Key takeaways
- Shift left: move testing earlier — reviews, examples, static analysis and fast automated tests — so bugs are prevented or caught when they're cheapest.
- Shift right: keep testing in and after release — staged rollouts, feature flags, monitoring and experiments — because some problems only appear with real users, data and scale.
- They're not alternatives. Shift left reduces how many problems reach production; shift right limits the damage of the ones that still do.
Where this picks up
You know that testing belongs in every stage of the life cycle. "Shift left" and "shift right" are names for two directions of that idea, on a timeline drawn from left (idea) to right (production).
Shift left: find it earlier
Practices that move testing towards the start:
| Practice | What it catches early |
|---|---|
| Reviewing requirements and designs | Ambiguities, contradictions, missing cases |
| Three amigos and example mapping | Misunderstandings between business, development and testing |
| Acceptance criteria as examples (BDD) | Disagreement about what "done" means |
| Static analysis and linting | Whole classes of coding errors, before anything runs |
| Unit tests and test-driven development | Logic errors, minutes after they're written |
| Tests on every pull request | Regressions before code is merged |
| Contract tests between services | Broken integrations before deployment |
The tester's role changes here: less "testing at the end", more asking questions early, helping others test and designing tests before code exists.
Shift right: learn from production, safely
Some problems can't be found before release: real traffic patterns, real data, rare devices and combinations nobody predicted. Shift right means planning for that, instead of hoping.
| Practice | What it does |
|---|---|
| Canary releases | Release to a small group first (1%, then 10%, then everyone), watching for problems at each step |
| Feature flags | Deploy code switched off, then turn it on for some users; switch it off instantly if something goes wrong |
| Monitoring and alerts | Watch error rates, response times and business metrics (orders per minute) in real time. See Observability as a testing tool |
| Synthetic monitoring | Automated checks that run against production every few minutes, like a robot customer |
| A/B tests and beta programmes | Learn how real users respond before a full rollout |
| Chaos engineering | Deliberately cause failures, like stopping servers, to check the system recovers. Netflix's Chaos Monkey is the best-known example. See Chaos engineering |
Why you need both
On 19 July 2024, the security company CrowdStrike pushed a content update to its Falcon sensor software for Windows. The update contained a defect that its validation checks didn't detect. When computers loaded it, they crashed and kept crashing on restart.
The update went to all online Windows machines running the sensor within a short window. Microsoft estimated that about 8.5 million Windows devices were affected. Airlines, hospitals, banks and broadcasters were disrupted around the world.
In its review, CrowdStrike committed to more testing before release, including stress testing, fuzzing and rollback testing. It also committed to staggered deployment, starting with a small canary group, and to giving customers more control over when updates arrive.
Lesson: more testing before release (shift left) would have reduced the chance of the defect. A staged rollout (shift right) would have limited its reach to a small group, where monitoring could have caught it.
What testers do on the right
- Define what "healthy" looks like for a release: which metrics, which thresholds trigger a rollback.
- Write synthetic checks for critical journeys and run them against production.
- Watch the canary: compare errors and behaviour between the canary group and everyone else.
- Test the way back: rollbacks and feature flag switches need testing too. A rollback nobody has tried may not work.
- Feed production back into testing: every incident becomes a new test, a new risk in the register or a new monitor.
- Before release: agree on metrics, thresholds and the rollback plan.
- Release to a small group and watch the agreed metrics.
- Expand step by step only if the metrics stay healthy.
- Roll back fast if they don't, and investigate afterwards.
- Learn: add tests and monitors for what you found.
Do
Treat monitoring dashboards and alerts as part of the test plan. Review them before release, just like test cases.
Don't
Use "we'll monitor it in production" as a reason to skip testing before release. Shift right adds a safety net; it doesn't replace the earlier checks.
Check yourself
Your team wants to release a new recommendation engine. It can't be fully tested before release, because its quality depends on real user behaviour. What shift-right practices would you suggest?
Show answer
Release it behind a feature flag, first to a small canary group or as an A/B test against the old engine. Define metrics in advance: error rate, response time, click-through and orders. Monitor them, expand gradually if they're healthy and keep a tested way to switch it off instantly. Shift-left checks still apply too, such as unit tests and API tests for its logic.
