Fix Manual Working-Capital Assumptions with ChatGPT-Powered Forecasts

Use ChatGPT to Derive Segmented DSO/DPO from Raw Transaction Data

Move beyond a single average DSO or DPO by letting ChatGPT analyse line-item transaction data and compute segmented drivers. Export invoice-level AR and AP data (customer, supplier, invoice date, due date, payment date, amount, terms, region, business unit) from your ERP. Use a secure environment (e.g. self-hosted or API-based deployment of GPT technology) to avoid exposing sensitive data to public models.

Once the data is available, prompt ChatGPT to calculate and explain segment-specific DSOs/ DPOs and highlight shifts over recent months. For early tests, you can aggregate data and anonymise names to stay within security boundaries while validating value.

Example prompt for segmented DSO/DPO analysis:
You are a senior finance analyst.
You receive AR and AP invoice data with the following columns:
- document_type (AR/AP)
- customer_or_supplier_segment
- region
- invoice_date
- due_date
- payment_date
- amount
- payment_terms

Tasks:
1) Calculate realised DSO (for AR) and DPO (for AP) by segment and region
   for the last 6 months.
2) Highlight segments where DSO or DPO has changed by > 5 days vs. the
   previous 6-month average.
3) Explain possible drivers for these changes in business language a CFO
   would understand.
4) Provide a table of suggested DSO/DPO assumptions by segment for the
   next 3 months.

Expected outcome: instead of one global number, you get a driver table by segment, with clear explanations that controllers can validate before using them in forecasts.

Generate Driver-Based Cash-Flow Models Directly from Historical Patterns

ChatGPT can help you turn historical data into a structured, driver-based cash-flow forecasting model. Start by describing your current model logic: how revenue flows into receivables, how payment terms translate into collections, how purchase orders turn into payables and inventory changes. Then provide ChatGPT with summarised historical series (e.g. monthly sales, collections, payables, inventory movements) and ask it to propose an explicit driver framework.

Use prompts that force ChatGPT to define formulas and data dependencies rather than just narratives. You can then transfer this logic into Excel, your planning tool or a Python model. Re-run the exercise whenever your business model or behaviour patterns change.

Example prompt for driver-based modelling:
Act as a corporate FP&A expert.
Here is our historical monthly data for the last 24 months (CSV):
- revenue
- AR_opening_balance
- AR_closing_balance
- AP_opening_balance
- AP_closing_balance
- inventory_opening
- inventory_closing
- cash_collections
- cash_payments

Tasks:
1) Infer the implicit DSO, DPO and inventory days by month.
2) Propose a driver-based model that links next month’s cash position to:
   - starting working capital
   - revenue and COGS forecasts
   - DSO/DPO/inventory days assumptions by segment.
3) Express the model as explicit formulas that can be implemented in Excel.
4) Suggest a minimal set of input drivers to maintain manually vs. those
   that can be updated automatically from data.

Expected outcome: a clear driver framework and concrete formulas that bring discipline and consistency to your cash forecasting.

Automate Variance Explanations Between Forecast and Actual Cash

One of the fastest wins is to use ChatGPT for variance analysis between forecasted and actual cash positions. Export your forecasted cash flows and actuals for the period, plus the working-capital assumptions used (DSO, DPO, inventory). Then prompt ChatGPT to reconcile the differences and attribute them to changes in volumes, prices, timing and behaviour (e.g. slower collections in a particular segment).

This turns laborious, manual variance explanations into a semi-automated workflow. Controllers can then refine and validate the narrative instead of starting from a blank page before each monthly performance review.

Example prompt for variance explanations:
You are supporting the monthly cash forecast review.
Inputs:
1) Our original 3-month cash forecast (CSV by week).
2) Actual cash movements over the same period.
3) The working-capital assumptions used in the forecast (DSO, DPO,
   inventory days by segment).

Tasks:
1) Quantify the variance between forecasted and actual cash by driver
   bucket: volume, price/mix, working-capital timing.
2) Identify which segments or regions contributed most to DSO/DPO drift.
3) Draft a concise variance explanation (max. 1 page) for the CFO,
   including charts or tables where helpful (describe them textually for
   now).
4) Recommend updated working-capital assumptions for the next forecast
   cycle based on observed behaviour.

Expected outcome: faster, more consistent variance reports and a closed feedback loop between forecast quality and updated assumptions.

Build Scenario Templates for Cash Shortfall and Liquidity Stress Cases

Use ChatGPT to build reusable scenario templates that test your liquidity under different working-capital conditions. Define a base case (current DSO/DPO/inventory days) and ask ChatGPT to design stress scenarios such as “DSO +10 days for riskier segments”, “major supplier shortens terms by 15 days” or “inventory correction of 20%.” Combine these with assumptions about sales and margin to create a matrix of scenarios.

Embed the resulting templates into your regular planning process: whenever new data is loaded, re-run them and let ChatGPT describe the implications and potential actions (e.g. activate credit lines, accelerate collections, slow capex).

Example prompt for scenario building:
You are designing liquidity scenarios for the next 13 weeks.
Base case inputs:
- Current DSO/DPO/inventory days by segment (table provided)
- Weekly sales and purchase forecast (CSV)
- Opening cash and credit facilities

Tasks:
1) Create 3 stress scenarios:
   a) Customers in high-risk segments pay 10 days later.
   b) Top 5 suppliers shorten payment terms by 15 days.
   c) Sales drop 15% while inventory targets stay constant.
2) For each scenario, estimate weekly cash impact vs. base case.
3) Describe practical mitigation actions for each scenario.
4) Summarise the results in a format suitable for a management meeting
   (bullets + concise narrative).

Expected outcome: a reusable scenario library and clear playbooks for how to react to emerging cash risks.

Use ChatGPT to Document Forecasting Logic and Governance Rules

As your use of AI in cash forecasting matures, documentation becomes critical. Ask ChatGPT to turn technical model descriptions, scattered emails and spreadsheet notes into a single, coherent forecasting handbook. Provide it with the sources (model specifications, policy documents, screenshots, extracts of formulas) and instruct it to produce clear process descriptions, RACI matrices and governance rules.

This not only supports auditability and onboarding of new team members, but also reduces key-person risk: your working-capital forecasting logic no longer lives only in someone’s head or in fragile spreadsheets.

Example prompt for documentation:
You are an internal process documentation specialist.
I will provide you with:
- Excerpts of our cash forecast model and formulas
- Email threads describing how we update DSO/DPO assumptions
- Our treasury policy document

Tasks:
1) Produce a structured process description for our cash forecasting,
   including:
   - data inputs
   - update frequency
   - roles and responsibilities
2) Document how working-capital assumptions (DSO/DPO/inventory) are
   currently set and updated.
3) Suggest a governance improvement plan for integrating AI-generated
   assumptions (approval workflow, thresholds, versioning).
4) Output in a format ready to paste into our internal wiki
   (headings, bullets, clear language).

Expected outcome: a living documentation set that supports control, transparency and continuous improvement of your AI-augmented forecasting process.

Integrate ChatGPT Outputs with Existing ERP/Planning Tools

To make these workflows sustainable, connect ChatGPT outputs to your existing ERP and planning tools. Technically, this often means building a small middleware layer: pull data from ERP, preprocess/anonymise, send to a ChatGPT API with a fixed prompt template, and write the resulting driver tables or narratives back into a database, Excel template or planning system.

Start with one or two high-impact flows: for example, a weekly job that calculates updated DSO/DPO by segment and updates a driver table for your rolling cash forecast. Over time, expand to variance explanations and scenario packs. Reruption’s engineering-heavy approach is designed for exactly this kind of pragmatic integration work – from script-level automations to production-ready services.

Expected outcome across practices: 20–40% reduction in manual forecasting effort, faster variance explanations, and materially improved visibility on short- and mid-term cash positions, with better alignment between finance, sales and procurement.