All case studies

Validating Conversational Analytics, Query by Query

How 1,140+ manual test cases across three database engines, AI-generated SQL, and enterprise-grade security gave a natural-language analytics platform the QA rigour mission-critical decisions demand.

AI-POWERED ANALYTICS MULTI-DATABASE VALIDATION NATURAL LANGUAGE QUERYING   RBAC & SECURITY TESTING   DASHBOARD QA
Validating Conversational Analytics, Query by Query
OVERVIEW

Quality assurance for a natural-language enterprise analytics platform.

The client's platform is an AI-powered enterprise analytics tool that lets business users interact with structured data using natural language — eliminating the need for SQL knowledge or dependency on analytics teams to answer everyday business questions. The platform allows users to connect multiple types of data sources and simply ask questions in plain English. The AI automatically interprets intent, generates optimised SQL, executes it across the connected database, produces business-friendly summaries, recommends the best visualisation, builds interactive dashboards, and enforces enterprise-grade security and role-based access — all within seconds. This lets both technical and non-technical users explore data independently, replacing scheduled reports and dashboard-building bottlenecks with real-time, conversational access to enterprise data.

Proving that conversational analytics can be trusted with mission-critical business decisions.

1,140+

FUNCTIONAL TEST CASES EXECUTED

3

ENTERPRISE DATABASES VALIDATED

447

CROSS-DATABASE TEST CASES

239

BILLING & SUBSCRIPTION CASES

PROJECT AT A GLANCE
Client
Enterprise Analytics / AI SaaS (name withheld at client’s request)
Website
Confidential (available on request)
Industry
Enterprise Business Intelligence / AI SaaS
Project Type
Manual QA — Multi-Database AI Analytics Platform
Databases Supported
PostgreSQL, Microsoft SQL Server, ClickHouse
QA Scope
AI query validation, multi-database validation, dashboard & collaboration testing, RBAC & security, billing validation
Functional Test Cases
1,140+ across all major product areas
THE CHALLENGE

Proving an AI can be trusted with the query, every time.

Organisations struggle to get fast, reliable insights from structured data spread across multiple systems — dependent on analysts, SQL expertise, and slow, inconsistent reporting. The platform's promise was to remove all of that friction; the QA challenge was proving it could do so safely, accurately, and securely across every supported database.

01

Multi-Database Consistency

The platform had to generate accurate SQL and return correct results regardless of whether it was connected to PostgreSQL, Microsoft SQL Server, or ClickHouse — each with its own SQL dialect and behaviour.

02

AI Accuracy at Scale

Natural language prompts spanning direct queries, aggregations, multi-table joins, filters, and time-based and multi-level aggregations all had to produce correct SQL and correct business insights, not just plausible-looking answers.

03

Enterprise Security & Access Control

Organisation isolation, role-based access control, and table- and column-level permissions all had to hold up under testing to ensure users could only ever reach authorised data.

04

Dashboard & Collaboration Complexity

Dashboard and widget creation, sharing, permissions, and access management all needed validation to ensure collaboration features didn't create unintended data exposure.

05

Usage-Based Billing Accuracy

A credit-based AI usage model — seat-based credits, workspace top-up pools, organisation usage analytics, and billing workflows — required careful validation to ensure customers were billed accurately for what they used.

WHAT WE DID

Testing the AI, the data layer, and the access model — together.

A comprehensive manual testing strategy was executed across every supported database platform, AI workflow, dashboard capability, and security feature — resulting in 1,140+ functional test cases across all major product areas.

01

Multi-Database Validation

Validated database onboarding and connectivity, schema discovery, metadata generation, table and column validation, SQL generation accuracy across dialects, data retrieval consistency, and error handling across PostgreSQL, SQL Server, and ClickHouse.

02

AI Query Validation

Executed a large set of natural language prompts covering direct and aggregate queries, single- and multi-table joins, search, sorting, filtering, and time-based and multi-level aggregations — validating AI-generated SQL and business insights across tables and columns.

03

AI Response Validation

Validated every generated response for SQL correctness, result accuracy, business summaries, visualisation recommendations, chart rendering, regeneration behaviour, query explanation, clarification handling, and error messaging.

04

Dashboard & Collaboration Testing

Validated dashboard and widget creation, deletion, copying, and moving, dashboard sharing and permissions, access management, and role-based access to individual widgets.

05

Security & RBAC Testing

Validated organisation isolation, role-based access control, table- and column-level permissions, shared dashboard permissions, and user invitation and access flows.

06

Billing & Usage Validation

Validated the AI credit consumption model — seat-based credits, workspace top-up credits, organisation usage, member consumption, billing workflows, subscription management, and invoice generation.

07

Comprehensive UI & Functional Testing

Covered AI chat, dashboards, widgets, dashboard sharing, billing and subscription, team management, roles and permissions, audit logs, system prompts, feedback, the embeddable chatbot, and responsive UI validation.

Layer
Focus
Validated
Outcome
Data Layer
Multi-database validation
PostgreSQL, SQL Server, ClickHouse
Consistent AI behaviour across all supported database engines
AI Layer
Query & response validation
SQL generation, result accuracy, summaries, visual recommendations
Reliable, business-correct outputs across query types
Collaboration Layer
Dashboards & widgets
Creation, sharing, permissions, access management
Secure, reliable dashboard collaboration
Access Layer
Security & RBAC
Organisation isolation, table/column permissions
Users confirmed to only access authorised data

1,140+

functional test cases executed across all major product areas

from AI chat and dashboards to billing, security, and multi-database connectivity.

447

cross-database test cases validated consistent AI and query behaviour across ClickHouse, PostgreSQL, and Microsoft SQL Server simultaneously.

239

billing and subscription test cases confirmed accurate credit consumption, workspace top-ups, usage analytics, and invoice generation.

200

user invitation test cases and 10 RBAC-specific test cases confirmed organisation isolation and role-based access held up as intended.

61

test cases validated domain-specific reporting scenarios, confirming the platform's natural-language querying held up on real-world, sector-specific data.

47

test cases validated the embeddable chatbot experience, and 21 validated AI-generated starter questions — ensuring first-touch AI interactions were accurate and reliable.

Faster decision-making, reduced dependency on technical teams, improved data accessibility, and increased adoption of self-service analytics — with QA validating cross-database compatibility, AI response accuracy, dashboard collaboration, and enterprise security ahead of production readiness.

TOOLS & TECHNOLOGY

The stack.

Databases Supported
PostgreSQL, Microsoft SQL Server, ClickHouse
AI Validation
Natural language query accuracy, SQL generation correctness, business summary validation
Visualisation Testing
Tables, bar/line/area/scatter/pie/donut charts, dashboard widgets
Security Testing
Organisation isolation, RBAC, table- and column-level permissions
Billing Validation
Seat-based credits, workspace top-ups, usage analytics, invoicing
QA Approach
Comprehensive manual testing across AI, data, dashboard, and security layers

Beyond validating what the platform does, the QA process proved the rigour behind it — cross-database compatibility, AI response accuracy, dashboard collaboration, and enterprise security all held up under 1,140+ test cases. That combination of breadth and depth significantly reduced deployment risk and gave the organisation confidence to adopt conversational analytics for decisions that matter.