Lanchester R&DTactical Exploration Lab
01
[01]

Define the prediction

“If you predict everything, you predict nothing.”

Start With: Time until next reactive job.

But now clarify: What is a “reactive job”?

Define it precisely:

  • -> Emergency leak callout
  • -> Water ingress complaint
  • -> Internal damage escalation
  • -> Insurance-triggered event
  • -> Safety hazard (loose slate, falling debris)

If you don't define this clearly?

Your labels will be inconsistent.

Your model will be confused.

02
[02]

Build a canonical data model

Now we expand it properly.

Core Hierarchy (Locked)
ManagingCompany
↳ Location
↳ Building
↳ RoofElement
↳ Sub-Element (optional/powerful)
↳ Project
Observation
Intervention
Outcome
🔥 Critical Data Engine

Building-Level Decay Signals

  • Construction year
  • Major refurb year
  • Structural material
  • Roof pitch angle
  • Building height
  • Surrounding height
  • Exposure (Urban vs Coastal)
  • Pollution index
  • Heritage constraints
Why?

Buildings decay differently based on environment and structure. This is slow-moving but powerful signal.

Roof Element-Level Signals

  • Material type (natural slate, EPDM, lead)
  • Install date & known lifespan
  • Warranty length
  • Repair frequency per element
  • Last repair method (patch vs renewal)
  • Flashing type & thickness
  • Joint type (welded, bonded, mechanical)
Why?

Decay is material-specific. Lead behaves differently than membrane. Patches behave differently than replacements.

03
[03]

Expand the Labeled Event

High-Value Signal Extraction

Your event row becomes:

( Date | RoofElement | ConditionSeverityScore | MoistureExposureIndex | WhatWasSeen | WhatWasDone | MaterialUsed | Cost | AccessType | WeatherAtTime | Outcome | DaysUntilNextEvent )
Massive Signal
A) Weather History
  • + Rainfall in last 30/90/365D
  • + Freeze-thaw cycles
  • + Wind gust exposure
  • + Storm proximity
  • + Temp volatility

Roofs fail under stress. Stress = weather × material age. Without weather? You’re predicting blind.

LiDAR / Physics
B) Water Flow & Drainage
  • + Roof slope gradient
  • + Drainage path length
  • + Low-point accumulation
  • + Gutter fall angle
  • + Downpipe count
  • + Overflow history

Water pooling predicts membrane failure. Blocked drainage predicts leak frequency. Physics > guesswork.

Operational Friction
C) Access Complexity
  • + Scaffold required?
  • + Council notice required?
  • + Neighbor access?
  • + Traffic management?
  • + Rope access only?

Access delays repairs. Delayed repairs increase failure risk. Operational friction drives decay escalation.

Culture Tracking
D) Maintenance Behavior
  • + Avg inspection frequency
  • + Planned vs reactive ratio
  • + Scope expansion freq
  • + % of jobs reworked
  • + Contractor consistency

Maintenance culture affects failure rate. Some portfolios fail because of physics. Some fail because of neglect.

Gradient Scoring
E) Defect Severity
  • + 1 = cosmetic
  • + 3 = early failure
  • + 5 = active leak risk
  • + 7 = temporary fix
  • + 10 = structural compromise

Models predict better when they understand gradient of decay. Binary leak/no-leak is crude.

04
[04]

The Stronger Baseline

Hazard Function Modeling

Your baseline now includes complex multi-variable interactions:

Static Arrays
  • + Roof material age
  • + Environmental exposure
  • + Building geometry
  • + Known lifespan vs age ratio
Dynamic Loads
  • + Weather load index
  • + Maintenance frequency
  • + Defect severity trend
  • + Repair type history
Operational Friction
  • + Access complexity
  • + Contractor patterns
  • + Delay between inspect and intervention

This becomes a hazard function model.

Risk = f(material_age × weather_stress × repair_history × drainage_physics × maintenance_behavior)

Now you’re modeling decay, not just job frequency.

05
[05]

Advanced Multimodal Signals

"Holding for now""Couldn't access""Customer declined""Temporary patch"

Transcript Decay Indicators

These phrases are gold. They predict future failure better than images.

Photo-Derived Time Decay

Decay has direction. You can model acceleration.

Crack length over time Δt
Growth progression Δt
Rust bloom expansion Δt
Membrane discoloration Δt
> Sagging progression
> Deflection change
> Drainage pooling growth
> Flashing lift displacement

3D Temporal Geometry Tracking

If geometry shifts year over year? You’re detecting structural fatigue.

// That’s enterprise-level predictive intelligence.

06
[06]

Evaluate Properly

“Accuracy is a vanity metric.”

"How accurate is the model AUC / F1 loss curve?"

Start asking:

01

Did we reduce emergency callouts per building per year?

02

Did we increase the planned-to-reactive ratio?

03

Did average repair cost per building drop?

04

Did variance in annual maintenance stabilize?

Reduced volatility. Predictability. Budget smoothing.
That’s executive-level value.
07
[07]

Deployment UX Strategy

Now your UI should visualize actionable prescriptions.

App View

Building Risk Index
84/100
  • Age Stress +12%
  • Weather Tmp +8%
  • Defect Accel -
  • Maint Lag -

Graph View

Decay Curve Projection

Predicted risk growth over the next 24 months.

Simulator Tool

Intervention Impact

"If we replace the flashing now, what does the risk curve look like?"

Now it’s not just predictive.
It’s prescriptive.

08
[08]

Industrial Ingestion

“Future-proof the machine.”

+ Weather API Integration+ GIS Environmental Layer+ Auto Material Lifespan+ Warranty Alerts

Every inspection becomes:

Structured EventvLabeled DatavModel Retrained.

Pipeline > Model.

09
[09]

Portfolio Strategy

Now We're Playing Big.

  • []Cluster buildings by decay pattern.
  • []Detect abnormal deterioration outliers.
  • []Predict capital replacement cycles.
  • []Rank contractors by long-term failure recurrence.
  • []Detect material batch failures.

You move from

Maintenance Prediction↳ Asset Lifecycle Optimization↳ Portfolio-level Risk Arbitrage
10
[10]

Finance & Capital Modeling

“Make the CFO love you.”

The Baseline

Expected Reactive Loss Curve

The Model

Predicted Preventative Curve

Then Calculate:

Net Present Value
of early interventions.

Now the CFO listens. Because you’re not talking about leaks.

You're talking about:
Risk-Adjusted Capital Planning
Volatility Reduction
Asset Lifespan Extension

Because Roofs Fail
In Patterns.

Water follows slope.
Flashings fatigue.
Membranes age.
Constraints repeat.
Building typologies behave similarly.

You’re not predicting chaos.

You’re modeling
decay curves.