Deciding What to Automate Next
On this page
Every manual check you repeat costs the same each time. Automating it costs more once and little after that, as long as the test stays reliable. Manual testers are in the best position to decide what is worth automating, because they know which checks they run every release, which ones find bugs, and which ones are tedious enough that people start skipping them.
Good candidates#
A check is a strong automation candidate when most of these are true:
- It is repeated. Run every release, or after every change to an area. A check you run once is cheaper done by hand.
- The expected result is unambiguous. "Total is 45.00", "the order appears in My orders with status Paid". If two testers might disagree on the verdict, a machine will not do better.
- The flow is stable. The feature has stopped changing week to week. Automating a feature in its first sprint means rewriting the test with every iteration.
- Failure is expensive. Log in, sign up, checkout, payment, the core action your product exists for. A broken checkout costs money every minute; a misaligned footer does not.
- It is tedious or error-prone by hand. The same form with 20 data combinations, the same flow on six browsers, checks that need exact arithmetic.
- It has caught a regression before. The escaped-bug lines you added to your checklists (chapter 2) are prime candidates: proven risk, precise expected result.
Poor candidates#
Keep these manual, at least for now:
- New and changing features still being designed. Explore them; automate once they settle.
- Judgement calls: usability, whether copy is clear, whether an animation feels right, whether a layout looks broken in a way nobody has defined. Visual comparison tools can catch changes in screenshots, but someone still decides whether a change is a bug.
- One-off checks such as a data migration verified once.
- Things that are very hard to drive automatically, like the interruptions that need a phone in your hand from the previous chapter. The cost of a fragile test can exceed the cost of a quick manual check each release.
Which of these is the best next automation candidate?
The coupon check is repeated, has an exact expected result, covers a flow that costs money when it breaks, and has already caught a real bug. The onboarding flow is still changing, so a test would need rewriting constantly; the pricing page check is a judgement only a person can make.
A simple scoring sheet#
When the list of candidates is longer than the automation team's time, score each one from 1 to 3 on four questions and automate the highest totals first:
| Question | 1 | 3 |
|---|---|---|
| How often do we run it? | Rarely | Every release or more |
| How bad is a failure in production? | Cosmetic | Lost revenue or data |
| How stable is the feature? | Changing every sprint | Unchanged for months |
| How costly is it by hand? | Two minutes, one browser | Many data sets or browsers |
This is not science; its value is that it forces the conversation into the open, so a tester and a developer can agree on the order in ten minutes instead of arguing from instinct.
Handing a check over for automation#
The person writing the automated test may not know the feature as well as you do. Give them what a good bug report gives a developer:
- The test case, written as in chapter 2: preconditions, data, intent-level steps, precise expected results. Rewrite vague checklist lines into this form.
- The test data and how to create it, including accounts, products and coupons, and what must be cleaned up afterwards.
- The environments it must run on, from your browser and device matrix.
- The edge cases you found while testing manually: the boundary values, the state transitions that should be refused. These are often worth more automated assertions than the happy path.
- What is out of scope, so the test does not try to verify things that need judgement.
Once the automated test is running reliably, remove the line from your manual checklist. The point of automation is to free your time for the exploratory and judgement work that machines cannot do.
The shipping-threshold check is now automated and has been green and reliable for a month. What happens to the line on your manual checklist?
Running a reliable automated check by hand as well doubles the cost for no new information. If you doubt the automated test, review what it asserts once; then use the freed time where nobody is looking yet.
Where to go next#
If you want to write the automation yourself, these free courses pick up where this one ends:
- Cloud Testing Fundamentals: how automated tests drive real browsers on a cloud grid, and your first test.
- Playwright Tutorial: a modern web automation framework in TypeScript, from first test to CI.
- Selenium Python Tutorial and Selenium with Java and TestNG: WebDriver in Python or Java.
- Maestro Mobile Testing: mobile app flows written in YAML, a gentle start for testers new to code.
- Visual Regression Testing: screenshot comparison, for the layout checks you currently do by eye.