Trovata AI

Large language model chat assistant that can answer complex questions about your cash data

Joseph Drambarean avatar
Written by Joseph Drambarean
Updated over a week ago

We are so thrilled to be launching the world's first completely secure and fully private generative AI chat assistant for corporate finance teams to use with their bank data.

As this technology is continuing to evolve quickly we wanted to take a moment to give you some answers to commonly asked questions about this technology and the experience you can expect within Trovata.

Is my data being sent anywhere outside of Trovata?

No. Your Trovata data is securely stored within your own private instance. That data is never shared with third parties in any form.

How does the AI come up with the answers?

Trovata AI uses the large language model capabilities of chatGPT to help infer the intent of your question and come up with an approach to answering it that can be described in a SQL statement. This SQL statement is then used by the powerful analytics capabilities of the Trovata platform to search for the answer to your question in your data privately. The resulting data is never transmitted to chatGPT or OpenAI in any way.

Are the answers accurate?

The mathematical computations that make up an answer to any question that you ask are being performed by the Trovata platform and can be trusted to be accurate using your data. The methodology that is being utilized to generate the answer, however, is being identified through the large language model capabilities of chatGPT and being translated into a machine readable SQL statement that can be used by our platform. If you are ever in doubt of the answer, you can always see the SQL that was being used to identify the answer to your question and if it is not the method you expected you can ask the question in a different way or ask the chat to modify the approach to calculating the answer.

What kinds of questions can I ask the Trovata AI chat?

Trovata AI can leverage the data stored in your instance for balances and transactions at this moment and time and is being given the context to approach problem solving your questions from the perspective of a corporate treasurer. As such the large language model is tuned to answer topics from areas such as cash flow, analysis, forecasting, statistics, strategies, and can also provide fine grained answers to questions about the specifics of your transactions.

Example questions can include:

  • Show a year over year comparison in a table of my total ending cash balance each month for 2022 and 2023

  • Can you give me a report of my daily spend in USD by currency for the last week?

  • Lets dig a little deeper, can you show me the daily historical exchange rates for all currencies for the last week?

  • Show me the 5 largest sources of revenue for the last month in a table

  • Give me a report of our monthly cashflow for the last year and plot it

  • I am trying to find the most cyclical transactions over the last 90 days, can you put them in a table for me?

  • Can you show me my average cash burn over the last 30 days plotted as a line graph?

  • Can you show me my cash allocations across all of my banks as a percentage of overall cash in a table?

  • Can you produce an FBAR report showing my highest balance across all accounts over the last year?

  • Can you produce a projected cash forecast based on the average cash flow across all of my accounts that provides daily expected cash for the next 7 days and plot it in a line graph?

  • How do I go about hedging my international balances from FX exposure?

  • Can you explain the technique of hedging currency exposure?

  • Can you make a bar graph showing this same information?

How do I know if the answers are correct?

The data that is being presented in any visualization within Trovata AI can be inspected in its table form with exact values as well as the SQL statement that was leveraged to produce the data set. If there is an inaccuracy in either of those two you can make modifications to the approach by asking the chat further questions or clarifying your request.

Does the system learn from my questions?

While Trovata AI can remember the context of the questions that you ask during a particular session it cannot learn outside of that session. No data is being communicated about your session outside of Trovata, and the information that is being processed during your session never leaves your Trovata instance. As a result the large language model that chatGPT is using cannot learn from your session or be influenced in any way by your data.

Why is the Trovata AI feature labeled as a Beta feature? Can I actually trust it?

While the promise of generative AI brings with it the possibility of endless creative approaches to solving a myriad of different challenges, the design approaches best utilized in helping the user navigate these problems will inevitably change. As a result of this Trovata AI is currently being marked as in the beta phase from a product development perspective. As more and more customers continue to use this breakthrough technology we will learn more about the edge cases that are unearthed as a result of this use and continue to refine the experience accordingly.

What should I do if I see something that is completely wrong or is not working?

While this feature is in beta we appreciate any and all feedback on the usefulness and accuracy of the answers provided by Trovata AI. Each answer gives you an opportunity to vote with a thumbs up or down with regards to the answers and when you provide this feedback you can add additional notes that will be read in detail by our product and engineering teams to look into your case further.

How much does the Trovata AI feature cost?

The Trovata AI feature is available as part of your subscription package and is available to use in an unlimited fashion. No further fees apply.

Did this answer your question?