New KPMG Agentic ERP, now available with KPMG in 140+ countries Read the announcement

AI-Powered Data Cleaning

Category
AI Features
Guide
Admin guides
Reading
3 min read
On this page
  1. 01Non-Bill Document Detection
  2. 02Duplicate Bill Detection
  3. 03Vendor and Bank Data Enrichment
  4. 04AI-Assisted Import Cleaning
  5. 05Review and Control
  6. 06Related Articles

What is this page about: Light uses AI to check incoming bills, catch duplicates, fill in vendor and bank details, and clean up data during CSV imports, all while keeping every AI suggestion reviewable before it posts.

On this page

  • Non-Bill Document Detection
  • Duplicate Bill Detection
  • Vendor and Bank Data Enrichment
  • AI-Assisted Import Cleaning
  • Review and Control
  • Related Articles

Non-Bill Document Detection

  • Upload a document to your payables inbox.
  • Light's AI checks whether it is a bill, invoice, or credit note.
  • If it's clearly something else (a marketing flyer, a contract with no amount due, a screenshot, a shipping label, or an unreadable file), Light flags it with a "Doesn't appear to be a bill" banner, asking you to review the attachment and archive it if it doesn't belong.
  • The AI only flags a document when it is highly confident the document is not a bill. This check never blocks a bill. If the AI misfires, a real bill still comes through, just with the warning banner on it.

Duplicate Bill Detection

  • Light compares each new bill's vendor and document number against existing bills.
  • If another bill from the same vendor has the same document number, Light moves the new one to a Duplicated state and skips your approval workflow, with a reference to the bill it matches.
  • If you cancel or archive the original bill, Light automatically returns the oldest duplicate to Draft, so it isn't lost.
  • This match is based only on vendor and document number. AI is not involved in the decision, so the result is the same every time.

Vendor and Bank Data Enrichment

Vendor enrichment:

  • When a vendor is created from a parsed document, Light looks up its brand data (website domain, logo, address) using an external brand database, combined with AI.
  • When a vendor is created from a card transaction, Light looks up the same brand data, but without the AI step.

Bank details enrichment:

  • When bank details are extracted from a document, Light structures and enriches them using AI.

AI-Assisted Import Cleaning

Chart of accounts import:

  1. Upload your file.
  2. Light's AI maps your file's columns to Light's account fields.
  3. Light processes each row into a structured ledger account, including account type and category.
  4. Any row the AI can't map falls back to standard handling instead of failing the whole import.

Vendor import:

  1. Upload your file.
  2. Light's AI maps your file's columns, including any custom properties, to Light's vendor fields. Review the suggested mapping, then click Continue.
  3. Imported lines are created as drafts.
  4. Review the draft lines, then run the import.

Vendor import only automates the column mapping step. There is no automatic check of the AI-suggested mapping against your file or company configuration, and no row-by-row AI processing or fallback. That level of automation is specific to chart of accounts import.

Review and Control

  • AI-extracted and AI-mapped values are suggestions. Review and edit them before anything posts.
  • Import lines stay in draft until you run the import.
  • Bills flagged as duplicates or non-bill documents stay visible in your workflow. Verify and resolve them yourself.

Also searched as: AI data cleaning, duplicate invoice detection, vendor enrichment, bank details enrichment, CSV import mapping, chart of accounts import, AP automation data quality.

Was this article helpful?