9 Use Cases. One Predictive Database.
Predictive accounting on Aito — multi-tenant, no model training
A working reference implementation of a predictive accounting SaaS. 255 customer companies, 128K invoices, one shared Aito instance — predictions scoped per-tenant via customer_id in the where clause. No model training, no retraining schedule, no per-tenant pipeline.
Predictive Ledger — 9 Features Grouped by Function
Every feature is a query against the same multi-tenant Aito table. Pick a feature to read its implementation guide on GitHub.
Payables Automation
Code an invoice, route it for approval, match a payment, flag what looks wrong — every step one query.
Invoice Processing
Predict GL code + approver per invoice with the full $why factor decomposition
View Use Case →View Source →
Smart Form Fill
Type any field, the rest predict. Vendor → fills GL, approver, cost centre, VAT
View Use Case →View Source →
Payment Matching
Bank transactions matched to invoices via schema link in one _predict invoice_id call
View Use Case →View Source →
Anomaly Detection
Inverse prediction — if the actual GL is not in the top-3 predicted, flag it
View Use Case →View Source →Accounting Intelligence
Discover statistical patterns and learn from human corrections — close the loop without retraining.
Rule Mining
Discover patterns with support and lift, drill into compounds via chained _relate
View Use Case →View Source →
Human Overrides → Rules
Two-pass _relate on overrides surfaces input → output rules from corrections
View Use Case →View Source →
Rule Performance
Mined rules per-customer with lift, support, and override coverage
View Use Case →View Source →
Prediction Quality
Real _evaluate accuracy on held-out samples with green/red diff vs majority baseline
View Use Case →View Source →Multi-tenancy & Quality
One Aito table, 255 tenants. Same vendor, different customer_id → different prediction. No per-tenant infra.
Implementation Details
Explore the source, data model, and architecture decisions
{
"from": "invoices",
"where": {
"customer_id": "CUST-0000",
"vendor": "Telia Finland Oyj",
"amount": 890.50
},
"predict": "gl_code",
"select": ["$p", "feature", "$why"]
}
Why This Pattern Works for Accounting SaaS
One predictive database, every tenant scoped by customer_id, no model deployment between corrections
Zero MLOps
- No model training, no retraining schedule
- Override a prediction, the next reflects it instantly
- Schema additions accepted in place
Multi-Tenant by Default
- One Aito DB, 255 tenants
- customer_id in where clause = isolation
- Same vendor, different tenant → different prediction
Production-Proven
- Same operators Posti runs at 95% accuracy
- 20-168ms _predict latency
- Sub-50ms when served from precomputed JSON
Evaluate in 15 Minutes
From live demo to your own data in three steps
Address
Episto Oy
Putouskuja 6 a 2
01600 Vantaa
Finland
VAT ID FI34337429
