For any SaaS finance team, the revenue waterfall is both a critical forecasting tool and a persistent source of friction. The process typically involves pulling data from your CRM, billing system, and revenue recognition platform, then manually stitching together opening backlog, new bookings, contractions, churn, and upgrades to produce a forward-looking view. Spreadsheets grow unwieldy, version control breaks down, and every month-end brings the same anxiety: did we double-count that expansion deal? Is that churn figure net or gross? The pain is compounded when leadership asks for scenario analysis—”What if we lose our top three accounts?”—and you need hours to rebuild the model from scratch.
ChatGPT, when used with structured prompting, transforms this workflow. Instead of spending days reconciling data and formatting outputs, finance teams can instruct the model to generate complete waterfall analyses from raw data exports, apply consistent logic for recurring vs. one-time revenue, and even produce sensitivity tables in minutes. The key is not asking ChatGPT to “do the waterfall” in one vague sentence—it’s providing the model with a clear anatomy of the task: what data to read, what logic to apply, what output format to use, and what success looks like. When you treat ChatGPT as a financial analyst who needs precise instructions, the results are remarkably reliable.
The prompts below are designed to be copy-pasted into ChatGPT (GPT-4 or later) with your actual data files. They follow a structured format that forces the model to ask clarifying questions before executing, reducing the risk of hallucinated numbers or incorrect logic. Use them as templates for your own waterfall analysis, then adapt the placeholders to match your company’s specific revenue categories and data sources.
Why the Anatomy-of-a-Prompt Approach Matters for Finance
Finance professionals are trained to be precise. A revenue waterfall that is off by 2% on a $50M ARR base is a $1M error—material for any board deck. The same standard applies to AI prompts. A vague request like “build me a revenue waterfall” will produce generic, often incorrect output. The structured template below forces you to specify the exact task, the files the model must read, the reference output you want emulated, and the success criteria. This turns ChatGPT from a black box into a controllable analytical tool that follows your firm’s specific revenue recognition policies.
Prompt 1: Full Revenue Waterfall from Raw Exports
First, read these files completely before responding:
[Q2_bookings_export.csv] — Raw bookings data with columns: Deal Name, Close Date, ACV, Term Months, Customer ID, Deal Type (New/Expansion/Contraction/Churn)
[Q2_customer_master.csv] — Customer master with: Customer ID, Name, Segment (Enterprise/SMB), Start Date, Current MRR
[Q2_churn_log.csv] — Churn events with: Customer ID, Churn Date, Reason Code, Lost MRR, Lost ACV
Here is a reference for what I want to achieve:
A standard SaaS revenue waterfall table showing: Opening ARR + New Bookings + Expansion + Contraction – Churn = Closing ARR, with a separate row for gross vs net retention.
Here’s what makes this reference work:
– Each line item is mutually exclusive (no double-counting between expansion and contraction)
– All numbers are annualized (ACV basis, not monthly)
– Churn is gross churn (full account loss), not logo churn
– The waterfall foots exactly: Opening + New + Expansion – Contraction – Churn = Closing
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: A table in markdown format with 8 rows (Opening ARR, New, Expansion, Contraction, Net New, Gross Churn, Net Retention %, Closing ARR) and 3 columns (Q1 Actual, Q2 Forecast, Q2 Actual). Include a notes section below the table explaining any reconciling items.
Recipient’s reaction: The CFO should be able to read this in under 60 seconds and trust the numbers without requesting a re-run.
Does NOT sound like: A generic template with placeholder percentages. Every number must be calculated from the uploaded data files.
Success means: The closing ARR in the Q2 Actual column matches the sum of opening ARR plus all net changes to within $500.
My context file contains my standards, constraints, audience. Read it fully before starting.
DO NOT start executing yet. Ask clarifying questions first.
Give me your execution plan (5 steps max) before you begin.
This prompt works because it forces ChatGPT to pause and ask you clarifying questions—for example, whether to treat annual contracts with monthly billing as ACV or TCV, or how to handle partial-month churns. Once you answer, the model executes the plan step-by-step, and you get a waterfall table that you can paste directly into your board deck. The structured format also reduces the risk of the model inventing data or applying incorrect logic, because you have explicitly defined what “success” looks like in measurable terms.
Prompt 2: Scenario Analysis with Downside Sensitivity
First, read these files completely before responding:
[Q3_forecast_base.csv] — Base case forecast with: Month, New Bookings ACV, Expansion ACV, Contraction ACV, Churn ACV, Opening ARR
[Q2_actual_waterfall.csv] — Actual Q2 waterfall from the previous analysis (used as a variance reference)
[top_10_customers.csv] — Top 10 customers by ACV with: Customer Name, ACV, Contract End Date, Renewal Probability (High/Medium/Low)
Here is a reference for what I want to achieve:
A three-scenario sensitivity table showing Base Case, Downside (20% reduction in new bookings + 15% increase in churn), and Severe Downside (40% reduction in new bookings + 30% increase in churn + loss of top 2 customers with Low renewal probability).
Here’s what makes this reference work:
– Scenarios are additive: each builds on the previous one with clear, documented assumptions
– The impact of losing specific named customers is calculated individually, not as a blanket percentage
– Gross retention and net retention are shown for each scenario
– A variance column shows the delta between Base and each downside scenario
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: A table with 4 sections (Opening ARR, Net New Bookings, Churn Impact, Closing ARR) across 3 scenarios. Below the table, a brief paragraph describing the most sensitive assumption driving the downside risk. Total output should fit on one printed page.
Recipient’s reaction: The board should immediately see the cash flow implications of a downturn and ask for specific mitigation steps, not question the math.
Does NOT sound like: A generic “what-if” analysis that ignores the actual customer names and contract values in my data.
Success means: The Severe Downside closing ARR is within 5% of what our finance team would calculate manually using the same assumptions.
My context file contains my standards, constraints, audience. Read it fully before starting.
DO NOT start executing yet. Ask clarifying questions first.
Give me your execution plan (5 steps max) before you begin.
Scenario analysis is where ChatGPT truly shines for SaaS finance. The model can process the logic of “if we lose Customer A and Customer B, and new bookings drop by 40%, what is the resulting ARR?” in seconds—a task that would take an analyst two to three hours in Excel. The key is that the prompt forces the model to use your actual customer names and contract values, not generic placeholders. When you ask clarifying questions first, you also ensure the model understands whether “loss of top 2 customers” means full ACV loss or gradual contraction over their remaining contract term. This level of specificity is what separates a useful AI analysis from a misleading one.
Practical Tips for Implementing These Prompts
Start with the first prompt and a single quarter of clean data. Do not attempt to merge multiple data sources until you have validated the output against your manual waterfall. After you confirm the numbers foot correctly, move to the scenario analysis prompt. A common mistake is uploading files with inconsistent date formats or missing customer IDs—clean your exports first. Also, be explicit about whether you want the waterfall in ARR or MRR terms; mixing the two is the most frequent error we see in practice. For CFOs and controllers reading this, we recommend having one team member run the ChatGPT analysis in parallel with the manual process for two cycles before relying on it for board reporting.
Finally, remember that these prompts are templates. Adapt the success criteria to match your firm’s specific materiality threshold—if your CFO tolerates a $5K variance instead of $500, adjust accordingly. The structured format ensures consistency, but the business logic must always be yours. Use the clarifying question step to teach ChatGPT your company’s unique revenue recognition policies, such as how you handle prepaid annual contracts or usage-based overage charges. Once the model learns these rules, you can reuse the same prompt structure month after month with fresh data exports, turning a painful manual process into a five-minute analysis.
Published on 29 June 2026 on growwithgpt.com