What is the model layer cake?
Model Layer Cake
The Model Layer Cake is Digits' multi-layer AI architecture that improves bookkeeping accuracy by consulting a sequence of specialized models, from the most specific to the most general. When a transaction enters the ledger, Digits works through the layers in order. The first model with enough context and confidence to correctly classify the transaction provides the answer. If no layer can determine the correct accounting treatment, the transaction is routed for additional investigation or human review.
The layers, in the order they are consulted:
Company Model: Consulted first. Learns the vendors, customers, and accounting patterns unique to a specific business.
Firm Model: Consulted next for accounting firms using Digits. Learns the firm's accounting standards and applies them consistently across its clients.
Global Model: Forms the foundation of the Model Layer Cake. Learns from hundreds of billions of dollars in real business transactions to provide a strong baseline, especially for new clients with little or no historical data.
Fallback Agents: When no model can confidently resolve a transaction, specialized AI agents perform additional reasoning and investigation. If uncertainty remains, the transaction is surfaced to an accountant for review.
Example: A transaction from a familiar vendor is categorized immediately by the Company Model. A transaction from a new vendor falls through to the Firm Model, which may recognize it from another client. If it's new to the firm as well, the Global Model provides guidance based on broader accounting patterns. Truly ambiguous transactions are investigated by Fallback Agents, and any remaining uncertainty is routed to an accountant.
Related terms: AI Agent, Agentic General Ledger™ (AGL®), Confidence Score, Tiered Intelligence
