Antti Rauhala
Co-founder
January 20, 2025 • 4 min read
The accounts payable automation landscape just got a major upgrade. Aito's latest releases introduce game-changing capabilities that solve one of the most persistent challenges in invoice processing: making accurate predictions when you have limited historical data.
Whether you're dealing with a new vendor, an unfamiliar expense category, or a recently hired employee, traditional ML approaches often struggle. Today, we're excited to show you how Aito's enhanced prediction capabilities automatically leverage entity properties and enterprise-scale performance optimization change the game entirely.
Let's start with a real scenario every accounting team faces:
It's Monday morning, and your automated invoice processing system encounters an invoice from a new cloud services vendor. The system has never seen this vendor before, but it needs to assign a GL code, predict the appropriate approver, and route it to the right department. Traditional rule-based systems would fail here, and most ML models would make random guesses.
This is where Aito's automatic property-based predictions become invaluable.
Instead of requiring explicit prior configuration, Aito automatically uses the properties of entities to make intelligent predictions. Think of it like an experienced accountant who uses what they know about a vendor's industry, size, and service type to make smart categorization decisions.
Here's how it works in practice:
Your system encounters a new vendor "NewCloudTech Solutions" with no historical data:
{
"from": "invoices",
"where": {
"VendorName": "NewCloudTech Solutions",
"Amount": 2400,
"Description": "Monthly cloud infrastructure"
},
"predict": "GLCode"
}
Traditional ML: Would make a random guess because this vendor has never been seen before.
Instead of requiring manual configuration, Aito automatically leverages vendor properties to make intelligent predictions. If your vendor table includes properties like this:
// Vendor table entry for the new vendor
{
"VendorName": "NewCloudTech Solutions",
"Industry": "Technology",
"ServiceType": "Cloud Infrastructure",
"CompanySize": "Medium"
}
Aito automatically uses these properties to find similar vendors and their GL code patterns. The prediction query remains simple:
{
"from": "invoices",
"where": {
"VendorName": "NewCloudTech Solutions",
"Amount": 2400,
"Description": "Monthly cloud infrastructure",
"VendorIndustry": "Technology",
"VendorServiceType": "Cloud Infrastructure"
},
"predict": "GLCode"
}
Result: Aito automatically uses all fields from the where clause (VendorName, Amount, Description, VendorIndustry, VendorServiceType) to find similar patterns in historical data, maintaining high accuracy (85-90%) even for completely new vendors - a significant improvement over traditional ML which typically drops below 50% accuracy for unknown entities.
Let's examine how these improvements transform high-volume invoice processing in production environments:
Challenge: Bob, a new IT manager, joins the company. He has zero historical invoice approvals.
Traditional approach: All invoices default to a senior manager, creating bottlenecks.
Aito's automatic approach: Uses Bob's employee properties to make intelligent assignments. If your employee table has:
// Employee record for Bob
{
"Email": "bob@company.com",
"Role": "Manager",
"Department": "IT",
"ApprovalLimit": 10000,
"Seniority": "Mid-level"
}
The prediction query automatically leverages these properties:
{
"from": "invoices",
"where": {
"Department": "IT",
"Amount": 3500,
"InvoiceType": "Software License"
},
"predict": "Approver"
}
Result: Aito automatically uses all fields from the where clause (Department, Amount, InvoiceType) to find similar patterns in historical data. The system learns from how other IT managers with similar roles and approval limits have been assigned invoices, ensuring Bob gets appropriate assignments despite having no historical data.
Challenge: Holiday season brings unique expenses (catering, gifts, events) that appear infrequently.
Aito's approach: Automatically uses descriptive keywords and timing to find similar expense patterns:
{
"from": "invoices",
"where": {
"Description": "Holiday catering services",
"Month": "December",
"Amount": 1200,
"VendorCategory": "Food Service"
},
"predict": "GLCode"
}
Outcome: Aito automatically uses all the fields from the where clause (Description, Month, Amount, VendorCategory) to find similar patterns in historical data, matching this to similar event-related expenses from previous years and other catering services, achieving 90%+ accuracy on seasonal expenses compared to 60% with rule-based systems.
Aito's enhanced predict endpoint is optimized for large-scale financial operations:
Based on our performance benchmarks:
With Aito's enhanced predict endpoint, you can now leverage all available metadata to make intelligent predictions, even for entities with no historical data. The system automatically uses linked table attributes when making predictions, eliminating the cold start problem.
Here's how to implement contextual GL code prediction for your invoice processing:
First, create a schema that includes your invoice data and any linked tables (vendors, employees):
{
"schema": {
"invoices": {
"type": "table",
"columns": {
"InvoiceID": {"type": "String"},
"VendorName": {"type": "String", "link": "vendors.VendorName"},
"Amount": {"type": "Decimal"},
"Description": {"type": "Text"},
"Department": {"type": "String"},
"GLCode": {"type": "String"},
"Approver": {"type": "String", "link": "employees.Email"}
}
},
"vendors": {
"type": "table",
"columns": {
"VendorName": {"type": "String"},
"Industry": {"type": "String"},
"ServiceType": {"type": "String"},
"CompanySize": {"type": "String"}
}
},
"employees": {
"type": "table",
"columns": {
"Email": {"type": "String"},
"Department": {"type": "String"},
"Role": {"type": "String"},
"ApprovalLimit": {"type": "Decimal"}
}
}
}
}
Upload your historical invoice data along with vendor and employee metadata:
# Upload schema
curl -X PUT https://your-instance.api.aito.ai/api/v1/schema \
-H "x-api-key: YOUR_API_KEY" \
-H "content-type: application/json" \
-d @schema.json
# Upload data
curl -X POST https://your-instance.api.aito.ai/api/v1/data/invoices/batch \
-H "x-api-key: YOUR_API_KEY" \
-H "content-type: application/json" \
-d @invoices.json
Now you can predict GL codes for new invoices, leveraging all available metadata:
{
"from": "invoices",
"where": {
"VendorName": "NewCloudTech Solutions",
"Amount": 2400,
"Description": "Monthly cloud infrastructure",
"Department": "IT"
},
"predict": "GLCode"
}
The enhanced predict endpoint automatically uses the vendor's properties (Industry, ServiceType) from the linked vendors table to make an informed prediction, even though this vendor has never been seen before.
# Python SDK example
from aito import AitoClient
client = AitoClient(api_key="YOUR_API_KEY", instance_url="YOUR_INSTANCE_URL")
# Make prediction
result = client.predict(
from_table="invoices",
where={
"VendorName": "Unknown Vendor",
"Description": "Miscellaneous expense",
"Amount": 150
},
predict="GLCode"
)
# Use confidence to determine automation level
if result.hits[0]["$p"] > 0.85:
# High confidence: Auto-assign
gl_code = result.hits[0]["GLCode"]
process_automatically(gl_code)
elif result.hits[0]["$p"] > 0.70:
# Medium confidence: Flag for review
queue_for_quick_review(result.hits[0]["GLCode"])
else:
# Low confidence: Manual processing
route_to_accounting_team()
The key innovation in Aito's enhanced predict endpoint is how it automatically leverages metadata from linked tables. When predicting a GL code for an invoice from a new vendor, the system:
This approach solves the cold start problem - new vendors get accurate predictions from day one based on their metadata, not random guesses.
Based on production implementations in enterprise accounting environments:
These prediction enhancements represent just the beginning. Here's what we're seeing in the accounting automation space:
Ready to transform your invoice processing with contextual predictions? Here's how to begin:
The future of accounting automation isn't just about speed—it's about intelligence that adapts, learns, and makes smart decisions even when facing the unknown.
Have questions about implementing contextual predictions for your accounting workflows? Reach out to our team - we're here to help you build the intelligent accounting platform your team deserves.
Back to blog listEpisto Oy
Putouskuja 6 a 2
01600 Vantaa
Finland
VAT ID FI34337429