12 AI Prompts for Credit Managers: Scoring, Dunning, Negotiation

23% of credit professionals already use ChatGPT in their day-to-day work, according to a 2025 NACM poll. For a credit manager, the gain is tangible: a demand letter drafted in 2 minutes, a well-documented customer score, a negotiation plan for a strategic account. Here are 12 ready-to-use prompts designed for a B2B credit manager who wants to industrialize the practice without compromising quality.

Why AI is transforming credit management

The credit manager's role blends three activities that LLMs handle well: information gathering (solvency, customer news), written communication (dunning, demand letters, negotiation emails) and analysis (scoring, aging, limit recommendations). ChatGPT does not execute the dunning, but it produces in seconds what takes a credit manager 20 to 40 minutes: a tailored letter, a file summary, a pricing rationale.

🇫🇷 Note for non-French readers: the prompts below reference French commercial law (articles L441-10 and D441-5 of the Code de commerce), which governs late-payment penalties in B2B transactions in France. Adapt the legal references to your jurisdiction (e.g., Late Payment of Commercial Debts Regulations in the UK, EU Directive 2011/7/EU for other EU member states, UCC § 2-709 and state-level statutes in the US).

Customer risk scoring and assessment

Prompt 1 - Quick scoring grid

Use case: assess a new customer before opening an account.

You are a B2B credit manager in France. Build a 100-point score for the prospect below using 5 weighted criteria: seniority (10), financial strength (30), payment history (20), sector (15), size / footprint (25). Answer in English, currency in euros.

Expected data (CSV or markdown):
- legal_name, SIREN / company registration number
- incorporation_date
- Revenue Y-1, Revenue Y-2 (if available), net income Y-1, equity, financial debt
- headcount
- industry code (NAF or NACE)
- payment history if pre-existing relationship (observed DSO, incidents)
- commercial request: expected volume, requested terms

Minimal example:
legal_name, ACME SAS
SIREN, 123456789
incorporation_date, 2018-03-12
Revenue Y-1, 4200000
net_income Y-1, 180000
equity, 950000
financial_debt, 620000
headcount, 28
industry_code, 6202A
history, none (new prospect)

Deliverables:
1. Global score /100 + detailed score per criterion (markdown table)
2. Recommended credit limit in euros + maximum outstanding advised
3. Vigilance level (low / medium / high) + 2-line rationale
4. 3 precise questions for the sales rep before validation

Guardrails:
- If a financial data point is missing, display "insufficient data" for that criterion rather than inventing.
- If the company is less than 2 years old, automatically cap the proposed limit and flag it.
- Do not compute ratios on negative values without flagging (e.g., negative equity = strong signal).

Prospect data: [PASTE HERE]

Prompt 2 - Annual credit limit review

Use case: review a limit at its annual renewal based on observed behavior.

You are a credit manager. Based on the last 12 months of payment behavior, propose an adjustment to the credit limit (hold, raise, lower, suspend). Answer in English, euros. Compare the customer's DSO to the sector DSO if known, otherwise flag it.

Expected data:
- customer: name, registration number, industry code
- Revenue last 12 months billed
- Customer DSO 12M (days)
- Sector DSO (benchmark, if available)
- Incidents: number + typology (>30-day delay, dispute, write-off, bounced check, garnishment, etc.)
- Current limit + last review date
- Contextual events (fundraising, acquisition, legal proceedings, etc.)

Minimal example:
customer, ACME SAS
SIREN, 123456789
Revenue 12M, 820000
Customer DSO, 58
Sector DSO, 45
incidents, 2 delays > 30d, 1 dispute resolved
current limit, 120000
last review, 2025-04-01
events, none

Deliverables:
1. Recommended decision: hold / raise / lower / suspend + target amount
2. 3 quantified bullet-point rationale
3. Internal email to the account manager (120 words max, factual tone)
4. Suggested next review date

Guardrails:
- Do not recommend raising if customer DSO > sector DSO + 10 days.
- Do not recommend suspending on a single isolated incident without aggravating context.
- If a major contextual event is flagged (insolvency proceedings, central bank rating downgrade), reflect it as priority.

Data: [PASTE HERE]

Prompt 3 - Solvency analysis from a public filing

Use case: analyze a prospect or customer from a public financial filing (Pappers / Infogreffe in France, Companies House in the UK, SEC in the US).

You are an SME credit analyst. Read the financial statements below and produce a solvency analysis. Answer in English, euros.

Expected data (at least the last 2 filings):
- Fixed assets, current assets, inventory, trade receivables, cash
- Equity, long-term / short-term financial debt, trade payables, tax payables
- Revenue, EBITDA (or operating profit), net income
- Headcount, industry code

Minimal example:
Year, 2024, 2023
Revenue, 4200000, 3800000
EBITDA, 310000, 280000
Net_income, 180000, 150000
Equity, 950000, 820000
Financial_debt, 620000, 680000
Cash, 140000, 95000
Trade_receivables, 720000, 610000
Trade_payables, 380000, 320000

Deliverables:
1. Key ratios (markdown table):
   - Financial autonomy (equity / total liabilities)
   - Current ratio (current assets / short-term debt)
   - Interest coverage (EBITDA / financial expenses)
   - Leverage ratio (financial debt / equity)
   - Days sales outstanding (trade receivables × 365 / revenue incl. tax)
   - Days payable outstanding (trade payables × 365 / purchases incl. tax, if available)
2. Altman Z'' score (1995 variant adapted to unlisted non-industrial firms) — only if data permits. Otherwise, flag it.
3. Estimated central-bank rating or credit score (range) with rationale.
4. 3 to 5 red flags (trend deterioration, inconsistency, alert).
5. Final recommendation: open / open with guarantee / refuse. Justify.

Guardrails:
- Do not use the original Altman Z-score (1968): calibrated on listed US industrial firms and unsuitable for SMEs. Use Z'' (1995).
- If financial expenses are missing or zero, do not compute the coverage ratio.
- If equity is negative, raise a red flag and recommend at minimum a guarantee.

Financial statements: [PASTE HERE]

Dunning and collection letters

Prompt 4 - 3-email dunning sequence

Use case: generate a complete B2B amicable dunning sequence.

You are a B2B credit manager in France. Draft 3 dunning emails for an overdue invoice. Answer in English. Each email is 120 words maximum, with a clear subject line and a precise CTA.

Variables to replace:
- [CUSTOMER_TYPE]: e.g. "strategic key account", "recurring SME", "one-off customer"
- [AMOUNT]: in euros
- [INVOICE_ID], [DUE_DATE]
- [CONTACT]: name, role, email
- [HISTORY]: prior payments, delays, known disputes

Sequence structure:
- Email 1 — D+7: courteous tone, friendly reminder, mentions that the EUR 40 fixed indemnity (article D441-5 of the French commercial code) is due but not yet claimed. CTA: confirm payment date.
- Email 2 — D+21: firmer tone, reminder of late-payment interest (rate stipulated in the seller's T&Cs, or ECB MRO rate + 10 percentage points as the statutory fallback) and the EUR 40 fixed indemnity due since day 1 of late payment. CTA: settle within 8 days.
- Email 3 — D+45: formal tone, pre-demand letter. Announces escalation to formal collection at D+55 if unpaid. States the fixed indemnity, cumulated interest, and possible additional indemnity on evidence. CTA: settle within 10 days, otherwise registered demand letter.

For each email: subject + body + standard signature.

Guardrails:
- Never threaten any measure not authorized by law (e.g., listing on a national register, public disclosure of the delay).
- Adapt the tone to [CUSTOMER_TYPE].
- If [HISTORY] flags an ongoing dispute, do not dun mechanically: propose a dispute-clarification email instead.

Context: [PASTE HERE]

Prompt 5 - Demand letter compliant with French law

Use case: produce a legally robust registered demand letter (mise en demeure).

You are a business-law counsel in France. Draft a demand letter for an unpaid B2B commercial claim, compliant with the French Commercial Code and Civil Code. Answer in English (letter will be translated to French before sending).

Variables:
- Creditor: legal name, registration number, registered office, legal representative
- Debtor: legal name, registration number, registered office, addressee
- Invoice(s): number, date, amount incl. tax, due date
- Applicable T&Cs: stipulated late-payment interest rate (if any), otherwise "absent a contractual stipulation"
- Prior dunning actions and dates

Mandatory items:
1. Subject: "Formal demand for payment"
2. Detailed restatement of claims (invoice by invoice, amount, due date, days overdue as of letter date)
3. Total principal due
4. Late-payment interest: rate stipulated in the creditor's T&Cs, or absent such stipulation, the European Central Bank refinancing rate (MRO) plus 10 percentage points. Compute the interest due as of the letter date.
5. Fixed indemnity of EUR 40 per unpaid invoice (article D441-5 of the French commercial code), enforceable from day 1 of late payment (article L441-10 II).
6. Statement that the creditor reserves the right to claim an additional indemnity on evidence if collection costs exceed EUR 40.
7. Mandatory settlement deadline: 8 days from receipt of the letter.
8. Enforcement avenues available in case of non-payment (order for payment, substantive action, summary proceedings as applicable).
9. Closing + signature of the legal representative.

Form:
- Legal register, formal.
- Format suitable for registered mail with acknowledgment of receipt.
- Length: maximum 1 A4 page.

Guardrails:
- If the T&C rate is not provided, explicitly use the wording "absent a contractual stipulation" to justify the ECB + 10 points fallback.
- Never state a flat rate drawn from outdated usage (e.g., "statutory rate" alone): B2B has a specific regime.
- Before sending, recommend that the creditor have the letter reviewed by legal counsel.

Data: [PASTE HERE]

Prompt 6 - Phone dunning call script

Use case: prepare a structured call script for a credit manager or collections officer.

You are a B2B collections lead in France. Draft a dunning call script for an overdue invoice. Answer in English, firm and courteous tone, solution-oriented.

Variables:
- [AMOUNT], [INVOICE_ID], [DUE_DATE], [DAYS_OVERDUE]
- [CONTACT]: name, role
- [RELATIONSHIP_CONTEXT]: history (first delay, recurring delays, ongoing dispute, etc.)
- [CALL_RECORDED]: yes / no (GDPR)

Required structure:
1. Opening (15 seconds): greeting + clear identification (name, company, purpose). If [CALL_RECORDED] = yes, GDPR notice: "Please note this call may be recorded for quality and traceability purposes. You may object."
2. Identity check of the contact (right department, authorized person for finance matters).
3. Factual reminder: invoice number, amount, due date, days overdue.
4. 3 open questions to understand the cause (never accusatory): receipt of invoice? internal approval? cash? dispute?
5. Prepared responses to the 3 most frequent objections:
   - "We never received the invoice" → immediate resend by email during the call + address confirmation.
   - "Wire transfer issue" → ask for transfer reference or proof, propose an alternative (SEPA).
   - "Payment scheduled next week" → ask for a precise date and a written commitment by email the same day.
6. Closing: verbal recap of the commitment (date, amount, means), confirmation email within 15 minutes, callback date if commitment not met.

Additional deliverables:
- Template of the post-call confirmation email (80 words max).
- 3 alert signals that must trigger internal escalation (silence, refusal to commit, new contestation).

Guardrail: never threaten a non-legal measure. No intimidation. Do not discuss the dispute on the merits over the phone — redirect to writing.

Customer context: [PASTE HERE]

Negotiation

Prompt 7 - Payment plan negotiation

Use case: build 3 payment plan options for a distressed customer.

You are a credit manager. Propose 3 payment-plan options to a customer owing [AMOUNT] EUR for an amicable negotiation. Answer in English, euros.

Variables:
- [AMOUNT]: total debt incl. tax
- [CUSTOMER_ANNUAL_REVENUE], [CUSTOMER_EBITDA]
- [SECTOR], [SITUATION]
- [RELATIONSHIP]: duration, historical volume

Expected data:
- Debt breakdown (invoices, dates, amounts)
- Existing guarantees (personal guarantee, pledge, credit insurance)
- Estimated monthly payment capacity

Deliverables:
1. Comparison table of the 3 options (short / medium / long term):
   | Option | Duration | Monthly payment | Interest rate | Required guarantees |
2. Per option: pros / cons from the creditor's standpoint.
3. Acceleration clause, drafted in 2-3 lines ("upon failure to pay any single installment on its due date, the full remaining balance becomes immediately payable, any agreed deferrals becoming void").
4. Email template proposing the payment plan (120 words max, firm and solution-oriented tone).
5. Break point: above what remaining balance at 6 months should the creditor refuse and trigger formal proceedings?

Guardrails:
- Verify the proposed monthly payment does not exceed 15% of the customer's estimated monthly payment capacity.
- Do not propose an unsecured option above [AMOUNT] > EUR 50,000.
- Flag if [CUSTOMER_EBITDA] is negative or missing: in that case, no payment plan without solid guarantee.

Context: [PASTE HERE]

Prompt 8 - Response to a request for extended terms

Use case: respond to a customer's request for extension without damaging the relationship.

You are a credit manager. Draft a professional response to a customer requesting [REQUESTED_DAYS] additional days to pay an invoice of [AMOUNT] EUR. Answer in English, firm tone that preserves the relationship.

Variables:
- [REQUESTED_DAYS]: extension requested by the customer
- [MAX_DAYS_GRANTED]: maximum extension the creditor is willing to grant (suggested default: half the requested days, capped at 30 days)
- [AMOUNT]: amount incl. tax
- [DEPOSIT_PCT]: deposit requested upfront (suggested default: 30%, adjust per customer type and amount)
- [CUSTOMER_TYPE]: strategic / recurring / one-off

Objectives:
1. Do not grant more than [MAX_DAYS_GRANTED] days.
2. Secure a deposit of [DEPOSIT_PCT] upfront.
3. Set a new written payment plan.
4. Preserve the commercial relationship.

Email structure (180 words max):
- Empathetic acknowledgment.
- Reminder of the framework: EUR 40 indemnity and statutory interest due since day 1, not claimed if agreement is reached.
- Proposal: [DEPOSIT_PCT] deposit immediately + balance within [MAX_DAYS_GRANTED] days, formalized by amendment or confirmation email.
- Conditions: strict adherence to the schedule, otherwise full application of penalties and retroactive interest.
- CTA: confirmation by return email within 48h.

Additional deliverable: a "firm" fallback version if the customer does not confirm within 48h (pre-demand letter).

Guardrails:
- If [CUSTOMER_TYPE] = strategic and [AMOUNT] < EUR 10,000, do not trigger a deposit (relational cost > cash benefit).
- If history shows 2 accepted extension requests in the last 12 months, no further extension without a written personal guarantee from the director.

Context: [PASTE HERE]

Prompt 9 - Argument for early payment with cash discount

Use case: convince a customer to pay earlier in exchange for a discount.

You are a credit manager. Draft a 150-word pitch to propose a [DISCOUNT_RATE]% cash discount to a customer in exchange for payment at [EARLY_DAYS] days instead of [STANDARD_DAYS]. Answer in English, euros.

Variables and default values:
- [DISCOUNT_RATE]: default 2%
- [EARLY_DAYS]: default 15 days
- [STANDARD_DAYS]: default 60 days
- [INVOICE_AMOUNT]: to quantify the pitch

Deliverables:
1. Preliminary calculation (show before the pitch):
   - Annualized equivalent discount rate (rate × 365 / (std_days - early_days))
   - Cost for the creditor in euros on [INVOICE_AMOUNT]
   - Verdict: is the discount economically rational for the creditor? (rule of thumb: yes if annualized rate < weighted average cost of capital, typically 8-12%)
2. 150-word pitch for the customer:
   - Cash benefit for the customer (% immediate saving)
   - Relational benefit (delivery priority, future terms)
   - Annual cumulative effect if applied to every invoice
3. 3 possible email phrasings (commercial, factual, short tone).

Guardrails:
- If the annualized rate exceeds 15%, flag that the discount is expensive for the creditor and suggest renegotiating terms rather than offering the discount.
- Do not use wording that implies a permanent commercial rebate (different VAT and accounting treatment).

Context: [PASTE HERE]

Analysis and reporting

Prompt 10 - Monthly aging synthesis

Use case: produce a monthly aging report analysis for the CFO.

You are a credit manager. Analyze the aging report below and produce a monthly synthesis. Answer in English, euros, markdown format.

Expected data (CSV, one line per open invoice):
customer, registration_number, invoice_id, amount, issue_date, due_date, days_overdue, last_contact, dunning_status, account_manager

Minimal example:
customer, SIREN, invoice_id, amount, issue_date, due_date, days_overdue, dunning_status
ACME SAS, 123456789, F2026-118, 32000, 2026-02-15, 2026-03-17, 36, reminder_2
BETA SARL, 987654321, F2026-142, 8500, 2026-03-20, 2026-04-19, 3, none

Deliverables:
1. Summary by aging bucket (markdown table):
   | Bucket | Amount | % of total | # invoices | Implicit DSO |
   Buckets: not due, 1-30d, 31-60d, 61-90d, > 90d
2. Top 10 accounts at stake ranked by (amount × age). For each: name, amount, max bucket, dunning status, recommended action.
3. 3 priorities for the month: which accounts, which actions, estimated cash recoverable.
4. 150-word commentary for the CFO: trend vs prior month (if provided), risks, actions. Factual, quantified tone.
5. 3 key KPIs: computed DSO, % aging > 90 days, number of accounts in litigation.

Guardrails:
- If issue_date or due_date is missing for an invoice, exclude it from DSO and flag.
- Do not recommend litigation if dunning_status = none (missing amicable history).
- If > 30% of the amount sits in the > 90 days bucket, trigger a red alert at the top of the commentary.

Aging report: [PASTE HERE]

Prompt 11 - At-risk customer detection

Use case: identify customers whose payment behavior is deteriorating.

You are a credit-management analyst. Identify customers showing deteriorating payment behavior over the last 12 months from the file below. Answer in English, euros, markdown.

Expected data (CSV, one line per invoice over 12 months):
customer, registration_number, invoice_id, amount, issue_date, due_date, payment_date, actual_days_overdue, dispute (yes/no), related_order_id (if available)

Minimal example:
customer, SIREN, invoice_id, amount, due_date, payment_date, actual_days_overdue, dispute
ACME SAS, 123456789, F2025-M07-118, 32000, 2025-08-17, 2025-08-25, 8, no
ACME SAS, 123456789, F2025-M10-142, 28000, 2025-11-17, 2025-12-14, 27, no
ACME SAS, 123456789, F2026-M01-198, 35000, 2026-02-17, 2026-03-28, 39, yes

Signals to detect (thresholds):
- Progressive increase in average overdue (>= +5 days rolling 3 months)
- Appearance or rise of disputes
- Order volume drop > 20% on last 3 months vs prior 3 months
- Shift from one bucket to a later one across successive invoices
- Shorter time between issuance and dispute

Deliverables:
1. Table of at-risk customers (only those triggering at least one signal):
   | Customer | Registration # | Deterioration score (/10) | Dominant signal | Open amount | Recommended action |
2. Deterioration score: weighted sum of signals (describe the weighting used).
3. For each top-5 customer: 3-line narrative explaining the factual deterioration.
4. Overall recommendation: how many customers to place under enhanced monitoring, how many to suspend commercially.

Guardrails:
- Do not alert on a customer with fewer than 4 invoices in the file (insufficient sample).
- Ignore delays < 3 days (normal administrative noise).
- If payment_date is missing, exclude from the overdue computation and flag.
- Distinguish a structural signal (3-month trend) from a one-off: do not classify a one-off as deterioration.

File: [PASTE HERE]

Prompt 12 - Monthly executive credit management report

Use case: produce the monthly credit management report for the executive committee.

You are the head of credit management. Draft the monthly report for the executive committee. Length: 400 words maximum. Answer in English, euros, factual and decision-oriented tone, no jargon (or define each technical term in a footnote).

Expected data:
1. Monthly KPIs: DSO, CEI (Collection Effectiveness Index), % aging > 90d, bad-debt ratio, provisioned amount. Same KPIs for M-1 for comparison.
2. Top 5 accounts at stake (customer, amount, status, action in progress)
3. Actions completed during the month (5 bullets max)
4. Identified risk events (insolvency proceedings, credit rating downgrade, etc.)
5. Objectives for the next month

Required structure:

## 1. One-sentence summary
One sentence summarizing the month: under control / under watch / executive action required.

## 2. KPIs (markdown table)
| KPI | M | M-1 | Change | Commentary |
| DSO | | | | |
| CEI (*) | | | | |
| % aging > 90d | | | | |
| Bad-debt ratio | | | | |
(*) CEI = (beginning receivables + period billings - ending receivables) / (beginning receivables + billings - ending current receivables). Typical sector target: > 80%.

## 3. Top 5 accounts at stake
Table: customer, amount, age, status, next milestone.

## 4. Actions completed this month
Max 5 bullets, each quantified (e.g., "12 demand letters sent, EUR 340k recovered").

## 5. Identified risks and action plan for M+1
Max 3 risks, each with: estimated cash impact, probability, plan.

## 6. Decisions required from the executive committee
Closed questions (yes/no) with quantified stakes.

Guardrails:
- If a data point is missing, write "to be confirmed" rather than inventing.
- No marketing adjectives ("excellent", "promising", "solid"). Facts and figures only.
- If the executive committee is non-financial, simplify CEI to "collection rate" with a short definition.

Data: [PASTE HERE]

Best practices for prompting in credit management

• Always provide customer context: industry, seniority, history, stakes. Without it, the model produces generic, unconvincing text.
• Set a tone: "firm but courteous", "cordial", "very formal legal". This is the single lever that most changes the output.
• Impose a maximum length: models are verbose. 120 words for an email, 400 for a report.
• Test 2 variants before sending: ChatGPT, Claude and Gemini produce slightly different tones.
• Build an internal library of prompts validated by legal counsel, especially for demand letters.

Limits and watchpoints

• Never paste identifiable customer data into a consumer-grade model. Use ChatGPT Enterprise, Claude Team, or an internal LLM.
• Always review demand letters: an LLM may omit a mandatory item (ECB + 10 points interest, EUR 40 indemnity, settlement deadline).
• Sector hallucinations: some norms or benchmarks cited may be invented. Verify.
• Retention: keep a timestamped record of AI-generated dunning, especially in case of litigation.
• Tonal bias: without guardrails, models sometimes sound too accommodating. Always enforce a firmness level in the prompt.

Conclusion

AI does not replace the credit manager; it removes 30 to 50% of the time spent writing. The real competitive advantage in 2026 will be having an internal library of validated prompts integrated into collections workflows.

That is precisely Cleavr's logic: rather than letting each credit manager copy-paste prompts, our AI agent directly executes the dunning cycle, adjusts tone to the customer profile and escalates to humans when needed. See how Cleavr automates your collections →