Demonstration · Illustrative data · BI Consulting Services

Electra Capital × BI Consulting Services

Collateral rent-roll surveillance — from 40 sponsor rent rolls in five formats to one portfolio view across the multifamily loan book. Every panel below recalculates when you change region or period.

Prepared for BenjaminChief Investment Officer · September 2026

Region

Period

Signature view · the pipeline this board sits on

From 40 rent rolls to one portfolio view

Each month every sponsor sends a rent roll in whatever their property-management system exports — Yardi, RealPage, Entrata, hand-built Excel, or a scanned PDF. A standardisation step (AI model for layout recognition + Python rules for unit/lease fields) lands them in one table; anything it cannot map with confidence goes to a review queue instead of the board.

Trend ·

Occupancy & in-place rent per unit

Physical occupancy (left axis) against weighted average in-place rent (right axis). Dashed line is the underwritten occupancy assumption for the selected properties.

OccupancyIn-place rent / unitUnderwritten occupancy

What this board answers

Three questions from the discovery calls

Ranking · NOI variance to underwriting

Properties furthest from underwritten NOI

Trailing actual NOI vs the NOI underwritten at closing, worst first. Centre line = on plan; grey code = state.

Rent-roll data quality · the standardisation step

Files received, auto-formatted, and sent to review

One file expected per property per month. Blue = mapped automatically with confidence ≥ 0.95; red = exceptions a person had to touch.

Unit mix & lease-expiry ladder

Where the roll-over risk sits in the next 12 months

Top: units by floor plan. Bottom: leases expiring by month from the standardised lease-end field — the number sponsors most often get wrong in their own exports.

Ask the data · private AI layer (mock)

A question typed by an asset manager

Also from BI Consulting Services

Private AI on your own server

The "ask the data" panel above is built to run on a GPU server inside your own network: an open-weight model over your own documents, with every answer citing the file it came from. No per-seat licence, works offline, and nothing leaves the building.

How it sits in your environment

One appliance-class server on your rack or in a secured closet. Your files are indexed where they already live; the model runs locally; answers come back with the source page attached.

What that means in practice

1
Your data only, on your hardwareAn open model over your own documents and exports. No third-party AI processor, no data-transfer agreement to review.
2
Every answer cites its sourceReplies point to the file and page they were drawn from, so people check the answer instead of trusting it blind.
3
No per-seat licenceThe whole team can ask questions. Cost is the server and our build, not a monthly number that grows with headcount.
4
Works offlineIf the internet connection goes down, the answers still come. Nothing depends on a vendor's uptime or pricing changes.