Every finance team knows the monthly dread of chasing overdue invoices. The accounts receivable process is often a manual slog through spreadsheets, email threads, and aging reports that are already outdated by the time they’re reviewed. Controllers spend hours segmenting receivables by customer, drafting collection emails, and reconciling partial payments. The friction is real: you have to balance firmness with customer relationships, prioritize which accounts to chase, and ensure every follow-up is compliant with your company’s credit policy. Meanwhile, aging reports—the lifeblood of AR analysis—are frequently static PDFs that don’t surface the insights you actually need, like which customers are trending toward delinquency or where payment patterns have shifted.
ChatGPT changes this entirely. By acting as a structured reasoning engine, it can ingest your raw AR data, apply your collection rules, and produce both prioritized action lists and narrative aging analyses in seconds. Instead of manually sorting a 500-row spreadsheet, you can ask ChatGPT to segment receivables by risk tier, generate personalized collection emails that reference specific invoice numbers and due dates, and even produce a written aging summary that highlights anomalies. The key is how you prompt it—not as a magic box, but as a methodical assistant that follows your logic step by step.
This post provides two production-ready prompts. The first automates your entire collection workflow from raw data to drafted emails. The second transforms a standard aging report into a narrative analysis that your CFO can actually use in a board meeting. Both follow the structured “Anatomy of a Prompt” format, which forces ChatGPT to reason through your constraints before executing—eliminating the vague, generic outputs that most users get.
Why Most AR Automation Attempts Fail
The typical approach to automating AR with AI is to dump a CSV into ChatGPT and ask “write collection emails for overdue accounts.” The result is a mess: generic language, no prioritization, and emails that ignore your company’s specific payment terms or discount policies. The problem isn’t ChatGPT—it’s the prompt. Without defining success criteria, reference patterns, and constraints, the model defaults to its training data, which is not your data. The two prompts below solve this by embedding your business logic directly into the instruction set.
Prompt 1: Automated Collections Workflow
This prompt ingests your raw AR export, applies your segmentation rules, and outputs a structured collection plan with drafted emails. It forces ChatGPT to first understand your data structure, then execute a multi-step reasoning process before generating anything.
First, read these files completely before responding:
[ar_export_2026Q2.csv] — Raw accounts receivable data with columns: customer_name, invoice_number, invoice_date, due_date, amount_due, amount_outstanding, days_overdue, customer_tier (A, B, C), last_payment_date.
[collections_policy.md] — Company policy document defining escalation rules: Tier A customers get one reminder at 15 days overdue, then a manager call at 30 days; Tier B gets automated reminders at 10 and 20 days, then a formal notice at 45 days; Tier C gets daily automated reminders starting at 5 days overdue.
Here is a reference for what I want to achieve:
I have attached a sample output file called [sample_collections_plan.md] which shows a table with columns: priority_rank, customer_name, invoice_ref, days_overdue, risk_flag, recommended_action, email_draft. The email drafts use a professional but firm tone, reference specific invoice numbers, and include a payment link placeholder.
Here’s what makes this reference work:
– Each customer is ranked by a composite score of days overdue and customer tier (higher tier = more leniency)
– Email drafts are personalized: they mention the customer’s name, the exact invoice, and the number of days overdue
– Risk flags are binary: “high” if days_overdue > 30 AND tier is B or C; “medium” if days_overdue > 15; “low” otherwise
– The tone shifts slightly by tier: Tier A emails are consultative, Tier C emails are direct and urgent
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: A table with 10 rows (top 10 priority accounts) plus one drafted email per row. Total output should be about 800 words.
Recipient’s reaction: The collections manager should be able to copy the email drafts directly into Outlook with minimal editing. The priority ranking must match what they would manually assign.
Does NOT sound like: Generic phrases like “we noticed your account” or “friendly reminder.” Avoid any language that sounds like a form letter.
Success means: The team saves 4 hours per week on collections email drafting and sees a 10% reduction in 30+ day overdue balances within two weeks of using the output.
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.
Bridging to Aging Report Analysis
Once collections are automated, the next logical step is making your aging reports actionable. Most aging reports are just tables—they show buckets but not the story behind them. The second prompt transforms your raw aging data into a narrative analysis that identifies trends, anomalies, and recommended actions. This is particularly useful for CFO presentations or weekly AR review meetings where you need to explain why certain buckets are growing.
Prompt 2: Narrative Aging Report Generator
This prompt takes a standard aging report (typically a pivot table or SQL export) and produces a written analysis with specific insights. It forces ChatGPT to calculate month-over-month changes, flag accounts that have shifted buckets, and recommend specific follow-ups.
First, read these files completely before responding:
[aging_report_june2026.csv] — Aging report with columns: customer_name, current_balance, 1-30_days, 31-60_days, 61-90_days, 90+_days, total_outstanding, previous_month_total. Each row is one customer.
[aging_report_may2026.csv] — Same structure for the prior month, used for comparison.
[credit_policy_summary.md] — One-page document defining acceptable aging thresholds: total AR over 90 days should not exceed 5% of total AR; any single customer over 60 days for more than $10k requires a credit hold review.
Here is a reference for what I want to achieve:
I have attached [sample_aging_narrative.md] which is a 500-word memo format with sections: Executive Summary, Key Changes vs Last Month, High-Risk Accounts, and Recommended Actions. The memo uses percentages and dollar amounts, and flags any account where the 60+ day balance grew by more than 20% month-over-month.
Here’s what makes this reference work:
– The executive summary states the total AR, the percentage in each bucket, and the month-over-month change in the 90+ bucket
– “Key Changes” highlights the top 3 accounts that moved between buckets (e.g., “Customer X had $45k move from 31-60 to 61-90”)
– High-risk accounts are those exceeding the credit policy thresholds, listed with specific dollar amounts and days
– Recommendations are action-oriented: “Place credit hold on Customer Y” or “Schedule call with Customer Z for payment plan”
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: A memo format, approximately 600 words, with clear section headings and bullet points for action items.
Recipient’s reaction: The CFO should be able to forward this memo directly to the board with zero edits. It must be professional, data-driven, and free of hedging language.
Does NOT sound like: A robot reading numbers. Avoid sentences like “the data shows that there is a change.” Instead, use direct statements: “The 90+ bucket grew by 12% this month, driven by Customer A and Customer B.”
Success means: The CFO approves this format for weekly AR reporting, replacing the current 3-page spreadsheet that no one reads.
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.
Practical Tips for Implementation
To get the most out of these prompts, start with a clean data export. ChatGPT works best with structured CSV data that has consistent column names and no merged cells. Before running the collections prompt, ensure your AR export includes the customer tier column—this is the linchpin for prioritization. If you don’t have a tier system, create a simple proxy based on total lifetime revenue or payment history.
For the aging report prompt, always include the prior month’s data. The magic happens in the comparison—without it, the narrative is just a static description. Also, note that both prompts ask ChatGPT to present an execution plan before generating output. This is critical: it forces the model to surface its reasoning and lets you catch errors before it produces final text. If the plan looks wrong, you can correct it before the model wastes tokens on bad output.
Try running the collections prompt this week with your current AR data. Even if you don’t use the emails verbatim, the priority ranking alone will likely surface accounts your team has been overlooking. For the aging report, run it as a test with two months of data and compare the narrative to what your senior analyst would write. You’ll likely find that ChatGPT catches trends that humans miss—simply because it can process every row without fatigue.
Published on 3 July 2026 on growwithgpt.com