GPT Invoice Reconciliation Software with LLM AI
Data Alchemy reconciles supplier invoices against purchase orders and delivery notes (DDT) using one AI engine per document model: Claude (recommended), GPT, Data Alchemy AI, Data Alchemy's proprietary model, or a compatible model through the European providers Scaleway and Regolo.ai. The platform extracts line-level data, performs 3-way matching, validates against your ERP master data and posts the result into SAP, Zucchetti or TeamSystem — 99.8% accuracy with Claude, around 12 seconds for a one-page document (average time with Claude Sonnet), no templates.
From raw invoices to reconciled ERP entries
GPT invoice reconciliation software automates the most time-consuming control in accounts payable: matching every supplier invoice with the corresponding purchase order and delivery note. Data Alchemy pairs each document model with a single AI engine — Claude by Anthropic (the engine we recommend by default), GPT by OpenAI, the proprietary Data Alchemy AI model or a compatible model through the European inference providers Scaleway and Regolo.ai — to read each document, extract structured line items and align them by SKU, quantity and price. The reconciled entry is validated against ERP master data and posted automatically, so your finance team stops re-keying invoices and starts reviewing only genuine exceptions. For a broader overview of the process, see our invoice reconciliation & three-way matching solution.
Everything you need from a GPT invoice reconciliation platform
LLM-powered extraction
A dedicated LLM reads every invoice, purchase order and delivery note. For each document model you choose one AI engine — Claude (recommended), GPT, Data Alchemy AI or a compatible model through the European providers Scaleway and Regolo.ai — and each document is processed by that engine alone. With Claude it reaches 99.8% accuracy, with no templates.
Automatic 3-way matching
Each invoice line is aligned with the matching purchase-order line and delivery-note line by SKU, quantity, unit of measure and price. Discrepancies surface before the entry reaches accounting.
ERP master-data validation
Suppliers, items, price lists and payment terms are checked against SAP, Zucchetti or TeamSystem in real time. Mismatches are flagged for review; clean invoices flow straight through.
Duplicate and fraud detection
The AI compares each incoming invoice with the history of already-posted documents, blocking duplicates, double submissions and tampered totals before payment is approved.
Email-first acquisition
With the Mail service installed at your premises, emails matching a model — invoices, POs, DDTs — are sent from your business mailboxes to the platform and queued for reconciliation, while the rest of the mail stays in the mailbox.
Licences that fit your needs
Choose the Basic, Business or Enterprise licence and add pay-as-you-go credits: 1 credit = 1 document, regardless of the number of pages. From Business up you also get the Mail service, external fields and remote validation against your ERP. Quote on request.
Compare what each licence includes and request a quote on the pricing page.
Four steps from inbox to reconciled ERP entry
1. Capture
Invoices, purchase orders and delivery notes are captured from email, scanners or upload — PDF, XML, FatturaPA and international formats.
2. Extract with the LLM
Data Alchemy uses the single AI engine chosen for each document model — Claude, GPT, Data Alchemy AI or a compatible model through the European providers Scaleway and Regolo.ai — and extracts structured line-level data; with Claude Sonnet or higher, reading takes around 12 seconds for a one-page document and around 18 seconds for two pages (average times).
3. 3-way match
Each invoice line is matched with the corresponding PO and DDT line by SKU, quantity, unit of measure and price.
4. Validate and post
Reconciled data is validated against SAP, Zucchetti or TeamSystem master data and posted into the ERP through the AI agent or the REST API.
How GPT reconciliation compares to traditional OCR workflows
Legacy OCR + manual reconciliation
- ✓Extracts text but not structured line items
- ✓Requires templates per supplier
- ✓Operators re-key data into the ERP
- ✓Matching done in spreadsheets or by hand
- ✓Duplicates and price mismatches surface late
GPT reconciliation by Data Alchemy
- ✓One AI engine per document model: Claude (recommended), GPT, Data Alchemy AI or a compatible model through Scaleway and Regolo.ai
- ✓Zero-template, line-level extraction
- ✓Automatic 3-way matching (invoice-order-DDT)
- ✓Real-time validation against ERP master data
- ✓99.8% accuracy with Claude, ~12 s for a one-page document (Claude Sonnet)
Built for finance, logistics and procurement teams
Accounts payable automation
Reconcile supplier invoices with POs and DDTs, validate against the ERP and post entries with no manual re-keying.
Learn more →LogisticsDelivery-note matching
Capture DDTs at receiving, align them with open POs and feed the 3-way match that closes each procurement cycle.
Learn more →IDP PlatformFull IDP solution
Beyond invoices: orders, contracts, price lists and bank statements processed by the same AI platform.
Learn more →Frequently asked questions about GPT invoice reconciliation
What is GPT invoice reconciliation software?
GPT invoice reconciliation software uses large language models (LLMs) to read supplier invoices, match each line against the corresponding purchase order and delivery note (DDT), and post the reconciled result into the ERP. Data Alchemy pairs each document model with a single AI engine: Claude (Anthropic), the engine we recommend by default; GPT (OpenAI); Data Alchemy AI, Data Alchemy's proprietary model, running on its own servers and GPUs in Italy; or a compatible LLM through the European inference providers Scaleway and Seeweb Regolo.ai, on the customer's own subscription. Your invoice model can run on GPT, while other models use Claude or Data Alchemy AI. With Claude the platform reaches 99.8% extraction accuracy (no accuracy figure is published for the other engines), with no templates; with Claude Sonnet or higher, reading takes around 12 seconds for a one-page document and around 18 seconds for two pages (average times).
How is GPT-based reconciliation different from traditional OCR?
Traditional OCR converts pixels into text but cannot decide whether an invoice line belongs to a specific purchase order or whether quantities match the delivery note. An LLM-powered reconciliation engine understands document semantics: it identifies suppliers, items, units of measure and prices, then performs 3-way matching against the ERP master data. The result is structured, validated and ready to post — no human re-keying required.
What is 3-way matching (invoice-order-DDT)?
3-way matching is the standard accounts-payable control that compares three documents — supplier invoice, purchase order and delivery note (DDT) — before approving payment. Data Alchemy automates this control: the LLM extracts line items from each document, the platform aligns them by SKU, quantity and price, and the system flags discrepancies (price mismatch, missing goods, duplicate invoice) before the entry hits the ERP.
Does the software integrate with SAP, Zucchetti and TeamSystem?
Yes. Reconciled invoices flow directly into SAP, Zucchetti, TeamSystem and other ERPs through the AI agent installed at your premises (a database, CSV/JSON/TXT files or APIs) or with an integration built on Data Alchemy's REST API and webhooks. Master data (suppliers, items, price lists) is read from the ERP in real time so the AI can validate against the system of record on every document.
Which invoice formats does the GPT reconciliation engine handle?
The platform processes PDF, XML, FatturaPA and international electronic invoice formats, as well as scanned paper invoices. It also ingests purchase orders and delivery notes from email (business mailboxes, via the on-premise Mail service), scanners or upload, classifying each document automatically before reconciliation.
How much does GPT invoice reconciliation cost?
Data Alchemy is sold on licences — Basic, Business and Enterprise — plus pay-as-you-go credits for the documents processed (1 credit = 1 document, regardless of the number of pages), and you can start with the free Demo (10 documents). Pricing is available as a quote on request: see the Pricing page for what each licence includes.
See GPT invoice reconciliation in action
Activate the free Demo to try AI reading on 10 documents for 10 days, with 1 model chosen from the predefined ones. To see the 3-way match against your POs and DDTs and calculate your ROI, contact us to plan a call with a demo tailored to your business and your documents.