Testability - Building in Lower QA Costs

Currently working on developing a Test Framework for a third-party's website and it is an excellent example of how involving testers early in a project could result in long-term cost savings and/or higher quality.

The website has a Menu navigation component on most of the site's 130 pages. The menu provides the ability for users to select and browse to the required page. While familiarising myself with the website I noticed that some menu items were missing on some pages and some menu items linked to the wrong pages. At this point I realised that I'd need to fully test the Menu component - IE. Ensure all menu items (130) were there and they all worked correctly (IE. Clicking on each one takes user to the correct page) - on every page. Potentially, over 15,000 tests (about 130 x 130)!

The menu system had been designed so that it does not use common code - or a single component - for the menu, it appears the code had been copy/pasted and possibly edited; hence introducing the errors. If a (good technical) tester had been used early in the design phase he/she would have highlighted that by not using a common menu component referenced by the pages, they would all need to be tested. Rather than just 260-300ish tests, over 15,000 tests - with the resulting cost implications (especially if the test-automation tool being used doesn't scale well). If a common component had been used, then the full set of items and their functionality would only have to have been tested on a single page and then just its presence and item-count on the other pages tested. As the menu component had separate code on every page that code has to be excersized fully and hence the high number of tests.

Luckily the test tool I'm using scales well, and going from the 130+ tests on the first page to 15K+ is relatively quick. However, with most tools this would be prohibitively expensive & time consuming (and not cost effective considering the low impact of defects the tests would find), which would mean that functionality would simply not be regression tested......