Tradeshift replaces legacy BI tool with Amazon Quick, cuts query times 30×

Tradeshift is an AI-powered accounts payable and e-invoicing compliance platform serving buyers and sellers in more than 70 countries, with a cloud-based network that processes millions of transactions. For several years, the company relied on a proprietary in-house BI tool, but as data volumes grew and customer expectations rose, the tool became a bottleneck. It supported a maximum of 10,000 rows per query, imposed a 25 MB ceiling on scheduled reports, retained only six months of historical data, and required approximately 50 percent of one full-time employee's maintenance effort. These constraints made large-scale trend analysis, anomaly detection, and predictive modeling impossible, and they locked external customers out of self-service access to their AP workflow data. Meanwhile, Tradeshift's Customer Success, Data and Analytics, and Commercial teams spent hours each week on manual reporting: exporting CSVs, running Excel macros, and assembling reports by hand.

Tradeshift evaluated several alternatives before selecting Amazon Quick, an agentic AI workspace that includes a chat agent for natural language Q&A over business data, Flows for automating multi-step workflows without coding, and Research for producing comprehensive analytical reports from multiple sources. The company deployed Amazon Quick in phases: a proof of concept in early 2024, an embedded analytics MVP from August 2024 through March 2025, and a full Reporting and Analytics app launch in June 2025. By August 2025, Quick had replaced the legacy BI tool entirely, achieving 98 percent adoption across the organization.

The deployment architecture served three layers of analytics capability. The first layer consisted of 16 embedded Amazon QuickSight BI dashboards delivered through secure iFrames, covering nine domains—document invoicing, purchase orders, scanning, network connections, user activity, workflow automation, compliance and anomaly detection, payment prediction, and goods receipt-invoice reconciliation. These dashboards process between one million and one hundred million transaction records and return results in under three seconds, compared to 45 to 90 seconds with the legacy tool. The second layer provided conversational AI access: users could ask natural language questions without SQL knowledge and receive instant visual responses. The third layer introduced tiered data autonomy: a Standard tier with pre-built dashboards and basic filters, and a Premium tier with Designer Mode, custom dashboard creation, what-if modeling, and access to agentic AI capabilities. Security was handled through four layers: Okta single sign-on for authentication, Amazon Quick custom namespaces to isolate each tenant, signed URLs for time-limited embedding, and row-level security (approximately 14,000 RLS rules) to filter data to each user's context.

The business impact was substantial. Tradeshift's internal accounts and support teams saved 8.5 hours per week that were previously spent on manual CSV reporting; external buyers saved 6 to 8 hours per week per user. Manual data manipulation—Excel macros, VLOOKUPs, pivot tables—decreased by 80 percent. The time required to identify operational bottlenecks dropped from one to two days to under five seconds. From a cost perspective, the company achieved a 40 percent reduction in total cost of ownership, a 35 percent reduction in infrastructure costs by shifting queries to SPICE (QuickSight's in-memory data engine), and a 30 percent consolidation of licensing costs. Routine maintenance fell from 50 percent of one FTE to just 0.5 FTE. Tradeshift realized a 2 percent incremental ARR expansion from existing buyer accounts through the premium reporting tier, achieved a 75 percent improvement in time-to-market for custom enterprise reporting (new reports now deployed in approximately one week, down from four to six weeks), and saw accounts using embedded analytics show a 10 percent higher retention rate over 12 months. Within the first year, 50 percent of targeted enterprise buyers actively used the platform monthly, non-technical staff analytics utilization increased 2×, and analytics-related support tickets dropped 80 percent.

Two use cases exemplified the platform's impact. First, Tradeshift built a generative AI-powered AP Auditor chat agent using Amazon Quick's custom chat agent capability. Previously, AP auditors lacked self-service access to document data and had to rely on manual effort and cross-functional requests to BI teams, creating bottlenecks. The new agent, defined through natural language instructions without code, let users ask questions like "Show me pending invoices from last month" and receive immediate answers. The agent was grounded in a comprehensive knowledge architecture: 11 reference documents, 9 dashboards, 14 curated query topics, and 68 automated action tools accessible via the Model Context Protocol. Second, Tradeshift used Amazon Quick's SPICE data cache layer to build a high-volume reporting solution for a client that needed visibility into manual operational overhead. The solution processed large volumes of line-level coding data at speed, distinguished manual corrections from AI-generated codes, tracked approval hierarchies, and provided near real-time visibility to help spot bottlenecks.

Internally, Tradeshift migrated its legacy BI tool to QuickSight and collapsed 40 separate dashboards that users previously navigated into more than 100 queryable datasets accessible through natural language. Teams in Finance, Operations, Product, and Customer Success could now ask questions in plain language and receive answers instantly without SQL expertise. The quarterly market analysis process, which previously required multiple weeks of manual data gathering and interpretation, now completed in days or hours using Quick Research, which generated multi-source analysis from a single prompt. Recurring reporting flowed through automated Quick Flows, a no-code automation engine that orchestrated dataset refreshes, report generation, and distribution as scheduled pipelines. More than 270 SPICE datasets refreshed daily, enabling teams to focus on analysis and insight generation rather than data plumbing.

Tradeshift's migration reflects a broader shift from monolithic, code-heavy business intelligence tools to conversational, agentic AI platforms that lower barriers to data access. The legacy in-house tool, while appropriate for early-stage operations, became a drag on growth: its technical constraints (10,000-row limit, six-month retention) made large-scale analytics impossible, its maintenance burden consumed engineering capacity that could have fueled product innovation, and its lack of self-service interfaces forced customers and internal teams into dependency bottlenecks. Manual reporting—CSV exports, Excel macros, hand-assembled dashboards—consumed 8.5 hours per week for internal teams and 6–8 hours per week for each external customer, creating friction that delayed decision-making on both sides.

Amazon Quick's agentic capabilities—chat agents that interpret natural language, workflow automation (Flows) that orchestrates recurring tasks without code, and research engines (Research) that synthesize multi-source analysis—directly addressed these pain points. The platform's architecture also enabled Tradeshift to turn analytics into a customer-facing product: a tiered model (Standard for pre-built dashboards, Premium for custom creation and what-if modeling) converted what was once an internal cost center into a source of incremental annual recurring revenue (2 percent ARR expansion from existing buyer accounts in the first year). By August 2025, 98 percent of Tradeshift's organization actively used the platform, and accounts using embedded analytics showed 10 percent higher retention over 12 months.

No comments yet. Be the first to share your thoughts!