be warranted.

Property-level adjustments

Beyond comparable-level adjustments, the report applies property-level adjustments to the median rent estimate for factors specific to the subject property:

• Transport accessibility score relative to comparable benchmark
• School proximity premium/discount based on Ofsted data
• Flood risk adjustment where applicable (GOV.UK Flood Monitoring API)
• Crime rate adjustment (data.police.uk)

Each adjustment shows its percentage impact and absolute change against the median rent.

AI valuation layer

An AI valuation model (OpenAI o4-mini) reviews the comparable evidence and produces a plain-English confidence statement and "What Could Move This" section. The AI is given the full comparable set, property details, transport scores, and government benchmarks. It does not override the quantitative rent range — it provides commentary on the evidence and surfaces factors the rule-based system cannot assess from structured data alone (e.g. recent local development, neighbourhood character, seasonal demand).

An independent market search via Perplexity Sonar runs in parallel and is merged with the OpenAI web search context before being fed to the valuation model, ensuring the AI commentary reflects both data sources' findings.