The Problem: Static Projections and Pre-Launch Blind Spots#
Like many founders preparing for launch, I relied on a spreadsheet to map our runway. But as we got closer to bringing our products to market, the static formulas in our old 6-year projection model started showing their limits.
It assumed flat marketing budgets, flat customer acquisition costs (CAC), and had no seasonality—which is a major blind spot for a direct-to-consumer (DTC) e-commerce brand that lives and dies by Q4 spikes and summer slumps. More importantly, it had a classic repeat purchase compounding error: it calculated customer retention across our entire historical pool indefinitely, ignoring the fact that maternity wear has a natural utility window of about nine months.
For a pre-launch startup, these static models hide the real risks. They mask the pre-launch cash burn, ignore supplier Minimum Order Quantity (MOQ) capital traps, and assume sales start from day one without list building.
The Python Experiment: Streamlit vs. Excel Transparency#
My first instinct was to move away from Excel altogether. I wrote a python-first financial engine and built a Streamlit web dashboard to run real-time sensitivity dials.
graph TD
A[assumptions.yaml
Master Inputs] --> B(Python engine/calculations.py
Streamlit Dashboard)
A --> C(openpyxl compiler/generate_maternity_model_6yr.py)
C --> D[Maternity_Wear_Financial_Model_6Yr.xlsx]
D --> E[Cover Page & Control Panel]
D --> F[Executive Charts Dashboard]
D --> G[Dynamic 3-Statement Model]
The Python dashboard was great for sliding levers and seeing immediate graphs, but I quickly hit a roadblock when it came to sharing the model.
When you sit down with investors, banks, or manufacturing partners during a pre-launch raise, they don’t want to run a local Streamlit server or trust a black-box Python script. They want a spreadsheet. Excel provides a level of row-by-row transparency and auditable cells that python dataframes simply cannot replicate for external stakeholders.
The compromise? Use Python as the compiler to build a dynamic, formulas-only Excel model.
Rebuilding the Model with AI Agents in Hours, Not Weeks#
To achieve this, I used Antigravity AI agents to pair-program the rebuild. What would ordinarily take weeks of manual spreadsheet writing, testing, and debugging took us a single afternoon.
I configured a master assumptions.yaml file with our unit economics, and had the agents write a compiler script using openpyxl. The script generates the entire workbook programmatically, injecting clean, dynamic formulas instead of hardcoded numbers:
- Pre-Launch Runway Toggle: Modeled a configurable pre-launch phase (e.g. 3 months of zero revenue but active fixed costs) to accurately project startup burn before the first product ships.
- Rolling Cohort Pools: Repeat purchases are linked to a rolling 9-month pregnancy window (
=SUM(start_col:prev_col) * repeat_rate), matching the maternity lifecycle. - Supplier MOQ Modeling: Replaced smooth inventory assumptions with stepped bulk purchases, exposing the cash flow dips when buying large fabric runs from suppliers.
- Dynamic Seasonality & Inflation: Clicks are seasonal based on monthly indices (
=INDEX(Seasonality, MONTH(Date))), and CPC inflates over time to represent ad auction fatigue. - Stepped Headcount & COGS Discounts: Salaries step up by +£1,500/month for every block of 500 monthly orders above our baseline, and fabric material costs automatically trigger a 5% bulk discount when monthly sales exceed 1,000 units.
The final output is a dynamically calculating workbook featuring an elegant cover control sheet, an executive charts dashboard tab (utilising native Excel charts), and integrated Three Statements (P&L, Cash Flow, and Balance Sheet) with corrected sign and debt treatments.
Key Takeaways for Pre-Launch Startups#
- Model the Cash Valley: For a pre-launch startup, the primary purpose of a financial model is to find the Cash Valley—the lowest point of your cash balance before the business becomes cash-flow positive. This number dictates exactly how much seed capital you need to raise to launch safely.
- Pivot Assumptions Instantly: In the pre-launch phase, supplier terms, shipping rates, and packaging costs change weekly. By compiling the Excel sheet programmatically from a config file, I can rewrite the entire model’s logic (like shifting from 60 inventory days to bulk MOQ batches) in seconds, rather than manually editing hundreds of Excel formulas.
- Excel Remains the Language of Funding: Investors and banks run on spreadsheets. A clean, fully formulaic Excel file builds trust because it allows stakeholders to inspect your math cell-by-cell.
