All projects

Automation at scale.

Conquering Salesforce login complexity. How a multi-layered automation architecture took the Cirrus Insight test suite from a 6-hour, flaky, lock-out-prone bottleneck to under 2 hours with near-zero flake rate — across Salesforce, Gmail, and Outlook simultaneously.

Playwright Parallelisation Claude-Assisted
Automation at scale. project cover
350+

Automated test cases

66%

Execution time saved

<2h

Full suite completion

~0%

Flaky test rate

Overview

Automation engineering for an integration-intensive SaaS platform.

Cirrus Insight is an AI-powered sales productivity platform that operates directly inside Gmail, Outlook, and Microsoft Teams. It bridges the gap between email inboxes and Salesforce CRM, eliminating manual data entry and helping sales teams book more meetings, track buyer signals, and maintain a cleaner pipeline — all without leaving their inbox.

The platform serves over 50,000 customer-facing teams worldwide, including Walmart, Tableau, TripAdvisor, and Samsung. Its deep integrations with Salesforce, Google Workspace, and Microsoft 365 make it one of the most integration-intensive SaaS products to test and automate — and the reason a conventional Playwright setup was unworkable from the start.

Project at a Glance
Client
Engagement
Automation Engineering
Automation
Playwright (TypeScript)
Environments
Salesforce u00b7 Gmail u00b7 Outlook
Industry
SaaS / CRM Productivity
Project Type
Automation Engineering
Onboarded
November 2025
Automation Framework
Playwright (TypeScript)
AI Coding Tool
Claude
Integrations Under Test
Salesforce u00b7 Gmail u00b7 Outlook
Version Control
GitHub
The Challenge

A test suite that couldn’t get past login.

Automating QE for Cirrus Insight meant testing an application deeply entangled with Salesforce, Gmail, and Outlook — platforms with strict, automation-resistant security policies. The team encountered a unique combination of challenges that made a conventional Playwright setup unworkable from the start. The brief: turn an unstable, 6-hour, login-blocked test suite into a fast, reliable automation foundation that could confidently support weekly production releases.

01

Salesforce auto-lock

Salesforce locks accounts after repeated or failed login attempts. With multiple parallel test scripts attempting authentication simultaneously, account lock-outs were constant u2014 halting entire test runs.

02

Multi-platform authentication

The app requires concurrent authentication across Salesforce, Gmail, and Outlook browser extensions. Each uses different OAuth flows and session behaviours, making a unified auth strategy extremely complex.

03

6-hour execution time

The initial full-suite run took over 6 hours due to sequential execution and repeated login cycles. Too slow for a continuous-delivery model with weekly production releases.

04

Widespread test flakiness

Login failures and inconsistent extension loading caused pervasive flakiness u2014 tests failed intermittently without any code change, eroding team confidence in the whole suite.

What We Did

The multi-layered automation architecture.

01

Shared session storage

Authentication tokens and cookies are established once and shared across all Playwright test files. This eliminated the root cause of Salesforce lock-outs u2014 repeated logins per test u2014 without sacrificing test isolation.

02

Project-level cookie architecture

Native Salesforce login bugs were fixed and a project-level cookie management system built, letting authentication state persist across runs. Five framework fixes were merged to the development branch.

03

Parallelised test execution

Project-level parallel execution was enabled for the entire suite. Specs that previously ran sequentially over 6 hours were restructured to run in parallel u2014 cutting total execution time to under 2 hours.

04

2FA and OAuth automation

Salesforce 2FA was implemented across all scripts to handle the computer-activation flows triggered by automated logins. Gmail and Outlook flows u2014 including the full Forgot Password journey u2014 were fully automated with ESLint compliance.

05

Claude-assisted development

Claude accelerated code generation, suggested refactors, and helped enforce ESLint compliance u2014 significantly reducing time-to-automation for each new test case and improving code quality across the suite.

06

Production regression and smoke suites

Structured manual production regression delivered for every weekly release, plus a dedicated smoke suite for rapid post-deployment validation u2014 keeping production stable after each release.

Engagement Journey

From blocked login to weekly releases.

The engagement progressed through three clear phases — first unblocking authentication, then restructuring for speed and stability, then locking in reliable coverage for every weekly production release.

Phase
Focus
Automation State
Coverage Scope
Foundation
Shared session storage and project-level cookie architecture to kill Salesforce lock-outs.
Auth unblocked
Core login flows
Scale & Stabilise
Parallelised execution, 2FA + OAuth automation, and Claude-assisted refactoring.
Parallel, near-zero flake
350+ cases, 3 platforms
Release Cadence
Weekly production regression, smoke suite validation, and QE dashboard reporting.
Maintained
Full suite + smoke
Results & Impact

What the work delivered.

Salesforce account lock-outs eliminated

The core blocker that previously caused entire test runs to fail was resolved through shared session storage and project-level cookie management.

Full suite execution cut from 6h+ to under 2h

A 66%+ improvement through parallelisation u2014 enabling rapid CI/CD feedback aligned with weekly production releases.

350+ automated test cases passing consistently

Near-zero flaky rate provides a stable regression safety net for every release.

End-to-end multi-platform auth automated

Covering Salesforce 2FA, Gmail OAuth, Outlook OWA login, and Forgot Password flows across all environments.

Claude accelerated development velocity

Reduced time-to-automation per test case and improved code quality through AI-assisted review and refactoring.

QE dashboard delivered early

Real-time visibility into manual vs. automated coverage, pending backlog, and non-automatable scenarios.

Tools & Technology

The stack.

Automation
Playwright (TS) u2014 multi-layered architecture with shared sessions and parallel execution
AI Development
Claude u2014 accelerated code generation, refactoring, and ESLint compliance
Version Control
GitHub u2014 five framework fixes merged to the development branch
Project Management
Jira u2014 backlog, defect, and coverage tracking
Platforms Under Test
Salesforce u00b7 Gmail u00b7 Outlook / OWA
Reporting
Custom QE Dashboard u2014 manual vs. automated coverage and backlog visibility
The Takeaway

The Cirrus Insight engagement demonstrates that automation for deeply integrated SaaS products requires architectural thinking, not just script writing. By solving the Salesforce authentication problem at the session level and restructuring execution for parallelism, the team transformed an unstable, six-hour test suite into a fast, reliable automation foundation that confidently supports every weekly production release.