Tier · Light
The underwriting support layer

Light.Your AI senior analyst.

A full underwriting AI that operates like a senior analyst — turning raw OMs, rent rolls, and T-12s into a structured investment view, an Excel model, and an IC-ready memo. Drop the deal. Walk away. Come back to a finished underwrite.

±3%
Answer drift across runs
30mins
Per full underwrite
6
Operating modes
10×
More work than a standard LLM
01
More work, less drift

One question. Five runs. Standard LLMs scatter. Light holds the line.

We took a single underwriting question — the kind every acquisition team asks every day — and ran it through ChatGPT and Claude five times in a row. Then through Light, five times. The LLMs returned a different bid price every run, drifting up to 27% across the set. Light came back within ±2.4%, every time, with the same Excel model attached.

A standard LLM is a fast, fluent guesser. It will write you a different bid at 09:14, 09:16, and 09:18 — with a confident tone every time. None of those is a number you can put in a memo.
The question under test
“312-unit value-add multifamily in Charlotte NC. 65% LTV. 5-year hold. What bid price clears a 15% IRR hurdle?”
Standard LLM— GPT-class or Claude-class, single-pass, prompt-driven
±27%
Drift across 5 runs
$40M$45M$50M$55M$60M
Run 01
$45.1M
Run 02
$57.6M
Run 03
$49.3M
Run 04
$54.8M
Run 05
$50.2M
Same deal · same prompt · five consecutive runs · range $45.1M – $57.6M · spread $12.5M
Light Model— underwriting support layer, mode-driven, model-anchored
0.5–1%
Drift across 5 runs
$40M$45M$50M$55M$60M
Run 01
$50.6M
Run 02
$52.0M
Run 03
$51.3M
Run 04
$51.9M
Run 05
$51.4M
Same deal · same prompt · five consecutive runs · range $50.6M – $52.0M · spread $1.4M
Standard LLM
±27%
Drift across five runs of the same question. Different deal every run. The number you cite in committee is a number that re-rolls itself the moment you close the window.
Light Model
±2.4%
Drift across the same five runs. Reproducible. Auditable. The Excel model behind every bid is the same Excel model on the next run, the next week, and the next month.
You can’t take a number into IC that moves while you read it.
Light returns the same answer because the work is the same work.
02
Tool vs. employee

ChatGPT and Claude are tools you operate. Light is an analyst that operates for you.

A tool gives you a paragraph. You then have to chase down the missing data, verify the comps, build the Excel model yourself, and reconcile the gaps. Light arrives knowing what underwriting actually requires — and goes and does it.

/ Standard LLM · the tool
A confident paragraph. No model. No sources. Your problem.
You prompt. It writes back. Then you sit there for two hours operating it — re-asking, fact-checking, hand-building the Excel, chasing the comp set, repairing the IRR. The tool is fast. The deliverable is half-finished, and the half it gave you may be wrong.
  1. You write a careful prompt
  2. It returns prose with assumptions buried inside
  3. You hand-extract the numbers into Excel
  4. You verify each one against your own data
  5. You re-prompt for the gaps. Re-verify. Re-build.
  6. Two hours later: rough draft only. No memo.
/ Light · the analyst
A finished underwrite. Excel model, comp set, memo — done.
You drop the OM and the rent roll. Light already knows what an underwrite requires. It pulls comps, verifies transactions, builds the model in Excel, runs scenarios, scores assumptions, and returns a complete IC-ready package. You review and decide. Two hours of you, replaced by 30 minutes of it.
  1. You drop the OM, rent roll, T-12 · pick a mode
  2. Light pulls comps & verifies recent transactions
  3. Builds full Excel underwrite (IRR, sensitivities)
  4. Validates assumptions against market data
  5. Generates downside / base / upside scenarios
  6. 30 minutes later: full underwrite + IC memo
03
Six modes

You don’t prompt-engineer Light. You pick a mode.

Light arrives knowing six full underwriting playbooks — one for each side of the deal table. Pick the mode, drop the data, walk away. The mode tells Light what work to do, what comps to pull, what scenarios to model, and what kind of memo to write at the end.

01
/ Acquisition
Deep Underwriting
Full institutional underwrite of a stabilised acquisition. Income reconciliation, expense normalisation, comp-set validation, sensitivity tables, IRR & cash-on-cash projections, and the bid price that clears your hurdle.
Output: Excel model · IRR / sensitivity · bid price · IC memo
02
/ Value-add
Add-Value Mode
Quantifies the rent lift, vacancy reduction, and CapEx stack required to take an underperforming asset to stabilisation. Models scenarios at each lift/spend combination and returns the IRR profile of each strategy.
Output: CapEx stack · rent lift schedule · lease-up curve · post-stab IRR
03
/ Debt origination
Lender Mode
Underwrites the deal from the lender’s side. DSCR sensitivity at current and refi rates, tenant credit, lease structure, recorded covenants, sponsor track record, and the loan-size envelope at each LTV.
Output: credit memo · DSCR sensitivities · LTV envelope · risk flags
04
/ Land & ground-up
Developer Mode
Full residual-land-value underwriting. Tests the parcel against current zoning, recent code amendments, infrastructure access, and entitlement constraints — and returns the by-right yield, RLV per door, and bid price for the land.
Output: by-right yield · RLV/door · entitlement risk · land bid price
05
/ Asset operation
Landlord Mode
Operates the asset like an asset manager. Lease-up schedule, renewal vs new-lease IRR, concession optimisation, expense-ratio targeting, capital plan budgeting, and the operational moves that lift NOI inside 12 months.
Output: 12-mo NOI plan · lease-up schedule · expense plan · CapEx priorities
06
/ Lease decision
Tenant Mode
Underwrites the lease from the tenant’s side. NPV of the lease, real cost per sqft inclusive of TI/escalations/op-ex pass-throughs, sublease optionality, and the right counter-offer terms to bring back to the landlord.
Output: lease NPV · effective $/sqft · counter-offer terms · downside
04
The 30-minute underwrite

Drop the OM. Go do important things. Come back in 30 minutes.

You don’t need to babysit Light. You don’t need to prompt-engineer it. You upload what you have, pick the mode, and get back to your day. When you return, the underwrite is finished, the Excel is built, and the IC memo is sitting in your inbox.

/ 00:00
Drop what you have
Whatever you’ve got — perfect or messy.
  • Offering Memorandum (PDF / DOCX)
  • Rent roll · T-12 / T-3 financials
  • Property photos · tax statements
  • Anything else the broker sent you
~ 30 seconds to upload
/ 00:00 — 00:30
Light does the work
You’re not in the room. Light is.
  • Pulls 8–12 verified comps
  • Validates recent transactions in the corridor
  • Reconciles in-place to market rent & vacancy
  • Builds the full Excel model (IRR, sensitivities)
  • Scenarios: downside / base / upside
  • Scores every assumption against market data
  • Drafts the IC memo with go/no-go signal
~ 30 minutes, end-to-end
/ 00:30
You review & decide
A finished underwrite, ready for committee.
  • Excel model (downloadable, fully editable)
  • Comp-set summary with quality scores
  • Assumption-validation report
  • 3-scenario IRR profile
  • IC memo · presentation-ready
  • Go / No-Go signal with the case for each
~ 15 minutes to review

The team isn’t sitting at their desks operating an LLM all day. They’re sourcing deals, talking to brokers, taking meetings, building relationships. Light handles the underwriting back at the office — and presents finished work when they’re ready to look at it.

05
What you get

The full analyst output stack. Every underwrite, every time.

/ 00
Market & comps analysis
Rents, vacancy, cap rates, liquidity at the sub-market and corridor level. The full evidence base behind every assumption Light plugs into the model.
/ 01
Comps quality scoring
Every comp graded on recency, asset-class fit, sub-market depth, and condition match. Weak comps flagged, defensible comps surfaced. The IC sees the score with the number.
/ 02
Assumption validation
Rent growth, exit cap, vacancy, OpEx inflation — each cross-checked against verified market data. Aggressive assumptions called out. Defensive ones backed up with sources.
/ 03
Downside / base / upside scenarios
Three full IRR profiles, each modelled separately, each with the assumption deltas explicit. The IC sees the spread, not just the midpoint.
/ 04
Development feasibility snapshot
For land & ground-up: by-right yield, RLV per door, basic entitlement check, infrastructure access. Not the institutional depth Heavy delivers — but enough to triage a parcel in 30 minutes.
/ 05
Go / No-Go signal
A clear recommendation with the case for both sides explicit. Confidence score attached. The IC isn’t guessing what Light thinks — the signal is on the cover page.
/ 06
Excel model
A real, downloadable, fully-editable Excel model with the full IRR build. You can take it into committee, share it with a partner, or extend it yourself. No black box.
/ 07
IC-ready memo
A presentation-grade investment memo — thesis, market context, financial summary, risks, recommendation. Walk into IC with the work already done.
06
Plan

Monthly credits. Unlimited seats. Full refund within month one.

1,000 credits / month
enough for ~50 full underwrites.
Light scales with your team, not against it. Add as many seats as you need; credits are pooled. The team that uses underwriting the most gets it the most.
  • Unlimited seats — bring the whole team, no per-seat licensing.
  • Pooled credits — the analyst, the partner, the associate share one bucket.
  • Roll-over included — unused credits carry over and stack with next month’s allowance.
  • Full refund in month one — if you use less than 50% of your first month’s credits, full refund. No questions.
  • All six modes — included on every plan, no upsell tiers.
  • Excel + memo on every run — the deliverables are the product.
07
Property coverage

The asset classes most teams underwrite most of the time.

Residential SFR
Multi-family
Build-to-rent
Office
Retail (all formats)
Industrial & logistics
Land (basic dev. logic)
Mixed-use (standard)
Student housing
Co-living
Senior living
Not included

For these, Heavy is the right tool — full customisation, deep specialisation, institutional Excel modelling.

  • Highly complex or niche assets requiring deep specialisation
  • Advanced hospitality underwriting
  • Specialised infrastructure-like assets
  • Tailored fit to your internal investment strategy
  • Integration with proprietary data & templates
  • Full institutional Excel modelling at deal complexity
08
Where Light fits

Three tiers. Pick the one that fits your operation.

Light replaces the analyst layer for most teams. Heavy replaces the entire underwriting infrastructure for institutional shops. ChatGPT and Claude are general-purpose tools that were never built for capital allocation in the first place.

Capability
ChatGPT / Claude
Light Model
Heavy Model
Answer drift
ChatGPT / Claude±20%
Light Model±3%
Light Model0.5–1%
Excel model
ChatGPT / ClaudeNo
Light ModelStandard
Light ModelInstitutional, custom
Comp set selection
ChatGPT / ClaudeRe-rolls each query
Light ModelStable, scored
Light ModelHyper-local, demand-pocket
Assumption validation
ChatGPT / ClaudePlausible-sounding
Light ModelCross-checked vs market
Light ModelSourced · click-to-verify
Workflow scope
ChatGPT / ClaudeTool you operate
Light ModelAI senior analyst
Light ModelAI investment team
Customisation
ChatGPT / ClaudeNone
Light ModelSix built-in modes
Light ModelFully tailored to you
Throughput
ChatGPT / Claude1 deal at a time
Light Model~50 underwrites / mo
Light Model100s of deals in parallel
Best for
ChatGPT / ClaudeBrainstorming
Light ModelMost acquisition teams
Light ModelInstitutional capital

Most teams start with Light, scale into Heavy when their volume or complexity demands the full custom stack. The two tiers are designed to coexist: Light for the everyday underwriting flow, Heavy for the deals that move the fund.

The bottom line

A senior AI analyst that does the underwriting.
You review and decide.

Stop spending two hours operating ChatGPT for every deal. Stop hand-building Excel models you didn’t want to build in the first place. Drop the OM. Go do important things. Come back to a finished underwrite.