I would like to put two key lessons that I learned in these years that you can use to make most of your money you are putting into automation
If a test (case) can be specified like a rule - that MUST be automated
Automation code is software - thus, obviously is built on some kind of specification. Most GUI automation (QTP, Selenium) is typically built based on so called "test cases" written in human language (say English). It is the first question that a automation guy will ask while starting automation - "where are the test cases?". In dev world - automation takes a different meaning. In TDD style automation (if you call TDD tests as automation) - test is itself a specification. A product requirement is expressed as a failing test to start with. The approach of BDD throws this context to other boundary, specify tests in the form of expected behavior. So, automated tests are based on specification that is a human language but expressed in business terms (mainly) and with a fixed format (Given-when-then).
Key lesson here is - if a test can be specified like a rule with a clearly defined inference to be drawn from the test - that should be automated. Automating a test means create a program to configure, exercise and infer results of what test is trying to validate. Michael Bolton calls such a test as a check - a meaningful distinction. If a test has human element in it for inference mostly - you cannot possible automate the test in its full form.
How do you implement this lesson in your daily life as tester? When designing a test - see if you can specify it like a rule. If you can then explore ways to write a program for it. Then that test becomes automated. In this way when you are building a suite of tests - some are specified like a way that makes it easy to automate and some are specified in a way that a human tester need to apply her intelligence to exercise and infer.
Automated tests (checks) are like guard to product code
A child asks his father "what is the use of brake in a car". "it helps to stop the car" says father. Kid responds back "no.. I guess break helps driver to drive the car as fast he wants to as he has a means to to stop when needed". On the similar lines - having automated tests around a piece of code - literally guarding the code - empowers the developer to make changes to the code faster. More often than not - bigger speed breakers for development is fear of breaking some working code. Developers are mostly worried about large chunk of legacy code that one rarely understands fully. Having automated test as guard - what happens is test will flag change in the code via failing test. Armed with support of guarded code - developers can now make changes faster and can depend on tests to tell them if any of change made has broken some other "working" code.
How do you implement this lesson? Work with developers and help them creating tests that guard their code. These tests should work like "change detectors". Writing test automation would require knowledge of product code and principles of unit testing. Not for weak hearted GUI QTP/Selenium folks.