Table of Contents
Overview
Trovata AI is a secure generative AI chat assistant built specifically for corporate treasury, finance, and accounting teams. It allows users to ask questions about their own bank data, without that data ever leaving their private Trovata instance. Powered by large language model (LLM) capabilities to interpret intent, Trovata AI combines the flexibility of conversational AI with the accuracy, transparency, and security of Trovata’s analytics platform. The result is fast, trustworthy insights into key cash flows, balances, forecasting, and transaction details, delivered with full visibility into the underlying data and calculations.
Powered By: Claude Sonnet 4.6 (Anthropic) on AWS Bedrock AgentCore
Trovata AI Security
Summary: You get the analytical power of a state-of-the-art AI model with enterprise-grade data isolation and security. Your financial data is processed securely, never used for training, and never accessible to anyone but you.
The entire system runs on AWS Bedrock AgentCore, a secure, isolated environment purpose-built for enterprise AI. When you ask a question, the AI retrieves your data through Trovata's own verified API layer, analyzes it, and presents the answer, all within that secure infrastructure. Here's what that means for you:
Your data is never used to train the AI. Not your balances, not your transactions, not your questions. AWS Bedrock guarantees that your inputs and outputs are completely isolated from model training. Your data goes in, your answer comes out, and nothing is retained or learned from.
Anthropic never has access to your data. Claude Sonnet 4.6 powers the intelligence, but Anthropic as a company has no access to anything processed in your Trovata sessions. That's a core guarantee of the AWS Bedrock architecture.
Nothing is stored beyond your session. Your financial data isn't persisted, cached, or logged outside of your Trovata instance. When your session is over, it's over.
Your conversations are private. Other users in your organization cannot see your chat history. Your sessions belong to you.
Errors are transparent. If a tool call fails or data isn't available, the AI tells you what happened. It never silently fills in the blanks with made-up numbers.
Getting Started
Check Your Access
To use Trovata AI Chat, two things need to be in place:
Feature Enabled: Trovata AI must be turned on for your organization.
Permission Assigned: Your user role needs the Read Trovata AI permission in the User & Entitlements section of Trovata. The video below displays a walkthrough on how to create an AI user group for your organization.
You'll find Trovata AI in the left sidebar navigation (or go directly to /trovata-ai). Log in with your standard Trovata credentials. No separate account needed. If you can't see Trovata AI in your sidebar, contact your organization's Trovata administrator and/or Trovata support (via chat or email at support@trovata.io) to verify that the feature is enabled and your role has the right permissions.
Your First Conversation
When you open Trovata AI Chat for the first time, you'll see a landing screen with your chat history, as well as three columns:
Chat History: Displays the last 30 days of chat activity. Chats can be deleted at the user's discretion.
Examples: Clickable sample questions you can quickly have the AI answer. Keywords are highlighted in bold so you can scan quickly.
Capabilities: Analyzes your financial data using natural language, keeps your data private within Trovata, and stays focused on cash related questions (i.e. No Netflix show recommendations)
File Upload: Allows users to attach file references and/or upload templates. Maximum file size is 100 MB. Supported file types include PDF, XLSX, CSV, TXT, and DOCX.
Limitations: An honest note that the AI may occasionally produce inaccurate information, with a link to this article for more detail. See below for a full list of current limitations.
Type a question in the input field (the placeholder reads "Ask Trovata AI about your bank data...") and press Enter, or tap one of the example cards to jump right in.
General Tips
Small changes to how you phrase a question can make a big difference in the result.
Be specific about time. "Last 30 days" or "Q1 2026" gives the AI a clear window to query. "Recently" leaves too much room for interpretation.
Name your accounts. "Chase operating account" gives the AI something to search for. "My main account" doesn't.
Say how you want to see it. Adding "and plot it" or "in a table" helps the AI format the response the way you need it.
Mention currency when it matters. "In EUR" or "converted to USD" makes sure you get the right numbers in the right denomination.
Use follow-ups. The AI maintains context within your session, so you can build on previous answers without re-explaining. Just keep the conversation going.
AI Output Types
AI responses can include three types of output. You'll see tabs below the response text to switch between them.
Tables
Interactive, paginated data tables with 10 rows per page. Currency values are automatically formatted. You can export any table to CSV or XLSX using the buttons above the table. Table columns can be adjusted (added/changed/removed) via user chat input.
Charts
Line graphs for trends over time (balance history, cash flow patterns) and bar graphs for comparisons (transaction breakdowns, category splits). Charts include interactive tooltips when you hover, and can be exported as PDF, PNG, or JPEG.
Queries
The underlying data query, displayed in a formatted code block. This shows you exactly what data the AI retrieved and from where. Helpful for verifying results or understanding the AI's approach.
Tip: You can switch between a chart view and a table view to inspect the same data in different ways. If you ever want to double-check a number in a chart, toggle to the table for exact values.
AI Chat Workflows
Trovata AI Chat works with the data in your instance. Here's what it covers:
1). Account Balances: Current and historical balances, trends over time, comparisons by currency, and balance changes across accounts.
2). Transactions: Search, filter, and analyze by date, account, amount, type, tag, description, etc.
3). Tags: Browse and/or filter data using your custom transaction tags. Users can also use AI to create tags one by one or in bulk using the file uploader (see below for more details).
4). G/L Tags: Browse and/or filter data using your custom G/L tags. Users can also use AI to create G/L tags one by one or in bulk using the file uploader (see below for more details).
5). Reporting: Daily, weekly, monthly, quarterly, or annual trends showing credits, debits, and net flow. Users can also use AI to create balance, transaction, and summary reports.
6). Data Streams: Users can use AI to create both manual and ML data streams one by one or in bulk using the file uploader (see below for more details).
7). Forecasting: Make projections on expected cash burn or future cash balances.
8). Data Integrity & Account Reconciliation: Account reconciliation status and discrepancy identification.
1). Account Balances
Example Prompts
"What's my current balance across all accounts?"
"Show me daily balance trends for my USD accounts over the last 30 days."
"Which accounts had the biggest balance change last week? Highlight everything with a >20% change."
"Compare balances by currency for Q1."
Best Practices
Specify a time window. "Last 30 days" or "Q1 2026" yields more precise results than "recently."
Name your banks or accounts when you can. "My JP Morgan accounts" gives the AI a clear filter to work with.
Ask for a chart when reviewing trends. Adding "and plot it" or "show a graph of this" surfaces a visual that's easier to read than a long table.
2). Transactions
Example Prompts
"Show me all debit transactions over $50,000 this month."
"Find all transactions tagged as 'Payroll' in the last 30 days."
"What are the 5 largest inbound wires this quarter?"
"Show me transactions from Paychex in the last 60 days."
Best Practices
Combine filters for sharper results. Leverage properties like "Amount", "Credit/Debit", and Date Ranges to narrow results as needed.
Use description keywords when you don't have a tag. If a transaction type isn't tagged yet, the AI can still search within a transaction's metadata.
Ask follow-up questions to either dig deeper or to query broader results.
3). Tags
Example Prompts
"Help me identify why my AR tag isn't capturing lockbox transactions."
"What tags do I have set up?"
"Show me month to date cash flow totals for XYZ tags."
"Create a tag called 'Payroll' for debit transactions containing the words ADP or Payroll Fees"
"Update the Payroll tag to also include transactions from Paychex."
"Delete the tag called 'Miscellaneous'."
Bulk Tag Creation via File Upload
To create, edit, or delete multiple tags at once, prepare a CSV or Excel file with one tag per row. Include columns for name, TQL filter criteria (keywords, direction, amount range), and any other filter criteria. Upload the file in chat and ask the AI to perform a specific action or series of actions with the data provided. If there are any open questions, the AI will summarize what it found and ask for confirmation before proceeding.
Upload Template: BULK TAG CREATION
Best Practices
Be specific with filter criteria. Keywords, direction (credit/debit), and account information are especially helpful for the AI to build accurate tags.
Confirm before deleting. If you want to bulk delete, be very specific with the tag name(s) you'd like removed.
For bulk uploads, start small. Test with 5 rows first to confirm your column format is in a readable format before uploading a larger file.
4). G/L Tags
Example Prompts
"What G/L tags do I have set up?"
"Help me identify why my ACH G/L tag isn't capturing ACH transactions."
"Show me month to date totals for transactions tagged to my Payroll G/L tag."
"Create a G/L tag called 'Lockbox Deposits' for credit transactions containing the description "lockbox", with debit code 1000-1100 and credit code 1000-1200."
"Update the Payroll G/L tag to also include transactions from Paychex."
"Delete the G/L tag called 'Lockbox Deposits'."
Bulk G/L Tag Creation via File Upload
To create, edit, or delete multiple G/L tags at once, prepare a CSV or Excel file with one tag per row. Include columns for name, TQL filter criteria (keywords, BAI codes, direction, accounts), and the credit and debit G/L codes to assign. Upload the file in chat and ask the AI to perform a specific action or series of actions with the data provided. If there are any open questions, the AI will summarize what it found and ask for confirmation before proceeding.
Upload Template: BULK G/L TAG CREATION
Best Practices
Be specific with filter criteria. BAI codes, direction (credit/debit), account, and description keywords are especially helpful for the AI to build accurate G/L tags.
Remember that each transaction can only be assigned one G/L tag.
Order matters. Tags are applied top to bottom, so higher-priority rules should appear first in your list.
Confirm before deleting. If you want to bulk delete, be very specific with the G/L tag name(s) you'd like removed.
For bulk uploads, start small. Test with 5 rows first to confirm your column format is readable before uploading a larger file.
5). Reporting
Example Prompts
"Show me all payments made out of my North America entity over the last month."
"Find all transactions tagged as 'Payroll' in the last 30 days."
"What are the 5 largest inbound wires this quarter?"
"Show me transactions from Paychex in the last 60 days."
Best Practices
Combine filters for sharper results. Leverage properties like "Amount", "Credit/Debit", and Date Ranges to narrow results as needed.
Use description keywords when you don't have a tag. If a transaction type isn't tagged yet, the AI can still search within a transaction's metadata.
Ask follow-up questions to either dig deeper or to query broader results.
6). Data Streams
Example Prompts
"Create a machine learning stream for AR deposits using my 'AR Deposits' tag."
"Build a manual stream called 'Rent' for a $45,000 outflow on the first of each month."
"Show me all data streams currently in my forecast."
"Update my Payroll stream to reflect a 5% increase starting in July."
"Which of my streams have the largest variance against actuals this quarter?"
Bulk Data Stream Creation via File Upload
To create, edit, or delete multiple data streams (ML or Manual) at once, prepare a CSV or Excel file using the upload template below. Upload the file in chat and ask the AI to create each stream. The AI will confirm what it found and flag any rows with missing or ambiguous information before proceeding.
Upload Template: BULK MACHINE LEARNING DATA STREAM CREATION
Upload Template: BULK MANUAL DATA STREAM CREATION
Best Practices
Know which stream type fits your cash flow before prompting:
Machine Learning streams work best for recurring, high-volume activity (AR, daily deposits, credit card receipts) where historical patterns exist.
Manual streams are better for fixed, known obligations like rent, loan payments, or one-time events.
Make sure your tags are set up first. ML streams are best powered by tagged transaction history. Aim for 90–100% of historical cash volume tagged before building your forecast so that your data stream is representative of the majority (or all) of your cash activity.
Be specific about the stream configuration. Tell the AI the stream name, the tag it should reference, the direction (inflow or outflow), and any date range or frequency details. The more context you provide, the less back-and-forth.
7). Forecasting
Example Prompts
"Using the data streams I just built, make me a forecast that includes these streams and tell me my forecasted cash balance as of 12/31/2026."
"What will my total cash position look like over the next 90 days?”
Show me the variance between my Payroll forecast and actuals for this month."
"Which of my regional forecasts have the largest gap against actuals this quarter?"
Best Practices
Combine the AI Chat's ability to create tags, data streams, and forecasts into an efficient forecast creator workflow.
If your forecast looks off, you can ask the AI to investigate before adjusting manually — try "Why does my AR stream show a large variance this month?" or "Are any of my streams in my forecast that are missing actuals data?" The AI can often identify whether the issue is a tagging gap, a missing stream, or a data feed problem before you escalate.
Keep your tags tight. Forecast accuracy often depends on how well your tags capture historical actuals — if material transaction activity is falling outside of your streams, that's a signal your tag TQL filters need refinement.
8). Data Integrity & Account Reconciliation
Example Prompts
"Show me why my Collections tag isn't capturing XYZ transactions."
"Flag any transactions in the last 30 days that are missing a tag."
"Show me accounts that haven't reported a balance in the past 5 business days."
"Are there any duplicate transactions across my bank accounts this month?"
"Which of my tags haven't matched any transactions in the last 60 days?"
Best Practices
If you see missing or unexpected data, ask the AI to investigate before escalating. Check your bank statement to confirm if data is in fact missing, then send us a message and we'll work to resolve with your bank.
Use results to tighten your tag rules. If the same transactions keep showing up untagged, that's a signal your TQL filters need refinement.
Embedded AI Insights
Trovata AI isn't limited to the chat interface. This means you don't have to leave your workflow to get answers, as contextual AI insights are embedded directly into the following modules:
Balances (Account Detail) — Drill into individual account behavior with AI-assisted context
Statements — Flag unusual activity or summarize statement-level detail on demand
Analysis & Reports — Get instant interpretation of what the data is telling you, and surface trends and anomalies without leaving your report view.
Data Streams — Validate incoming data and spot irregularities in real time
Forecasts — Interrogate forecast assumptions and variance against actuals
Home — These agents provide domain-specific analysis without you needing to know the exact right question to ask
Data Integrity: Surface gaps, anomalies, or inconsistencies in your account data.
Transaction Health: Assess the quality and patterns of your transaction flow.
Transaction Categories: Break down and analyse transactions by category.
Currency Exposure: Understand your multi-currency positions and FX risk.
Idle Cash: Identify cash sitting in low-yield or dormant accounts.
Note: The main Trovata AI Chat and each insight agent operate as separate sessions with their own conversation context.
AI Chat Limitations
Trovata AI Chat cannot perform the following actions:
Override FX Rates
Create/Edit/Delete Dashboards
Create/Edit/Delete Entities
Map Entity Details
Create/Edit/Delete Manual Accounts
Map Account Details (Account Alias, Type, or Legacy Groups)
Create/Edit/Delete Groups
Create/Edit/Delete Statement Reports
Create/Edit/Delete Cash Positions
Upload/Delete Manual Transactions
Upload/Edit/Delete G/L Codes
Create/Edit/Delete Reconciliation Reports
Create/Edit/Delete Forecast or Data Stream Factors
Create/Edit/Delete Workbooks
Establish Connectivity via MultiBank Connect
Initiate/Approve/Release Payments
Create/Edit/Delete Payment Templates
Create/Edit/Delete Payment Workflows
Investments Module
Create/Edit/Delete Developer Portal Applications
Create/Edit/Delete Notifications
Create/Edit/Delete User Access
Create/Edit/Delete User Group Permissions
FAQs
What's the difference between Trovata AI Chat and Agents? Trovata AI Chat is mainly used for interactive, one-time questions or resource creation (i.e. tag building) where you ask and get answers in real time. Agents automate those same kinds of questions on a schedule. Think of Chat as asking a question and Agents as assigning a recurring task.
Where does processing happen? Everything runs on AWS Bedrock AgentCore, a secure, isolated AI environment built for enterprise workloads. Your data is processed within this infrastructure and never leaves it.
What retrieves your data? Trovata's own verified API layer, the same infrastructure that powers every other feature in the platform. The AI coordinates the analysis; Trovata's tools handle the data retrieval.
Does Anthropic have access to my data? No. Claude Sonnet 4.6 powers the AI, but Anthropic as a company has zero access to anything processed in your sessions. That's a core guarantee of the AWS Bedrock architecture.
Is my data used to train the AI? Never. AWS Bedrock ensures complete isolation between your sessions and model training. Your inputs and outputs are not retained, logged, or used to improve the model.
Is anything stored after my session? No. Your financial data isn't persisted, cached, or logged outside your Trovata instance.
Who can see my conversations? Only you. Chat sessions are private to your user account. No one else in your organization has visibility into your history.
What happens when something fails? The AI tells you. If a query returns no results or a tool call doesn't work, you'll get a clear explanation.
What does "may produce inaccurate information" mean? The data itself comes from verified API calls and is reliable. The disclaimer refers to the AI's natural-language reasoning and interpretation, which may occasionally contain errors. The numbers are real; the narrative is the AI's best interpretation.
Why does the response appear word by word? That's streaming, and it's by design. Instead of waiting for the full response to generate, Trovata AI Chat shows the answer in real time so you can start reading immediately.
Can I export the data? Yes. Tables export to CSV or XLSX. Some charts can be exported as a PDF. You can also copy values directly from any table.
Can I go back to earlier messages? Yes. Scroll up to see your full conversation history. The AI retains context from earlier messages, so follow-up questions work without re-explaining.
How do I switch between the main chat and insight agents? Use the back arrow in the header to return to the main Insights Dashboard. The main chat and insight agents maintain separate sessions.
What if the AI is taking a long time to respond? Complex questions that require multiple tool calls naturally take longer. Watch the progress messages to see which step the AI is on. If a response seems stuck, refreshing the page will reset the session.
What if I get an error response? This can happen when data is unavailable, a date range is invalid, or there's a temporary service issue. Try rephrasing with a different time window or more specific filters. If errors persist, contact Trovata support.
Why can't I see Trovata AI in my sidebar? Two things need to be in place: the Trovata AI feature must be enabled for your organisation, and your user role needs the Read Trovata AI permission. Check with your Trovata administrator.
I've used Trovata's AI features before... Is Trovata AI 2.0 included in my subscription? The AI experience you used previously was Trovata AI 1.0. With the launch of Trovata AI 2.0, AI Insights, AI Chat, and AI Agents are now paid premium features. All existing customers can try Trovata AI 2.0 for free during the trial period — no action needed to get started. After the trial ends, reach out to your Customer Success Manager if you want to add AI 2.0 to your plan.
How much does Trovata AI Chat cost? Consult your CSM to see if it's included in your subscription.