Automated tests for online store checkout

Our store gets updated almost every week, and after nearly every update something quietly breaks on the way to the order. A promo code stops applying, the delivery cost disappears from the total, the payment step throws the customer back to an empty cart. We find out about it from buyers who write to support — which is the most expensive way to learn that checkout is broken, because for every person who writes there are several who simply leave. Manual checking after each change is not realistic: there is nobody on the team with a spare hour for it, and after the third release in a row the person checking stops looking carefully.

What I want is a set of automated scenarios that walk the same path a real customer walks. From the product card to a paid order: adding to the cart, changing quantity and removing an item, applying and cancelling a promo code, choosing between delivery methods and seeing the total recalculate, registration and login with a wrong password among the cases, and the personal account afterwards — order history, saved addresses, repeating a previous order. The mobile layout needs its own scenarios, not a resized copy of the desktop ones: most of our orders are placed from phones, and that is exactly where the cart usually breaks first. Everything must start by itself on every build, without anyone remembering to press a button.

The report matters as much as the tests. A manager should be able to read it without asking an engineer: which scenario failed, at which step, what was on the screen at that moment, and what the run looked like — a screenshot of the point of failure and a video recording of the failed test. I also need written instructions for our own developers, so that adding a new scenario later is a normal task and not an archaeology project.

Conditions I am stating upfront: tests run on the staging environment with a test payment key, real payments are never made; test orders must not reach the shop's accounting and reporting; a test that fails every other run is either fixed or deleted — an unreliable test is worse than no test at all, because people stop reading it.

What I expect to receive:
— end-to-end scenarios from product card to paid order, including promo codes and delivery options;
— registration, login and personal account scenarios;
— a separate set of scenarios for the mobile layout;
— automatic start of the whole suite on every build, with a clear pass or fail result;
— a report a non-engineer can read, with a screenshot of the failing step;
— video recording of every failed run, kept long enough to investigate it;
— written instructions for our developers on adding and maintaining scenarios;
— a short handover session so the team can run and extend the suite on their own.