Lead Scorer

McMahon & Nagle · Logistic Regression Model · v65.1

Not loaded Import Model

How the Lead Scorer Works

Estimate → Order Model

This app uses Logistic Regression to predict, for each estimated lead, the probability that it converts to a won order. The model is trained on all leads (warm + cold) with lead source as a feature, plus Source × Postcode interaction cells so the model can learn that “D08 referrals” and “D08 internet” behave differently.

Signals Used (unified model)

Signal	Detail
Lead Source	All sources incl. Previous Customer / Referral / Friend / Family (no flat multiplier)
Postcode	Eircode routing key, e.g. D06, K67
Source × Postcode	Interaction terms for cells with n ≥ 5 — captures e.g. K67 referrals being stronger than K67 internet
Job Value (RRP)	<€2k / €2–5k / €5–10k / €10–20k / €20k+
Discount	Quoted discount bracket (known at quote stage)

Sales Rep is not a feature in the model (kept out by design — avoids scoring identical leads differently per rep). v64.31 replaces the v64.29/30 cold-only + flat-multiplier architecture: warm-source effects are now learned by the LR directly, and Source × Postcode interactions capture channel-specific area patterns.

Score Bands (relative to base rate)

vs Base	Label	Meaning
≥ 1.3× base	High Priority	Well above average — prioritise
≥ 1.0× base	Good Prospect	At or above average — follow up promptly
≥ 0.7× base	Moderate	Below average — standard follow-up
< 0.7× base	Low Priority	Well below average

Base rate = the cold-lead conversion rate in training (typically ~30%). Thresholds are multiples of that, so a "Good Prospect" sits at or above the average, not at an absolute percentage.

What Counts as a Conversion?

Estimate → Order: A lead is converted when its job type becomes an Order. Leads still at Estimate stage are not converted.

If the same enquiry appears as both an Estimate and an Order, the Order takes priority.

Excluded from the Model

Office/admin staff are not scored as reps: Joanne, Vanessa, Lorraine, Jane Rooney, Stephanie Paul, Caroline Flynn, Donna Higginbotham, Maria Redmond, Jeff.

Lead sources excluded: Remake, Source, Service.

Estimate → Order

—

Import model first

Train or import a model to score leads.

Lead Source

Postcode

RRP (€)

Train the model to see performance metrics.

Data Range Impact

Test whether older training data helps or hurts. All tested against 2025.

Train the model to see signal analysis.

Classification

Train the model to see conversion rates by source.

From

Hide not interested

Train or import the model first, then load leads.

From

In progress only Hide job copies (highest RRP showing) Hide Dead (>6mo)

PPI Target: €/inst

Train the model first, then load the pipeline.

Conv From

Conv To

Min price € Min discount % Min €/Inst Min units Max units

Train the model first, then load orders.

From

Live from MBNI · getJobDataset2 (no mirror)

Load the model, set a date range, then Load All Jobs.

From

UTM only (matches webservice) Hide job copies (latest showing) UTM attribution · live from MBNI · all leads by entered date

Set a date range, then Load.

UDF Conversion Analysis

Analyse how UDF field values, lead sources and areas correlate with Est→Order conversion.

⚠ UDF fields are only mandatory at order stage. Blank values on estimates may mean "not yet filled in" rather than "No". Treat results as directional, not definitive.

Profitability Analysis

Average profit and margin breakdown by rep, source, and area. Only orders with line items are included.

Price Elasticity Analysis

Conversion rate and expected profit per lead at each margin level (40–55%). Helps answer: would lowering prices increase overall profit?

Connection Check

Test the database connection and pull a sample job to verify everything is working.

Pipeline Trace

Load the pipeline first, then enter a job or lead number to trace where it drops out of the pipeline stages.

Backfill Data

Sync MBNI job data into the database for a date range. This fetches via SOAP and caches in Supabase.

From

UDF Mapping

MBNI User Defined Fields — field keys, names, types and values.

Key	Name	Type	Values
uf-1	Offer Discussed	dropdown	Yes, No
uf-2	Budget	text	—
uf-3	Other Prices	dropdown	Yes, No
uf-4	Timeline	text	—
uf-5	D.C. Feedback	textarea	—
uf-6	Other Info	textarea	—
uf-7	From Who?	text	—
uf-8	Discuss in Meet	dropdown	Yes, No
uf-9	Decision Makers	dropdown	Yes, No
uf-10	First in	dropdown	Yes, No
uf-11	Referral From	text	—
uf-12	Referring DC	text	—
uf-13	RFT	dropdown	Yes, No
uf-14	RFT NWD	dropdown	Yes, No
uf-15	RFT CAREYS	dropdown	Yes, No
uf-16	RFT DIAMOND	dropdown	Yes, No
uf-17	McMN	dropdown	Yes, Sales, Survey, Processing, Stores, Install
uf-18	RFT ORIGIN	dropdown	Yes, No
uf-19	RFT VERTICAL SLIDER	dropdown	Yes, No
uf-20	RFT DEKKO	dropdown	Yes, No
uf-21	Showroom Visit	dropdown	Yes, No
uf-22	Photographs	dropdown	Yes, No, Not Asked
uf-23	Feedback Call Made	dropdown	Yes, No
uf-24	On Time	dropdown	Yes, No
uf-25	Happy With Appt	dropdown	Yes, No
uf-26	Window Sample	dropdown	Yes, No
uf-27	Heat Lamp	dropdown	Yes, No
uf-28	Photostream	dropdown	Yes, No
uf-29	Next Step	dropdown	No Decision, Follow Up, McMN Order, Other Order
uf-30	Other	textarea	—
uf-31	Grant	dropdown	No, Yes-SEAI, Yes-COUNCIL, Yes-OTHER

Training Data by Rep

Job counts per rep in the trained Est→Order dataset.