LogoTailride
Settings

AI Rules

Customize how Tailride's AI processes your documents with natural language rules.

Overview

AI Rules are natural language instructions that adjust how Tailride processes your invoices. Each rule is applied independently to the document currently being processed, with no access to historical data.

For simply skipping emails from specific senders or domains, use Skip Rules instead — those are deterministic filters, while AI rules guide the AI's judgment.

How It Works

Each rule acts as an instruction to the AI during document processing. The AI considers your rules alongside its base knowledge and the document's content.

Available Input Fields

Email Context

When the document was extracted from an email:

  • Sender email address
  • Recipients email addresses
  • Reply-To email address
  • Email subject
  • Original filename

Document Content

  • Full text content of the document

Output Fields You Can Influence

  • Amount and currency
  • Invoice ID / reference
  • Creation and due dates
  • Document classification (invoice / not invoice)
  • Vendor name

Example Rules

Sender Filtering

"If the email is from @uber.com or @bolt.com, classify it as not an invoice as these are ride receipts"

Document Type

"If the document contains 'Quote', 'Estimate', or 'Pro Forma' in the title, mark it as not an invoice"

Vendor Name Standardization

"For invoices from AWS, always use Amazon as the vendor name"

Managing Rules

  • Add: type the rule in the input field and confirm — the AI assistant reviews it before it is saved (see below)
  • Edit: click the edit icon, change the text inline, and save — changed rules go through the same AI review
  • Remove: click the trash icon and confirm

AI Review Before Saving

When you add a rule or save a change to one, Tailride first analyzes it and shows you:

  • What will happen — a plain-language explanation of which documents the rule will affect and how
  • A risk assessment — whether the rule looks safe, needs attention, or is risky. A rule is flagged as risky when it could cause real invoices to be skipped, for example "ignore everything from Amazon" would also drop genuine Amazon invoices
  • Warnings — overly broad matching, ambiguous wording, or overlaps with your existing rules
  • Alternatives — your rule as written plus one or more more precise rewrites

Pick the version you want and save it, or cancel to rewrite the rule yourself. The explanation is shown in the language you use Tailride in.

If the analysis is temporarily unavailable, you can still save the rule as written.

Important Notes

Non-Deterministic Processing

Rules provide context to improve AI decisions but are not guaranteed to be followed exactly — the AI balances them with document evidence.

Scope

  • Rules apply per document, during processing of new documents only
  • They cannot modify already-processed documents
  • They have no access to historical data or other invoices

Best Practices

  • Keep rules simple and specific; use clear natural language
  • Avoid contradictory rules
  • Complex rules may affect accuracy — contact support if you need help optimizing them

On this page