Sketch only.None of these integrations are wired up in the reference implementation. They show the surface where you'd plug in your ERP. The shared mechanism is theprediction_logtable — every field decision (predicted, accepted, overridden) is recorded with provenance, ready to feed an outbound webhook.

NetSuiteERP

Post each predicted invoice to NetSuite as a draft VendorBill via SOAP/REST. Use the prediction_log row's predicted_value as the proposed Account; let the user confirm in the NetSuite UI.

POST https://<account>.suitetalk.api.netsuite.com/services/rest/record/v1/vendorBill
{
  "entity":   { "internalid": "<vendor_id>" },
  "tranDate": "2026-04-25",
  "memo":     "Auto-coded by Aito (CUST-0000) — prediction_log id LOG-abc",
  "expense": [{
    "category": { "internalid": "<gl_4400_id>" },
    "amount":   4220.00
  }]
}

Microsoft Dynamics 365 FinanceERP

Use the OData /VendorInvoiceHeaders endpoint to create the header, /VendorInvoiceLines for each line. Aito's predicted GL goes into MainAccount; predicted approver routes via the workflow assignment.

POST /data/VendorInvoiceHeaders
{
  "DataAreaId": "DAT",
  "InvoiceAccount": "<vendor_id>",
  "InvoiceDate": "2026-04-25",
  "PurchaseOrder": "",
  "InvoiceId": "INV-CUST-0000-001"
}

Procountor (Finnish AP)ERP

Procountor's REST API is well-suited because its account / dimension structure matches the predicted fields here directly. Each prediction maps cleanly to a single API call.

POST https://api.procountor.com/api/purchaseinvoices
{
  "supplier":    { "businessId": "0223640-4" },
  "invoiceDate": "2026-04-25",
  "totalSum":    4220.00,
  "rows": [{
    "accountNumber": "4400",
    "dimensions":    [{ "name": "Cost centre", "value": "CC-210" }],
    "totalSum":      4220.00
  }]
}

Generic outbound webhookGeneric

If your ERP isn't listed, point an HTTPS endpoint at the prediction_log table and consume new rows on a schedule. Aito predictions and human overrides both end up here, so your ERP gets a single feed with provenance.

POST <your_endpoint>
{
  "log_id":          "LOG-abc",
  "customer_id":     "CUST-0000",
  "field":           "gl_code",
  "predicted_value": "4400",
  "user_value":      "4400",
  "source":          "predicted",
  "confidence":      0.91,
  "accepted":        true,
  "timestamp":       1714060800
}