Supplier Rebate Checker

Turn a supplier rebate agreement into reviewable rules, then reconcile those rules against purchase records and rebates already received. Designed for bookkeepers, finance teams and procurement analysts checking one supplier and completed period at a time.

AI reads the contract; a separate decimal calculator checks the money. Source excerpts and included/excluded ledger rows make the result traceable. The result identifies a difference to review, not a guaranteed payment or an automatically submitted claim.

The input form starts with a labelled fictional example. AI processing and all review/completeness confirmations are off, so running it unchanged returns a needs_review report with no final rebate or difference and no calculation fee. Replace every example field with your own records and review the agreement before confirming it. API requests must supply their own audit fields; missing data is never filled with the example.

Two operations

1. Read an agreement

Select Read contract with GPT-6 Luna. Supply complete contractText, set allowAI to true, and enter your OpenAI API key in the encrypted openaiApiKey field. Your OpenAI project must have access to gpt-6-luna. No other model is substituted.

You can instead supply document as an object with name and base64-encoded content. Accepted files: text PDFs, TXT and Markdown, up to 5 MB and 80 PDF pages. Extraction accepts up to 120,000 characters. Scanned pages need OCR before upload; oversized or unreadable input is rejected rather than silently omitted.

The run writes DRAFT (JSON) and DRAFT_REPORT (HTML). Review every proposed field and source quote against the complete agreement. Missing fields, unverified quotations and unsupported conditions remain unresolved. AI can miss or misunderstand clauses even when its quotations match.

Every returned agreement has confirmed: false. Extraction never calculates money or confirms data completeness.

2. Audit reviewed rules

Select Audit confirmed rules. Copy the reviewed agreement object from the draft into this input, supply the purchase and statement CSVs, and set asOfDate to the review date. Only set agreement.confirmed, ledgerComplete and statementComplete to true once you have checked them.

Do not delete a genuine unresolved condition to force a calculation. Caps, growth targets, discretionary payment, missing schedules and other unsupported obligations require separate review or another calculation method.

The audit supports retrospective and marginal percentage tiers, net goods after discounts excluding tax and freight, linked same-period credits, and final-period half-up rounding to two decimal places. Supported currencies are GBP, EUR and USD; one currency per audit. Tier minima and percentages are decimal strings, with the first minimum at zero. An empty included-product list means all products; an empty excluded-product list means no exclusions. Enter these only when the scope has been established.

Purchase CSV headers:

lineId,invoiceId,date,supplierId,currency,productCode,type,netAmount,originalLineId

Use purchase for positive goods lines and credit for negative returns. Each credit links to an included original purchase through originalLineId. Up to 20,000 rows per CSV, maximum 8 MB of CSV text.

Statement CSV headers:

lineId,supplierId,agreementId,periodStart,periodEnd,currency,rebateAmount,status

Statuses are paid, credited or pending. Supply header-only CSV if the complete statement genuinely contains no payments. Avoid counting the same settlement as both paid and credited. Overlapping periods, pending amounts, incomplete records and unresolved credits prevent a final comparison.

Results

Operation	Key-value store records	Dataset summary
Extract	DRAFT JSON, DRAFT_REPORT HTML	Draft status, model, source fingerprint, missing fields, unresolved count
Audit	OUTPUT JSON, REPORT HTML, ROWS CSV	Review status, spend, expected/received rebates, difference, findings and input fingerprint

Only records for the chosen operation exist. needs_review withholds final expected rebate and difference as null. A failed input or model request produces an ERROR record and a failed run, with no successful dataset item. Correct input or credentials before retrying.

A fictional worked example with £24,000 eligible spend and a 3% retrospective rate calculates £720. With £600 received, the difference to review is £120. This is synthetic demonstration data, not a customer recovery or earnings claim.

JSON monetary values remain decimal strings. CSV exports escape potentially executable spreadsheet cells, including negative amount strings; use JSON for machine processing.

Processing, privacy and costs

The Actor report fee is $0.50 per completed calculation. It applies once to an audit with a final calculated rebate and difference, including a matched result or an amount received above the calculation. AI drafts, blocked audits (needs_review) and failed inputs have no report fee. Set a run spending limit that covers one calculation. OpenAI usage remains separate and is billed to your own key.

Apify receives the input and stores run outputs under its storage and access settings. Keep contracts, purchase records and reports private. AI extraction additionally sends the supplied agreement text to OpenAI; the model request excludes separate purchase and statement inputs. Do not include credentials in contract text.

OpenAI usage is billed to your own key separately from Apify run and any displayed Actor charges. The encrypted key input is read through the Apify SDK and is never deliberately included in results or application logs. Audit mode makes no model requests. Extraction makes one request with a 90-second connection timeout, an 8,192-output-token limit and no automatic retries. A timed-out request may still incur provider usage.

Responses use store: false; this does not mean zero retention. The OpenAI account's applicable data controls still apply. See OpenAI data controls.

This version is a review aid for the supported calculation rules. It does not establish legal entitlement, connect to accounting systems, send supplier claims, perform OCR or guarantee that every contract condition has been identified.