Utility bill data extraction

Three teams, three reasons to read a utility bill

• Onboarding and KYC analysts want a name, a service address and a date, because bills are common proof of address. Acceptance windows vary: one UK council asks for a bill dated within the last three months. Bill Date and Service Address answer that; the KYC workload has the wider picture.
• Accounts payable and property teams pay hundreds of bills across sites and need amounts, due dates and account numbers.
• Energy and sustainability reporters need consumption, not money: units per meter per period, for every site.

Providers rarely change their format, so a portfolio of sites reduces to a handful of layouts, each learned once and remembered.

Utility bill fields docRefinery extracts

There are 55 definitions, the most after invoices. Mailing and service addresses are separate because they often aren't the same place, and that difference matters for proof of address.

Example utility bill JSON

A trimmed excerpt with invented values. Total Amount Due is read, but its location is null: the figure is printed in the summary and again on the tear-off stub, with nothing to say which copy it is. See Extract a document for the complete response.

{
  "document_type": "utility_bill",
  "result": {
    "fields": [
      { "alias": "Service Address", "text": "1 Example Road, Sampletown", "field_confidence": 0.95,
        "location": { "page": 1, "bbox": [0.08, 0.20, 0.42, 0.24] } },
      { "alias": "Total Amount Due", "text": "84.17", "field_confidence": 0.97, "location": null },
      { "alias": "Deposit Held", "text": null, "location": null },
      { "alias": "Meter Item", "table": [[
        { "alias": "Meter Number", "text": "E00000001", "field_confidence": 0.96,
          "location": { "page": 2, "bbox": [0.08, 0.30, 0.24, 0.32] } },
        { "alias": "Reading Type", "text": "Estimated", "field_confidence": 0.90,
          "location": { "page": 2, "bbox": [0.60, 0.30, 0.72, 0.32] } },
        { "alias": "Consumption", "text": "312", "field_confidence": 0.94,
          "location": { "page": 2, "bbox": [0.76, 0.30, 0.82, 0.32] } }
      ]] }
    ],
    "extraction_approval": "auto",
    "low_confidence_fields": []
  }
}

Why bill portfolios suit docRefinery

Across a property portfolio, bills are a volume problem first. A stack scanned into one long PDF is segmented into the bills inside it, and each provider's layout is recalled once it has been learned. Meter Item and Charge Item come back as rows, so a site with four meters stays four rows. Insights shows vendors and flagged fields across the account, so a provider that keeps failing stands out.

Several teams can share one account without arguing over the output. The organisation governs which fields are on, and a Return as alias names each value for where it is going: Consumption Amount can arrive as kwh_used for a carbon report, Service Address as service_addr for onboarding. Every bill, from any provider, returns the same format as an invoice or a statement, so one parser feeds AP, onboarding and reporting.

Bills from several countries are covered by 70+ languages and scripts. The XLSX export gives each table its own sheet, corrections included, and a reviewer's fix can become a suggested fix, tested on that provider's own bills before anyone applies it.

What makes utility bills hard to read

• Units aren't comparable. A UK gas bill turns metered volume into kWh using a correction factor and a calorific value, so one bill can print cubic metres and kWh side by side. Reporting on the wrong one is an easy mistake.
• Estimated readings. Actual and estimated reads look alike, and an estimate can distort an energy report.
• Balance arithmetic. Previous balance, payments, adjustments and current charges all sit near the amount due, and only one of them is what's owed now.
• Two amounts due. Some bills print a different figure for payment after the due date.
• Multi-meter and multi-service bills. One document can cover several meters, or gas and electricity together, with tiered rates, standing charges and levies.
• Charts. Usage graphs carry figures that look like data but belong to other periods.

How low-confidence utility values are handled

What counts as too uncertain depends on who is reading. An onboarding team can hold Service Address at a high threshold and ignore the charges; an energy analyst can do the reverse with Consumption and Consumption Unit. Values under the bar go to the review queue with their box highlighted. Where a bill doesn't print a field, you get null: docRefinery won't work out a kWh figure the page doesn't show, and it won't fill in a reading. If the numbers need to agree, a workflow rule can say that previous and current readings match the consumption, and a failure is flagged.