Cashflow Model

Cashflow Model

Build a multi-year DCF for a real estate asset. Saves the model to .cashflow-models/<asset-key>.json in the workspace.

Step 1: Identify the Asset

Ask for the asset name if not already clear from context. The asset name becomes the memory key (e.g. "Prose Frontier" → key prose-frontier).

Check whether .cashflow-models/<asset-key>.json already exists:

ls .cashflow-models/<asset-key>.json 2>/dev/null && echo "EXISTS" || echo "NOT FOUND"

If it exists, ask:

"I have an existing model for [asset]. Would you like to update it, or use it as-is?"

If "use as-is", stop here. The model is ready for capex-analysis.

Step 2: Establish Asset Profile

Ask three questions (one at a time):

Q1 — Asset type:

"What type of asset is [name]?"

1. Multifamily / Residential
2. Office
3. Industrial
4. Retail
5. Hotel
6. Mixed-use

Q2 — Region:

"Where is the asset located?"

1. United States
2. United Kingdom
3. Germany
4. France
5. Australia
6. Other Europe
7. Asia-Pacific / Other

Q3 — Hold convention:

"What level of analysis do you want?"

1. Unlevered (NOI and IRR only — no debt)
2. Levered (add debt terms)
3. Full waterfall (debt + LP/GP splits)

Store the profile:

{
  "asset_type": "<type>",
  "region": "<region>",
  "currency": "<USD|GBP|EUR|AUD|...>",
  "hold_convention": "<unlevered|levered|levered-with-waterfall>"
}

Step 3: Check for Uploaded Documents

Before eliciting inputs, check whether the user has uploaded an Excel model or PDF. If yes, extract the relevant figures from it:

• For Excel: identify sheets named "Summary", "Cash Flow", "Operating Budget", or similar. Read key fields: unit mix, market rents, vacancy %, opex line items, hold period, exit cap rate.
• For PDF (OM): read the "Financial Analysis" / "Pro Forma" section for stabilized NOI, unit mix, and operating expenses.

Pre-populate inputs from documents and skip elicitation questions for fields already found. Note which fields came from the document vs. were assumed.

Step 4: Elicit Inputs

Ask only for fields not already extracted from documents. Group by template:

Multifamily inputs:

• Unit mix: for each unit type — count, avg SF, current market rent/unit/month
• Going-in occupancy % and loss-to-lease %
• Vacancy %, concessions %, bad debt %
• Other income per unit per year (RUBS, parking, fees, pet — can use $2,000–$2,500 as a starting benchmark for US suburban Class A)
• Opex: payroll $/unit, O&M $/unit, marketing $/unit, G&A $/unit, utilities $/unit, mgmt fee % of EGI, insurance $/unit, taxes $/unit, reserves $/unit
• Hold period (years), exit cap rate, sale costs %

Office / Industrial / Retail inputs:

• Total rentable SF, current occupancy %
• Passing rent $/SF/year, market (ERV) rent $/SF/year
• Lease structure: gross / NNN / net
• Opex $/SF (gross leases), capex $/SF/year, TI at renewal $/SF, LC % of lease value
• Avg lease term, renewal probability
• Hold period (years), exit cap rate, sale costs %

Hotel inputs:

• Number of rooms
• ADR (Average Daily Rate), going-in occupancy %
• Rooms expense %, undistributed expense %, management fee % of total revenue
• FF&E reserve % of total revenue (typically 3–5%)
• Insurance $/room/year, taxes $/room/year
• Other revenue % of rooms revenue (F&B, spa, etc.)
• Hold period (years), exit cap rate, sale costs %

Regional rent growth defaults (suggest to user, let them override):

If the user does not provide growth rates, use the regional defaults above and note them explicitly.

Step 5: Confirm Before Computing

Present a confirmation table of all inputs. Note which came from documents vs. assumed. Ask: "Ready to build the model with these inputs?"

Step 6: Run DCF Engine

Find and execute the DCF engine:

SCRIPT=$(find ~/.claude/plugins -name 'dcf_engine.py' 2>/dev/null | head -1)
if [ -z "$SCRIPT" ]; then
  SCRIPT=$(find ~/audette-skills -name 'dcf_engine.py' 2>/dev/null | head -1)
fi
python3 "$SCRIPT" --inputs '<INPUTS_JSON>'

Replace <INPUTS_JSON> with the full inputs object including asset_name and profile.

Parse the JSON stdout. If the script fails, show the stderr to the user and stop.

Step 7: Handle Levered / Waterfall (if requested)

If hold_convention is levered or levered-with-waterfall, ask for:

• Loan amount or LTC %, interest rate, I/O period (months), amortization (years), closing costs %

Compute levered cashflows:

• Annual debt service = compute from loan terms (I/O: rate × principal; amortizing: standard mortgage formula)
• Levered CF = Unlevered CF − Debt Service
• DSCR Year 1 = NOI Year 1 / Annual Debt Service (flag if < 1.20)

If waterfall:

• Ask for: LP % equity, GP % equity, hurdle tiers (rate, promote %) (Common US structure: 10%/13%/16% hurdles, 20%/30%/40% promotes)
• Compute tiered distributions across the hold period

Present levered IRR and LP/GP IRRs in the summary card.

Step 8: Save Model

Create the .cashflow-models/ directory if it does not exist:

mkdir -p .cashflow-models

Write the model JSON:

{
  "asset_name": "<name>",
  "asset_key": "<key>",
  "profile": { ... },
  "unit_mix": [ ... ],
  "going_in_occupancy": 0.95,
  "loss_to_lease_pct": 0.02,
  "vacancy_pct": 0.05,
  "concessions_pct": 0.01,
  "bad_debt_pct": 0.0015,
  "other_income_per_unit": 2049,
  "opex": { ... },
  "growth": { "rent": [...], "expense": [...] },
  "hold_period_years": 5,
  "exit_cap_rate": 0.05,
  "sale_costs_pct": 0.015,
  "debt": null,
  "waterfall": null,
  "dcf_output": { "annual": [...], "going_in_noi": ..., "stabilized_noi": ..., "exit_value": ..., "unlevered_irr": ... },
  "scenarios": {},
  "last_updated": "<ISO date>"
}

Step 9: Present Summary Card

Show the summary card from the DCF engine output.

Then say:

"Model saved for [asset name]. Use capex-analysis to model the impact of any interventions — solar, unit renovations, EV chargers, operational changes, etc."

Rules

• Never fabricate inputs — ask if missing. The only exception is regional rent growth defaults which must be stated explicitly.
• Always note which inputs came from uploaded documents vs. user-provided vs. defaults.
• Never assert specific green premium percentages — these must be user-provided.
• Always confirm inputs before running the engine.
• Always write the model JSON to disk before presenting the summary card.
• If the engine produces an IRR outside 0–30%, flag it: "The IRR of X% is outside typical range — please review the inputs."
• Hotel assets: remind the user this is an investor-level model, not a hotel management system.