FETech Analytics

1. Executive Summary

FETech Analytics is a full-stack Business Intelligence web application built to serve retail businesses that need to understand their sales performance without hiring a data analyst or purchasing expensive BI tools. The platform ingests raw order-level CSV data and transforms it into a structured, interactive dashboard with automatic insights, revenue forecasting, customer segmentation, and an AI-powered assistant.

The core design principle is that charts alone are not intelligence. Any spreadsheet can produce a chart. The differentiating feature of FETech Analytics is the Insight Engine — a rules-based analysis layer that reads the data, identifies patterns against industry benchmarks, and outputs structured recommendations with specific actions the business should take. This transforms the platform from a data viewer into a decision support system.

The platform is fully deployed on Vercel and publicly accessible. It operates in a graceful degradation model: it works completely out of the box with demo data, and progressively unlocks features (authentication, cloud persistence, AI chat) as environment variables are configured.

2. Problem Statement

Small and mid-size retail businesses generate significant transaction data but lack the tools or expertise to extract meaningful insights from it. Enterprise BI tools (Tableau, Power BI, Looker) are expensive, complex, and require dedicated analysts. Spreadsheets are flexible but cannot generate forecasts, segment customers by behaviour, or automatically flag performance anomalies.

3. Technical Architecture

The application is a single Next.js 16 project using the App Router. All data processing happens on the client side after CSV upload — there is no proprietary backend; the Next.js API routes handle only authentication and database persistence, both of which are optional.

4. Application Pages

5. Key Features

6. Insight Engine Detail

The Insight Engine (src/lib/insightEngine.ts) is the intellectual core of the platform. It is a pure TypeScript module — no external dependencies, no ML model — that applies domain knowledge rules against the DashboardData object and returns structured Insight objects.

Each Insight has: a type (critical / warning / opportunity / positive), a title, a description explaining why the metric matters, a recommended action, and the key metric that triggered it. Insights are generated fresh on every render from live data — there is no pre-computed cache.

7. CSV Data Pipeline

The CSV parser (src/lib/csvParser.ts) accepts any order-level CSV file and produces a complete DashboardData object. It uses case-insensitive column matching with 20+ alias names per field, meaning a CSV with columns named Amount, Total, Revenue, or Price all resolve to the same revenue field.

From a single CSV of orders it computes: monthly revenue with YoY comparison (multi-year detection), daily sales rolling window, category/region/channel breakdowns, top products with trend analysis, RFM customer segments, monthly acquisition and churn, cohort retention heatmap, and a 24-month ETS forecast — all in the browser, in under one second for typical datasets.

8. Revenue Forecasting Model

The forecasting engine (src/lib/forecastEngine.ts) implements Exponential Smoothing (ETS) with multiplicative seasonal adjustment. It produces a 24-point series: 12 months of smoothed historical actuals plus 12 months of projection.

9. AI Assistant Integration

The AI assistant uses Claude Haiku (Anthropic) via the Vercel AI SDK v6. A floating chat widget sits on every page. On each message, the client serialises a compact snapshot of the current dashboard — revenue figures, top category, at-risk customer count, forecast accuracy, and other key metrics — and sends it as context alongside the conversation history.

This means the assistant can answer questions like "what should I do about my at-risk customers?" or "is my 8.3% return rate a problem?" with responses grounded in the user's actual data, not generic advice. The API route streams the response using streamText().toTextStreamResponse() and the client reads chunks incrementally via the Fetch ReadableStream API.

The assistant gracefully disables itself when ANTHROPIC_API_KEY is not configured, returning a clear 503 message rather than crashing.

10. Deployment & Infrastructure

The application is deployed on Vercel using Fluid Compute (Node.js runtime, not Edge). Static pages are pre-rendered at build time; API routes are server-rendered on demand. The platform uses three optional external services, each activating automatically when environment variables are present:

Data persistence uses a two-tier fallback: localStorage (always available, browser-local, zero accounts needed) and Neon cloud DB (cross-device, per-user, requires Clerk + Neon). On load, the app tries cloud first; if unavailable it reads from localStorage; if empty it shows demo data.

11. Project Structure

12. Conclusion

FETech Analytics demonstrates that a full BI platform — with forecasting, customer intelligence, automated insights, AI assistance, and PDF reporting — can be built as a single Next.js application with no proprietary backend infrastructure. The entire analytical pipeline runs on the client side, making the platform fast, portable, and free to host.

The key architectural decision that makes this project more than a dashboard is the Insight Engine: by explicitly separating data computation from insight generation, the platform can tell a business not just what happened but why it matters and what to do next. That is the difference between a chart viewer and a decision system.