The categorization engine your ERP was never built to be.
Spendaq sits between your transaction sources and your general ledger — reading, mapping, flagging. Without touching your existing systems.
Built on your chart of accounts, not a generic taxonomy.
Generic categorization models fail because account code 6040 means SaaS subscriptions at one company and professional development at another. Spendaq trains on your specific chart of accounts, cost center hierarchy, and historical vendor-to-GL mappings — not an industry average.
Custom GL training: Spendaq reads your chart of accounts and builds a categorization model specific to your account codes and cost center structure.
Vendor enrichment: We cross-reference vendor names against a normalized vendor database — AMZN MKTP and Amazon are the same vendor.
Policy rule engine: Upload your T&E policy PDF. We parse expense limits, approved vendors, and reimbursement caps into rule checks applied to every transaction.
Anomaly detection: Statistical baseline per vendor × cost center × time window. Deviations surface as flags before they get coded into your ERP.
Confidence scoring: Every mapping gets a confidence score. High-confidence mappings auto-approve. Low-confidence routes to your reviewer queue — so you only see what actually needs attention.
Six modules. One categorization layer.
Each module targets a specific failure mode in how ERPs handle modern corporate spend. Spendaq is not an expense management platform, not a corporate card, not an AP system — it is the categorization intelligence layer that sits upstream of your ERP and gets the GL coding right before entries land.
Maps any transaction description — including truncated merchant codes like AMZN MKTP or DLTD*DELTA — to your exact account code. Handles vendor aliases, foreign-currency labeling, and cryptic card processor strings.
Cross-checks amount, normalized vendor name, and date window across a rolling 30-day period. Surfaces probable duplicates before month-end coding locks — not after they've been buried in a journal entry.
Upload your T&E policy document. Spendaq parses per-diem limits, approved vendor lists, hotel rate caps, and reimbursement ceilings into rule checks applied to every transaction before it reaches the approver queue.
Shared expenses — a cross-departmental SaaS license, a catered all-hands, a shared office lease — get proportionally allocated across cost centers by rules you define once. Not manually every month-end.
Real-time spend pace against approved budget for every cost center. Alert thresholds configurable at 70%, 85%, and 100% — so cost center owners get a warning before they create a budget variance, not after you're presenting it on a slide.
Push categorized, reviewed transactions directly into NetSuite, QuickBooks Online, or Sage Intacct. Entries land in the right GL account with the right cost center tag. No manual re-entry, no copy-paste step.
98% accuracy isn't a claim. It's how we measure ourselves.
We benchmark every account's GL code match rate weekly. When a Controller overrides a Spendaq suggestion, that mismatch becomes a training signal — fed back into the model within 48 hours. Accuracy compounds: accounts typically move from 92% at onboarding to above 97% within 60 days as the model learns the edge cases specific to your vendor mix and GL structure.
GL code match accuracy (median across active accounts)
Transactions requiring human review after AI pass
Typical time to reach >96% accuracy for new accounts
| Date | Vendor | Amount | AI Category | GL Code | Conf. | Status |
|---|---|---|---|---|---|---|
| 06/24 | ZOOM.US | $149.00 | SaaS Subscriptions | 6040 | 99% | PASS |
| 06/24 | AMZN MKTP | $87.43 | Office Supplies | 7210 | 72% | DUPLICATE |
| 06/23 | UNITED 1K | $412.00 | T&E — Air Travel | 6510 | 97% | PASS |
| 06/23 | MARRIOTT HTL | $319.00 | T&E — Lodging | 6510 | 95% | POLICY FLAG |
| 06/22 | DELL TECH | $1,899.00 | Capital Equipment | 1510 | 91% | PASS |
| 06/22 | ACME CONSULT | $4,200.00 | Prof. Services | 6830 | 61% | REVIEW |
See Spendaq categorize your actual transactions.
We run a live walkthrough on a sample of your transaction types — not a demo dataset. 30 minutes. No slides.