Skip to content

CAP 135

Category: Expense intelligence
Business Goal: BG-003 — Intelligent financial assistant; BG-009 — Automate everything that can be automated
System Domain: SD06 — Snowflake Analytics & Risk Platform
Satisfying Module: MOD-088 Expense classification engine (module in design)

Description

Every transaction is classified across five dimensions: ownership (personal / business / mixed), purpose (client meeting, supplies, commute, etc.), tax treatment (fully claimable / 50% entertainment / non-claimable / personal), accounting category, and confidence score. The classification model is a hybrid of ML inference and user-defined rules, improving over time as users confirm or override suggestions.

This is the engine behind the "5 hours in Xero → 5 minutes a month" proposition. See Expense Intelligence platform for full context.

Classification dimensions

Dimension Values Notes
Ownership Personal / Business / Mixed / Rental Primary split
Purpose 15+ intent categories e.g. Client entertainment, Office supplies, Commute
Tax treatment Fully claimable / 50% entertainment / Non-claimable / N/A NZ/AU rules built in
Accounting code Chart of accounts mapping Xero/MYOB compatible
Confidence 0–100% Drives automation threshold

Classification inputs (ranked by weight)

  1. User-confirmed rules for this merchant/context
  2. Merchant category (MCC) and enriched name
  3. Behavioural pattern (time, recurrence, location cluster)
  4. Geo-spatial context (office cluster, trip period)
  5. Population signal (privacy-safe aggregate: most users classify this as X)

Automation thresholds

  • ≥95% confidence → auto-classify, no prompt
  • 70–94% → classify, show subtle indicator, easy one-tap override
  • 40–69% → surface as suggestion in daily triage queue
  • <40% → require manual input

Learning loop

User confirmation / override → rule candidate created → confidence updated → future transactions auto-classified → prompt rate decreases over time. Goal: new users reach >80% auto-classify within 90 days of first use.