Resi: lead scoring from proof of concept to production.
Resi helps thousands of UK homeowners design and plan home extensions every month. Fifty One Degrees built the company a predictive lead scoring model, proved it on three years of historical data, and shipped it as a production API, from signed contract to live system in under three months.
Thousands of leads a month, and no way to rank them
Resi is the UK's largest residential architecture platform, guiding homeowners from first idea through design, planning permission and building regulations. Its pipeline starts with a huge top of funnel: thousands of registrations a month, from homeowners ready to start next week to browsers years away from a project. The call team treated them broadly alike, because there was no reliable way to know the difference. Resi wanted its team calling the people who most needed the conversation, and the rest served well by self-serve tools.
A scoring model built on evidence, then an API
A predictive lead scoring model
The model scores each new registration on its likelihood to convert, using 57 features spanning the lead's segment and behaviour, and property-level signals from public UK datasets: EPC records, census demographics, land registry prices and deprivation indices, matched to the lead's address.
Findings Resi could interrogate
The proof of concept concluded with a findings report built on three years of Resi's historical registrations, covering not just model performance but when and how leads convert, so the commercial team could challenge the evidence before committing.
A production scoring API
The model shipped as a containerised Python FastAPI service on Google Cloud. Resi's systems send a lead's address, urgency and acquisition channel; the API returns a predictive score and decile in real time, with the IP transferred to Resi.
Prove it works first, then scale
Modern models, validated properly
The model was built with a state-of-the-art tabular foundation model, cross-validated on held-out data, then distilled into a fast model for production scoring, so accuracy was proven before speed was optimised.
Public data as a force multiplier
Resi's own CRM and behavioural data was enriched with UK public datasets at postcode and property level, lifting the model's address-match rate and giving it signals no single internal dataset contains.
A decision gate, not a sunk cost
The engagement was structured with an explicit gate after the proof of concept: hard evidence in a findings document, then a separate decision to productionise. Resi committed to the API build only after seeing the model perform.
Validated on live data before switchover
Rather than flipping call routing overnight, Resi is recording the model's decile scores on all incoming customers first, so the switch to model-driven routing is made on live evidence.
PoC to production in under three months
The engagement was deliberately staged: a fixed-price proof of concept with a decision gate, then a fixed-price productionisation phase. Contract to live production API took under three months.
Connect to Resi's CRM, warehouse and behavioural data, and enrich it with public UK property datasets.
Build and cross-validate the scoring model on three years of historical registrations.
Deliver the findings: model performance, conversion patterns and targeting analysis, as evidence for the go/no-go decision.
Ship the model as a containerised FastAPI service on Google Cloud, integrated with Resi's systems.
Score all incoming leads in production while Resi validates deciles against live outcomes before routing decisions switch over.
What the engagement delivered
From call routing to marketing
Once the live validation window completes, the model moves from scoring to steering: routing the right leads to the call team and the rest to self-serve, analysing lead allocation across Resi's brands, and putting the same scores to work in marketing segmentation and targeting.
Their approach, prove it works first, then scale, is exactly the right fit for us.Joe Whitworth, CEO, Resi Design
What did Fifty One Degrees build for Resi?
A predictive lead scoring model and production API. The model scores every new registration on its likelihood to convert using 57 features from Resi's CRM, behavioural data and public UK property datasets, and runs as a containerised FastAPI service on Google Cloud, returning a score and decile in real time.
How fast was the delivery?
Under three months from signed contract to live production API: the proof of concept and findings landed within the first month, and the productionised API was confirmed live in production roughly ten weeks after kickoff.
How did Resi know the model worked before paying to productionise it?
The engagement had an explicit decision gate. The proof of concept was built on three years of Resi's historical registrations and cross-validated on held-out data, with the evidence presented in a findings document the commercial team could interrogate before committing to the API build.
Can Fifty One Degrees build lead scoring for my business?
Yes. Fifty One Degrees builds predictive models on a business's own data, proof of concept first, and ships them as production systems you own. Book a Discovery Call and we will assess whether your data can support a scoring model.
Want to know which leads are worth the call?
Book a 30-minute Discovery Call and we'll assess what a predictive model could do with the data your business already holds.