For many enterprise data teams, the most dangerous phrase in the office is "we built it in-house." What begins as a cost-saving measure often evolves into a technical debt trap, where the initial low entry price is slowly eclipsed by the mounting cost of maintenance and the frustration of scaling limits. At Tradeshift, an AI-driven platform for accounts payable and e-invoicing compliance, this reality manifested as a weekly ritual of manual labor. Customer success managers and data analysts found themselves trapped in a loop of downloading massive CSV files and wrestling with complex Excel macros just to assemble a basic report. The tools they relied on were no longer serving the business; the business was serving the tools.

The 10,000-Row Wall and the Migration Path

The breaking point arrived when Tradeshift's proprietary business intelligence tool hit a hard ceiling. The system could only fetch 10,000 rows per query, capped scheduled reports at 25MB, and limited data retention to a mere six months. For a company handling global financial transactions, these constraints made large-scale trend analysis and predictive modeling an impossibility. The internal friction was palpable, as the gap between the data available and the insights required by customers grew wider.

To break this deadlock, Tradeshift pivoted to an agentic AI framework powered by Amazon QuickSight. This was not a simple software swap but a strategic overhaul of how the organization interacts with data. The new platform introduced a suite of AI-native capabilities: chat agents that allow users to query data using natural language, Flows for automating multi-step workflows without writing a single line of code, and Research tools designed to synthesize deep-dive reports from disparate sources. This shift effectively democratized data, allowing finance professionals with zero technical background to perform complex statistical analyses that previously required a ticket to the BI team.

The transition followed a disciplined roadmap. A Proof of Concept began in early 2024, leading into the development of an embedded analytics MVP between August 2024 and March 2025. By June 2025, Tradeshift launched formal reporting and analytics applications for both buyers and sellers. By August 2025, the legacy internal tools were completely decommissioned, and the organization saw a staggering 98% adoption rate across the company.

From Cost Center to Revenue Engine: The 3-Tier Architecture

The real transformation, however, lay in the architectural shift from static dashboards to a tiered intelligence system. Tradeshift implemented a three-layer analysis environment. The first layer consists of 16 QuickSight dashboards delivered via iFrame, covering nine critical domains including invoice processing, purchase orders, network connectivity, user activity, compliance, and anomaly detection. This layer handles transaction volumes ranging from 1 million to 100 million records. The performance gain was immediate: query response times plummeted from a sluggish 45 to 90 seconds down to under 3 seconds.

Beyond simple visualization, the second layer introduces a conversational AI interface. Here, users bypass SQL entirely, asking questions in plain English to receive instant visual responses. The third layer introduces data autonomy based on user permissions. Standard tier users receive basic filters and CSV exports, while Premium tier users gain access to What-If modeling, allowing them to design their own dashboards and simulate various business scenarios.

Scaling this to a multi-tenant SaaS environment required a rigorous security stack to prevent data leakage. Tradeshift deployed a four-stage verification process: Okta SSO for identity management, custom namespaces for tenant isolation, session-limited Signed URLs for embedding access, and a massive array of approximately 14,000 Row-Level Security (RLS) rules to ensure users only see their own data. For those looking to implement similar isolation, the Support multi-tenant applications for SaaS environments using Amazon Quick Sight documentation provides the technical blueprint.

This architectural pivot turned a maintenance burden into a profit center. By productizing these analytics capabilities into a paid Premium reporting tier, Tradeshift expanded its Annual Recurring Revenue (ARR) by 2%. The impact on customer loyalty was equally significant, with accounts using embedded analytics showing a 10% higher 12-month retention rate than those who did not. Within one year, 50% of target enterprise buyers became Monthly Active Users (MAU) of the analytics suite.

The Quantifiable Impact of Agentic BI

The operational dividends of the migration were felt across every department. Internal support teams reclaimed 8.5 hours of manual reporting work per week, while external buyers saved between 6 to 8 hours weekly. The reliance on manual data manipulation via VLOOKUP and Excel macros dropped by 80%. Most critically, the time required to identify operational bottlenecks was reduced from a window of one to two days to less than 5 seconds.

From a financial perspective, the Total Cost of Ownership (TCO) decreased by 40%. Infrastructure costs fell by 35%, largely because Tradeshift shifted query processing away from the operational database and into SPICE, Amazon's in-memory data engine. This reduced the load on the primary DB and streamlined performance. Additionally, fragmented BI license costs were consolidated, resulting in a 30% saving. The human cost of maintenance also vanished, with the effort required to maintain the tools dropping to roughly 0.5 FTE, allowing those engineers to pivot toward core product innovation.

Speed to market for custom enterprise reports saw the most dramatic improvement. The deployment cycle for new reports, which previously took four to six weeks, was slashed to approximately one week—a 75% improvement in time-to-market. Internally, the utilization of analysis tools by non-technical staff doubled, and the volume of support tickets directed at the BI team dropped by 80%.

Tradeshift's journey proves that the value of modern BI is not found in the ability to store data, but in the speed at which that data becomes an actionable decision. By leveraging an in-memory engine and a tiered agentic architecture, they didn't just fix a slow tool—they transformed data access into a competitive product feature.