For most finance leaders, working capital is a silent leak. You review the balance sheet at month-end, see days sales outstanding (DSO) creeping up, days payable outstanding (DPO) shrinking, and inventory days stretching beyond policy. The reaction is always the same: a flurry of emails, a spreadsheet update, and a promise to “look into it.” But the underlying friction remains—manual data extraction, siloed ERP exports, and a lack of daily visibility. The cost of that friction is real: higher borrowing costs, missed supplier discounts, and cash tied up in slow-moving stock.
The problem is not a lack of intelligence; it is a lack of operational cadence. Your team spends 80% of their time pulling reports and reconciling figures, leaving only 20% for actual analysis. ChatGPT changes that dynamic. Instead of waiting for a month-end snapshot, you can now query a conversational model that reads your latest AR aging, inventory turnover, and payables schedule. It identifies the top ten delinquent accounts, flags SKUs with negative contribution margin, and suggests a dynamic discounting offer to your largest supplier—all in minutes, not days.
This is not about replacing your ERP or your treasury management system. It is about building a decision-support layer on top of your existing data. By feeding ChatGPT structured extracts and holding it to a rigorous prompt framework, you turn a general-purpose language model into a working capital analyst that never sleeps. The prompts below show you exactly how to structure that interaction, moving from vague requests to precise, actionable outputs.
Why Prompt Anatomy Matters for Finance
A generic prompt like “analyze my working capital” yields generic advice. A CFO needs a specific output: a prioritized action list, a negotiation script, or a variance explanation tied to a specific business unit. The difference lies in how you frame the task, what context you provide, and how you constrain the response. The following pre-box template is designed to force clarity. It asks the model to read your files, understand your reference, and ask clarifying questions before executing. This prevents the most common failure mode—the model guessing your assumptions.
When you use this structure, you are essentially briefing a junior analyst. You give them the files, the reference output, the success criteria, and the constraints. You also require a plan before execution. This reduces hallucination and ensures the final output is usable in a board meeting or a bank covenant review, not just a theoretical exercise.
First, read these files completely before responding:
[ar_aging_q2_2026.csv] — Contains customer names, invoice numbers, invoice dates, due dates, and outstanding balances by aging bucket (current, 30, 60, 90+ days).
[customer_credit_terms.md] — Lists each customer’s agreed payment terms, credit limit, and historical payment behavior (prompt, slow, or default risk).
[collection_notes_q2.md] — Notes from the collections team on disputes, promises to pay, and any ongoing negotiations.
Here is a reference for what I want to achieve:
[Upload a sample cash release report from a prior quarter as markdown, or describe the structure: a ranked list of customers with specific action items, expected cash impact, and risk level.]
Here’s what makes this reference work:
– Each customer row has a clear “next action” (call, send statement, offer discount, escalate to factoring).
– The expected cash impact is quantified in USD and tied to a specific aging bucket.
– The risk level is color-coded (green, yellow, red) based on payment history and dispute status.
– The report ends with a summary of total addressable cash and a recommended weekly collection cadence.
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: A prioritized action plan, maximum 2 pages, table format.
Recipient’s reaction: The CFO should be able to assign tasks to collectors immediately without further analysis.
Does NOT sound like: A generic advice article or a list of accounting definitions.
Success means: I can reduce DSO from 52 days to 47 days by identifying the specific invoices that are collectible within 30 days and the ones that need a discount offer.
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.
The prompt above forces the model to act as a collections strategist, not a data dump. Notice the explicit success metric (DSO reduction) and the requirement for a table format. The clarifying questions step is critical—it allows the model to ask about FX rates, disputed amounts, or whether certain customers are strategic exceptions. Without that, you risk a generic list that ignores your commercial reality.
Once you have the AR plan, the next logical step is inventory. Excess stock is often the largest hidden cash user. The challenge is distinguishing between strategic buffer stock and obsolete dead weight. A simple inventory turnover ratio hides the nuance. You need a line-item analysis that considers lead times, demand variability, and product lifecycle stage. The second prompt below tackles this directly, forcing the model to segment the SKU base and recommend disposition actions.
Moving from Receivables to Inventory
Inventory optimization is politically sensitive because it touches sales forecasts and operations. A CFO cannot simply cut stock without risking stockouts. The prompt structure below addresses this by requiring the model to consider service level targets. It asks for a risk-adjusted recommendation, not just a list of slow movers. This is where the “clarifying questions” step becomes a safeguard. The model will ask about your target fill rate, your supplier lead times, and whether you have any contractual minimum order quantities.
This approach turns ChatGPT into a cross-functional facilitator. It does not just flag excess inventory; it proposes a phased reduction plan that aligns with the sales forecast. That means you can present the output to the VP of Sales without triggering an argument. The prompt includes a reference file that shows how a successful inventory optimization was structured in a similar company, giving the model a template to follow.
First, read these files completely before responding:
[inventory_snapshot_july_2026.csv] — Contains SKU, description, current on-hand quantity, unit cost, warehouse location, last movement date, and 12-month sales history.
[demand_forecast_q3_q4_2026.csv] — Contains SKU, forecasted monthly demand, forecast confidence interval, and lead time from supplier.
[supplier_contracts.md] — Contains MOQs, lead times, and any flexible ordering clauses or consignment stock options.
Here is a reference for what I want to achieve:
[Upload a past inventory write-off analysis or a best-practice SKU rationalization report as markdown, or describe the structure: a table with SKU, current stock value, days of cover, disposition recommendation (keep, reduce, liquidate, write-off), and cash release potential.]
Here’s what makes this reference work:
– Each SKU is classified by a “days of cover” threshold (e.g., > 90 days is excess unless it is a slow-moving spare part).
– The recommendation distinguishes between “reduce to safety stock” and “liquidate entirely.”
– The cash release is calculated as (current on-hand – target on-hand) x unit cost.
– There is a commentary column for exceptions (e.g., “mandatory spare part for service contract”).
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: A prioritized disposition table with a summary dashboard, maximum 3 pages.
Recipient’s reaction: The operations team should see a clear list of SKUs to stop ordering and a timeline for liquidation sales.
Does NOT sound like: A vague suggestion to “improve forecasting” or “reduce inventory.”
Success means: I can present a plan to the board that shows a $2M cash release with a risk matrix showing which SKUs have a 10% stockout risk.
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.
Your next step after running these prompts is to validate the output against your ERP data. ChatGPT is excellent at pattern recognition and prioritization, but it is not a substitute for your system of record. Use the generated plan as a starting point for your weekly cash call. Assign owners to each action item and track progress in your existing workflow tool. The real value appears when you run this prompt weekly—the model will learn your exceptions and become more accurate over time.
One practical tip: do not paste your entire ERP export into the prompt. Instead, pre-aggregate the data in a pivot table or a summary CSV. This reduces token usage, speeds up the response, and minimizes the chance of the model getting lost in transactional noise. Also, keep your context file updated with your latest credit policy and inventory targets. That file is your guardrail—it prevents the model from suggesting actions that violate your risk appetite.
Finally, remember that working capital optimization is a continuous process, not a one-time project. The prompts above are designed to be reusable. Save them as templates, change the date ranges, and re-run them every Monday morning. Over time, you will build a historical record of recommendations and outcomes. That data will allow you to measure the accuracy of the model’s suggestions and refine your prompts further. The goal is not to automate the CFO’s judgment, but to automate the preparation work so that judgment can be applied where it matters most.
Published on 29 August 2026 on growwithgpt.com