How Zera Books maps extracted transactions to QuickBooks and Xero charts of accounts — confidence scoring, category mapping logic, and pattern learning methodology.
After Zera AI extracts transactions from a document, the categorization engine runs a multi-layer classification pipeline. It does not use rigid keyword rules — it uses a trained model that understands transaction context, merchant type, amount patterns, and historical firm behavior.
Raw transaction descriptions are cleaned: POS codes, card suffixes, geographic noise, and date stamps removed. "POS PURCHASE 1234 AMAZON.COM WA 12/14" becomes "Amazon.com".
Normalized description is matched against a merchant database trained on 847M+ transactions. Each merchant carries a primary category vector (e.g., Amazon → Office Supplies / Software / COGS depending on amount/context).
Amount, transaction type (debit/credit), day-of-week, frequency, and surrounding transactions are weighed. A $49.99 monthly charge to Adobe maps differently than a $4,200 one-time charge.
If the firm has previously categorized this merchant differently, the firm-specific rule takes priority over the global model. This is how Zera Books personalizes to each practice's chart of accounts usage.
Each transaction receives a 0–100 confidence score. Score ≥85 → auto-categorized. Score 60–84 → flagged for review with suggestion. Score <60 → uncategorized for manual assignment.
Every transaction is scored 0–100. The threshold determines whether the AI auto-assigns the category or flags it for human review.
| Score Range | Status | Action in Export | Typical Cause |
|---|---|---|---|
| 85–100 | Auto-categorized | Category applied, no flag | Known merchant, consistent amount pattern, prior firm history |
| 60–84 | Review suggested | Category suggested, highlighted yellow | New merchant, ambiguous description, multiple plausible categories |
| 0–59 | Uncategorized | Left blank, flagged for manual | Unknown payee, missing description, ATM/cash withdrawal |
Upload a bank statement and watch Zera Books categorize every transaction against your chart of accounts — with confidence scores.
Try for one week →Zera Books maps against three chart of accounts targets. Users can use the default QuickBooks or Xero standard chart, or upload a custom chart of accounts in CSV format.
| Target CoA | Account Types Covered | Custom Override | Auto-Sync |
|---|---|---|---|
| QuickBooks Online | Income, COGS, Expenses, Other Income/Expense, Assets, Liabilities | ✓ | Direct API push |
| Xero Standard | Revenue, Direct Costs, Overheads, Assets, Liabilities, Equity | ✓ | Direct API push |
| Custom / Firm-Specific | Any account type, any numbering system (3-digit, 4-digit, alphanumeric) | ✓ | CSV/Excel export |
Adobe, Microsoft 365, Slack, Zoom → "Software Subscriptions" or QBO account 6030. Confidence typically 90+.
Google Ads, Meta Ads, LinkedIn → "Advertising & Marketing." Confidence 88–95 for known platforms.
ADP, Gusto, Paychex → "Payroll Expenses." Confidence 92+ when amounts are regular.
Monthly service charges, wire fees, NSF fees → "Bank Service Charges." Confidence 95+ for standard bank fee descriptions.
Airlines, hotels, Uber/Lyft, parking → "Travel & Entertainment." Confidence 85–92.
Legal, consulting, accounting fees → "Professional Fees." Confidence 80–88; often flagged for review.
Measured across 50,000 real-world transactions from 200+ firm accounts using QuickBooks and Xero. Benchmarks reflect post-pattern-learning accuracy (after ≥3 months of firm usage).
* Accuracy measured as category match to human reviewer's final assignment. Data from Q1 2025 benchmark study.
| Transaction Type | Auto-Cat Rate | Avg Confidence Score | Notes |
|---|---|---|---|
| Software subscriptions (recurring) | 97.8% | 93.2 | Highly consistent amounts and descriptions |
| Bank fees | 96.4% | 91.7 | Standard bank fee language well-recognized |
| Payroll / direct deposit | 95.1% | 90.3 | Regular amounts, consistent payee names |
| Advertising platforms | 94.0% | 89.1 | Known platform names, variable amounts |
| General merchant purchases | 91.3% | 85.6 | Varies by merchant recognition rate |
| Cash / ATM withdrawals | 42.0% | 51.2 | No merchant context; usually uncategorized |
The categorization model improves over time per firm account. Every correction a user makes becomes a training signal for that firm's future categorizations.
When a suggested category is overridden, Zera Books stores the merchant → correct category mapping. Future transactions from that merchant are categorized using the firm's preference.
Define a split rule once (e.g., "Meals 60% Entertainment / 40% Business Development") and it auto-applies to future transactions matching that merchant.
Each client folder has independent category memory. Client A's "Amazon → COGS" rule doesn't affect Client B's "Amazon → Office Expense" mapping.
After 10+ corrections for a merchant, the model's confidence score for that merchant within the firm rises toward 95+, reducing the review-required flag rate.
Zera Books maps to QuickBooks Online default chart of accounts, Xero standard chart of accounts, and custom charts. Users can upload their own chart of accounts and the AI learns firm-specific mappings over time.
Transactions with confidence score ≥85% are auto-categorized without review. Scores 60–84% are flagged for review with a suggested category. Scores below 60% are marked uncategorized for manual assignment.
Yes. When a user overrides a suggested category, Zera Books records the correction and applies it to future transactions with similar descriptions from the same vendor or payee. Learning is scoped per firm account.
Zera Books flags transactions that frequently appear as split in similar documents. Users can define split rules (e.g., 70% Office Expense / 30% Travel) which are auto-applied on future imports.
Explore related features and documentation.
Zera Books auto-categorizes 94%+ of transactions on import — mapped to QuickBooks or Xero with confidence scores.
Try for one week →