How AI Works in Digits
Learn how Digits uses AI to automatically book, categorize, and review transactions—so you only step in when human judgment is actually needed.
Welcome to Digits!
If your Inbox looks busy right after you connect data, you didn’t do anything wrong — Digits is being intentionally conservative while it learns your business. Confirm what’s right, correct what isn’t, and Digits applies that learning immediately, so the Inbox shrinks over time.
We’ll walk through how it all works below.
Overview
Digits is an AI-native accounting platform, designed to automate routine bookkeeping and surface only the items that need human judgment—without sacrificing accuracy, control, or auditability.
Unlike legacy systems where AI is layered onto an existing workflow, Digits’ AI is embedded directly into the general ledger itself. This foundation—called the Agentic General Ledger™—allows Digits to continuously book, categorize, reconcile, and review transactions as they come in, so you spend less time on repetitive bookkeeping and more time on exceptions and decisions.
Throughout the Help Center, the ✨ sparkle icon highlights features powered by Digits AI.
Where Digits AI Shows Up
Digits AI primarily works across five places in the product:
Ledger (categorization, matching, splitting, review)
Inbox (exception queue for items needing attention)
Reconciliations (statement matching + exception handling)
Documents (extraction + routing into the right workflow)
Ask Digits (search + navigation + analysis)
What Does Digits AI Do Automatically?
✨ Automatic Transaction Categorization
As transactions sync from banks, cards, payroll, and connected tools, Digits automatically:
Identifies the likely party (vendor/customer/payroll source)
Selects a category/account in your chart of accounts (Category Manager)
Applies what it has learned from your prior decisions
Routes uncertain items to the Inbox instead of guessing
✨ Continuous Reconciliation (Statement-Aware)
When statement data is available (from a connected account or an uploaded PDF), Digits automatically:
Matches statement lines to Ledger activity
Drafts the reconciliation
Surfaces exceptions like missing, duplicate, or unmatched items
Keeps reconciliation status current as items are confirmed or resolved
✨ Document Extraction and Workflow Routing
Digits can ingest documents like:
Bank/credit card statements
Bills and invoices
Receipts
Bank check images (where available)
Digits identifies the document type and routes it into the appropriate workflow (Documents, Reconciliations, Bills, etc.).
✨ Ask Digits Search and Workflow Shortcuts
Ask Digits helps you:
Search transactions, vendors, categories, and balances
Jump into workflows quickly (Ledger, Reconciliations, Documents)
Ask questions in natural language
How Digits Learns & Training Your AI
When you connect a new client’s data, Digits starts with high baseline accuracy, but it doesn’t know your client's unique business context... yet.
Expect the first 1–2 months to need the most review. That’s normal. Digits is learning your client’s vendors, naming quirks, and category preferences.
Digits automates 97% of your bookkeeping, so most transactions never require your attention. Anything that needs a human eye is automatically routed to your Inbox for review.
Think of the Inbox as your training queue, not your permanent workload. Every time you confirm or correct a transaction, Digits applies that feedback immediately to similar future transactions, so the system gets smarter while your review list gets quieter.
Digits learns at multiple levels:
Client learning (each client’s unique vendors, customers, categorizations, and preferences)
Firm learning (patterns across your firm’s client base, for accounting firms)
Global learning (patterns across all users)
Most firms see noticeable improvement by month 2–3, and accuracy continues to improve over time.
How Automatic Categorization Works
Digits categorizes a transaction when it appears in your Ledger (from a bank connection or an uploaded bank statement/csv file). Then, ✨ Digits evaluates context such as:
Merchant text/transaction description
Historical categorization patterns
Amount ranges and frequency
Related activity (e.g., transfers, deposits, fees)
Your chart of accounts structure
Digits then:
Proposes a category
Assigns a confidence score
Either posts automatically (high confidence) or routes to the Inbox (lower confidence)
This is why Digits doesn’t rely on rules you have to build and maintain. Instead, Digits learns patterns from your data and from what you confirm or correct, and it automatically creates and strengthens its own “rules” over time.
Confidence Scoring & the Inbox
Even with high accuracy, Digits is intentionally conservative when certainty is low.
Every automated categorization is assigned a confidence score.
High confidence: posted automatically
Lower confidence: routed to the Inbox
How to resolve an Inbox item
Review the suggested category and party
Confirm if correct, or change to the right category
Save/confirm
Digits learns and applies it going forward
Why the Confirm Button Might Be Greyed Out
A required field is missing (most often: no category selected). Select a category first, then Confirm becomes available.
Tip: The more consistently you confirm and correct in the Inbox, the faster Digits learns. Early investment in review pays off quickly — within a few weeks, your Inbox volume should decrease noticeably.
Reconciliations: What AI Does and What You Do
Digits reconciles by matching statement activity to your Ledger. When Digits has a statement (from a connection or an upload), it can keep reconciliation status up to date and surface only what needs attention.
How it works
Your statement gets into Digits by:
Connecting a bank/credit card (when statement data is available), or
uploading a statement PDF in Accounting → Reconciliations, or
emailing a statement to Digits (see below)
Digits matches statement lines to Ledger transactions
You get a clear reconciliation status plus a short list of exceptions (duplicates, missing items, or timing differences) to review and automatically apply necessary changes
For complete instructions, see: Using AI Bank Reconciliations in Digits
Documents, Statement Uploads, and Imports
You can bring data into Digits by connecting accounts or uploading files. What happens next depends on how and where you upload.
Below, you’ll see what happens to each file type (statements, CSVs, bills, receipts, and more) based on how it enters Digits—including Connections, Ledger import, Reconciliations, Documents, drag & drop, and email.
Hint: Where you upload a file determines what Digits does with it.
Document Type: Bank or Credit Card Statement (PDF)
Connections
Retrieves statement info when available
Ledger → Import Bank Transactions
Imports transactions into Ledger (does not start reconciliation)
Accounting → Reconciliations
Starts reconciliation
Documents
Tries to reconcile; if nothing matches, stays in Documents
Drag & Drop
Routed to Documents; tries to reconcile; if nothing matches, stays in Documents
Lands in Documents; tries to reconcile; if nothing matches, stays in Documents
Document Type: CSV
Ledger → Import Bank Transactions
Imports transactions into the Ledger and auto-categorizes them.
Upload to Documents
Stored in Documents (does not import transactions).
Drag & Drop
Routed to Documents. Stored in Documents (does not import transactions).
Email to Digits
Stored in Documents (does not import transactions).
Document Type: Bills
Upload to Documents
Automatically creates a bill for review and approval.
Drag & Drop
Routed to Documents. Creates a bill for review.
Email to Digits
Lands in Documents. Creates a bill for review.
Document Type: Check image
Upload to Documents
Automatically attaches to the matching check transaction in your Ledger, if one is found.
Drag & Drop
Routed to Documents. Attaches to a matching check transaction (if found).
Email to Digits
Lands in Documents. Attaches to a matching check transaction (if found).
Document Type: Receipts
Upload to Documents
Automatically attaches to a matching transaction for easy reference.
Drag & Drop
Routed to Documents. Attaches to a matching transaction (if found).
Email to Digits
Lands in Documents. Attaches to a matching transaction (if found).
Document Type: Other files (Word, Excel, general PDFs)
Upload to Documents
Stored in Documents for organization and reference.
Drag & Drop
Routed to Documents. Stored in Documents.
Email to Digits
Lands in Documents. Stored in Documents.
I got my data into Digits — what does it categorize?
✨ Digits AI categorizes transactions, not documents.
Here’s what that means once your data is in Digits:
Connected accounts (Connections): As transactions flow into your Ledger, Digits categorizes them automatically.
Ledger → Import Bank Transactions (CSV or PDF): Digits imports transactions into the Ledger and categorizes those transactions.
Files uploaded to Documents (including drag & drop or email): Digits stores the file, but does not categorize anything just because a document exists in Documents.
AI Bank Reconciliation: Reconciliation matches statement activity to Ledger transactions. If Digits flags a missing transaction and you click Add Transaction, that new transaction enters the Ledger and is categorized like any other.
In short, Digits categorizes transactions as they enter the Ledger.
Control, Auditability, and Safety
Digits is designed to support professional judgment, not replace it. You always retain control:
Transactions can be edited, reversed, or reclassified
Bulk changes are supported
Manual journal entries are supported
Changes are tracked in each transaction’s Activity Log, so you can see what changed, when, and by whom
Digits is built with bank-grade security practices, including SOC 2 Type II compliance, TLS/HTTPS for in-transit data, encryption at rest, and layered infrastructure controls.
Visit https://digits.com/security/ for more security information.
What Digits’ AI Doesn’t Do
To set clear expectations, Digits intentionally avoids certain behaviors:
🚫 It does not auto-post transactions it isn’t confident about
🚫 It does not override human decisions
🚫 It does not answer tax or legal questions
🚫 It does not perform foreign currency conversion or revaluation
✅ AI in Digits is designed to assist and accelerate professional judgment, not replace it.
Frequently Asked Questions
Does Digits use rules?
Yes and no. Digits doesn’t use rules you build and maintain. Instead, Digits relies on machine-learning models that learn patterns from your data, and you confirm/correct actions. Over time, Digits effectively creates and reinforces its own “rules” automatically—without you managing rule logic.
How accurate is Digits AI?
Digits’ AI is built for accounting — not retrofitted for it. It’s trained on 170M+ real transactions totaling $825B, so it recognizes the patterns behind actual books (and the weird edge cases that show up in the real world).
We believe accuracy claims should come with receipts. That’s why we benchmarked Digits against leading general-purpose LLMs (including GPT-5, Claude, and Gemini) on accounting-specific tasks. In internal testing, Digits performed 43% better.
For full methodology and results, download a copy of the full Agentic General Ledger whitepaper.
What does this mean for you? Digits automates up to 97% of your bookkeeping over time — so most transactions never require your attention. The exceptions that need a human eye are automatically routed to your Inbox.
Digits improves quickly as it learns each client’s vendors, naming conventions, and category preferences. Every confirmation or correction is applied immediately to similar future transactions, so accuracy increases as you work. Most firms see noticeable improvement by month 2–3.
Will Digits ever auto-post something incorrectly?
Digits applies confidence thresholds to avoid risky automation. When confidence is low, items are surfaced for review instead of being auto-posted.
Does Digits replace accountants or bookkeepers?
No. Digits removes repetitive work so accountants can focus on judgment, review, and advising—where humans add the most value.
Does Digits improve over time?
Absolutely! Digits improves over time as it learns your patterns.
When Digits first connects to a new business, it doesn’t yet know your vendors, naming conventions, or category preferences, so you’ll usually see more items routed to the Inbox.
Every time you confirm a category or correct a transaction, Digits applies that feedback to similar future transactions, so fewer items need review.
Does Digits use AI to extract information from uploaded documents?
Yes. Digits uses AI to extract and structure data from supported statements, bills, and receipts.
What's the difference between the Inbox and the Ledger?
The Inbox is for review. The Ledger is the system of record.
Digits separates review work from recordkeeping so you only spend time where judgment is needed.
If I upload statements, will Digits automatically categorize the transactions?
Yes — if you upload a bank/credit card statement in the Reconciliations workflow, Digits will extract the statement transactions, match them to what’s already in your Ledger, and help categorize anything that becomes a new Ledger transaction.
The key detail:
Uploading a statement does not directly “categorize the PDF.”
Categorization happens when transactions enter the Ledger.
What typically happens when you upload a statement PDF
Digits reads the statement and starts an AI Bank Reconciliation.
It matches statement lines to existing Ledger transactions.
If the statement contains missing items (i.e., not in the Ledger), Digits surfaces them as exceptions.
When you click Add Transaction (or otherwise create those missing Ledger transactions), Digits will categorize them using its normal AI categorization models.
If you want Digits to create Ledger transactions from a statement
Use the upload path that creates/feeds Ledger transactions (e.g., statement upload via Reconciliations + adding missing items, or “Import Bank Transactions” where available). Once those transactions exist in the Ledger, Digits will categorize them.
What’s the difference between CSV import and PDF statement upload?
The difference depends on where you upload the file.
Ledger → Import Bank Transactions (CSV or PDF) Imports transactions into the Ledger and automatically categorizes them. Does not start reconciliation.
Accounting → Reconciliations (PDF statement) Starts reconciliation. Digits reads the statement, matches transactions to the Ledger, and surfaces exceptions.
Upload to Documents, Drag & Drop, or Email Files are stored in Documents. If the file is a bank or credit card statement PDF, Digits will try to start reconciliation. If it can’t find anything to reconcile, the file stays in Documents. CSV files uploaded this way are stored only — they do not import transactions.
In short: Use Ledger → Import Bank Transactions to add transactions. Use Accounting → Reconciliations to reconcile a statement.
Got a question or feedback? Our team is all ears—come say hello!