Assigning accounting attributes such as Suppliers, Expense Accounts, VAT Rates, Cost Centres, Cost Carriers and other dimensions to transactions, invoices and reimbursements follows different rules, patterns and practices in every organisation.
The Pre-Accounting Agent uses AI to automatically pre-code the majority of your transactions, invoices — and business expenses in the Reimbursements module — for you.
Here is how we do it:
Learning from your data
The Agent reviews the accounting attributes you assign when exporting expenses. Those historical mappings are the foundation the Agent uses to learn how your organisation codes spend.
Building an AI model for your organisation 🤖
Once you’ve exported your first 150 transactions, 25 invoices or 25 business expenses (Reimbursements), we automatically train an organisation- and spend-type-specific AI model using your data. The model analyses how Expense Accounts, VAT rates and other dimensions were assigned previously and uses those patterns to predict the correct accounting details for future items.
Prediction with AI 🤖
For all future transactions, invoices and reimbursements the model attempts to predict VAT, Expense Account and other dimensions. Predictions are probability‑based: when the Agent's confidence for a single value falls below the minimum threshold, it will leave the field empty rather than guess, reducing the risk of incorrect matches.
When there are multiple higher probability options, the Agent surfaces up to five suggestions in the dropdowns, pinned to the top so accountants can quickly choose the correct value without browsing the full list. If no suggestion meets the minimum confidence threshold, the dropdown will not show suggestions and fields remain empty for manual input.
Our internal data shows the Agent assigns correct accounting attributes more frequently than static rule sets, so we recommend enabling Agent overrides rules in Settings so the Agent’s top prediction(s) take priority. The default is rule‑first, so rules remain authoritative until you opt in.
Self-learning with AI 🤖:
The Agent continuously improves through a feedback loop. If it makes an incorrect prediction, you can correct it and the Agent will learn from that correction. Each time you export transactions, invoices or reimbursements, we (re)build an updated model that incorporates your previous corrections and the newest exported data.
If you delete a specific Expense Account, VAT rate or other value, we rebuild the model so those values are no longer predicted.
Which spend types are supported and which fields can be predicted
Card transactions
Fields that can be predicted: Expense Account, VAT Rate, Cost Center, Cost Carrier, Custom Dimensions (if configured), Supplier
Invoices
Fields that can be predicted: Expense Account, VAT Rate, Cost Center, Cost Carrier, Custom Dimensions (if configured)
Reimbursements (business expenses only)
Fields that can be predicted: Expense Account, VAT Rate, Cost Center, Cost Carrier, Custom Dimensions (if configured)
When the Agent can’t predict your data entries (and what to do)
New or rarely used Expense Accounts or VAT rates
The Agent needs examples to learn. If a value appears very rarely in your data (for example fewer than ~10 transactions), it’s unlikely the Agent will predict it reliably — and it may be ignored.
Identical items linked to different accounts
If identical transactions or invoices are routinely coded to different accounts (e.g. the same supplier is sometimes charged to different Cost Centres), the Agent will struggle to infer a clear pattern and will tend to pick the most common value.
Low confidence in predictions
When the model lacks confidence — typically due to insufficient or noisy data — it will leave fields empty instead of making a risky prediction. This behaviour protects against incorrect matches.
Note: The Pre-Accounting Agent's models are (re)trained using data that has been exported to your accounting system. Assigning accounting attributes without exporting the transaction, invoice or reimbursement will not improve future predictions. To get the best results, export consistently and keep your exported data accurate.