Embedding conversational AI into Tradeshift’s Reporting & Analytics App

Published on: 30 September 2025

What happens when AI, Data and Finance meet

What if you could ask your financial data a question in plain English (or any other language) and get an instant, intelligent answer? That vision is at the heart of our work at Tradeshift. Our previous reporting infrastructure, the Insight Center, was built on an outdated architecture that was difficult and costly to maintain. Its reliance on thousands of lines of SQL stored procedures made it brittle and difficult to scale, providing only static, downloadable reports with no interactive data visualization.

We envisioned a better way: what if financial insights could be as conversational and intuitive as asking a trusted colleague?

This vision was the driving force behind the Tradeshift Reporting & Analytics App, a solution built in close collaboration with Amazon Web Services (AWS) and powered by Amazon QuickSight and Amazon Q. We didn’t want to just create another reporting tool. Instead, our goal was to engineer a dynamic system where data, algorithms, and user interactions continuously refine the relevance and accuracy of financial insights, creating a powerful engine for discovery and decision-making.

✨ On June 10th 2025, I had the pleasure to present for the first time the Reporting & Analytics App together with the Amazon Web Services team at London Tech Week. 

Why we added conversational AI to our Reporting & Analytics App

Finance managers navigate a world of immense complexity, from intricate compliance regulations and supplier relationships to tight budget constraints and the relentless pressure to find new savings. While data is abundant, extracting meaningful value from it is often a monumental challenge. We saw an opportunity to bridge the gap between data accessibility and business decision-making.

Instead of spending hours digging through dashboards, we wanted finance leaders to simply ask their data questions in plain language. That’s why we integrated Amazon Q, the generative AI capability within the Analytics App.

Amazon Q transforms complex reporting into natural, conversational interactions, giving finance managers the power to:

• Understand spending trends instantly.
• Spot compliance gaps before they escalate.
• Benchmark performance across suppliers.
• Explore scenarios without waiting for analysts to build custom reports.

👉 Read more about our success story with Amazon QuickSight and Amazon Q.

The Reporting & Analytics App AI engine at work

At the core of the Reporting & Analytics App is a multi-layered AI engine that combines:

• Generative AI (GenAI) for Conversation: With Amazon Q, users can pose natural language questions that are automatically translated into queries against structured financial data.
• Semantic Layer for Finance: We built a domain-specific semantic model for finance and procurement. This ensures that when a finance manager asks about “late payments” or “supplier risk,” the AI interprets the meaning in their business context.
• Analytics Powered by Amazon QuickSight: Once a question is parsed, QuickSight generates the relevant visualization or dataset, surfacing insights through intuitive charts and dashboards.
• Feedback Loops for Continuous Learning: Every interaction feeds back into the system. The AI learns from corrections, clarifications, and frequently asked questions, steadily improving accuracy and relevance.

👉 Visit the product page to learn more about all the functionalities the Reporting & Analytics App provides to your team.

Reporting & Analytics App architecture: Where AWS meets Tradeshift

Designing this AI engine required bringing together AWS’s powerful analytics stack with Tradeshift’s expertise in B2B finance.

The architecture can be summarized in three layers:

1. Data layer: Transactional data from the Data Lake flows into Amazon S3. Because QuickSight does not natively ingest Parquet files from S3, the data is first loaded into Amazon Redshift for efficient querying.
2. Infrastructure layer: This layer, composed of a Node.js backend service and the QuickSight SDK, orchestrates user provisioning and generates embedding URLs. It leverages Amazon Q to provide the natural language understanding (NLU) and generative AI capabilities. A custom semantic layer ensures financial terminology is mapped correctly to data models.
3. Analytics & visualization layer: Amazon QuickSight delivers dynamic dashboards and interactive visualizations directly to the user’s browser via an iFrame. The React front-end uses the QuickSight Embedding SDK to deliver a seamless user experience.

This architecture was designed not just for accuracy, but also for scalability and security, ensuring that finance teams can unlock the flexibility of AI while keeping sensitive data governed

To make the user experience feel seamless, we leveraged several AI techniques:

• Natural Language Processing (NLP): Converts everyday finance questions into machine-readable queries.
• Entity recognition: Identifies suppliers, currencies, and categories mentioned in queries.
• Reinforcement learning from human feedback (RLHF): Every time users accept or refine an answer, the system adjusts to become more accurate over time.

In addition to the conversational AI in our Premium version, we also provide more advanced features, like:

• Customize reports and dashboards to align with your specific goals.
• Automate data analysis with generative AI to explore “what if” scenarios and get to answers faster.
• Transform complex data into clear, actionable stories, with AI-generated first drafts powered by Amazon Q.

A Multi-Layered Security Model

Ensuring proper access to customer data when embedding QuickSight dashboards is paramount. Our solution implements a robust three-layer security model to enforce tenant isolation, a fundamental principle for platforms that host embedded analytics.

1. User group-based access controls: Access to datasets, dashboards, and other resources is managed at the group level, not the individual user level, for efficiency and scalability. Dynamic user groups are provisioned automatically at the master-tenant level, simplifying management.
2. Row-level security (RLS): RLS ensures that individuals view only the data pertinent to them. Even though a single dashboard may be served to many users, each user will only see the subset of data they are authorized to access.
3. Runtime filtering: This dynamic filtering capability allows us to dynamically alter a dashboard’s view based on a user’s context, such as the company branch they are currently logged into. This ensures a personalized, context-sensitive dashboard experience without requiring multiple versions of the same dashboard.

Real impact: How finance teams use the Reporting & Analytics App

The true measure of any AI system is how it improves lives for its users. So far, finance managers using Tradeshift’s Reporting & Analytics App have seen:

• Time savings: Hours of manual reporting reduced to minutes with conversational queries.
• Smarter budgeting: Real-time insights that guide spend optimization.
• Empowered teams: Non-technical users can now explore data without waiting for analyst support.

The future of finance is conversational

We believe the future of finance technology won’t be built on static dashboards, but on conversational AI experiences. Just as smartphones replaced landlines and streaming replaced DVDs, conversational AI is set to redefine how finance managers interact with their data.

GenAI brings us closer to a world where:

• Finance leaders ask business questions in natural language.
• AI engines instantly translate those into insights.
• Continuous feedback ensures the system gets smarter every day.

At Tradeshift, our mission is to remove friction from finance. With the Reporting & Analytics App, powered by Amazon QuickSight and Amazon Q, we’re taking a major step in that direction.

The journey doesn’t stop here. Tradeshift is continuing to expand AI capabilities across our platform, helping finance leaders move from reactive reporting to proactive intelligence.

Curious to see how conversational AI can transform your finance function?