diff --git a/ADR/ADR027-scheduled-smoke-tests.md b/ADR/ADR027-scheduled-smoke-tests.md
new file mode 100644
index 000000000..7447dc0f6
--- /dev/null
+++ b/ADR/ADR027-scheduled-smoke-tests.md
@@ -0,0 +1,43 @@
+# ADR027: Scheduled Smoke Tests
+
+Date: 2024-02-22
+
+## Status
+
+Accepted
+
+## Context
+
+We want to run scheduled smoke tests that check critical user journeys like a
+user completing and submitting a form. If a smoke test fails then the team
+should be alerted through various channels depending on the required response
+(e.g. Slack or Pager Duty out of hours).
+
+## Decision
+
+1. Use Ruby and Capybara to write the test code to maintain consistency with
+   the end-to-end and application code already in use.
+2. Use AWS CodeBuild to run the tests from an image containing the smoke test
+   code along with Chrome and ChromeDriver; this is consistent with how the
+   end-to-end tests are run in our pipelines during deployment. AWS Synthetic
+   Canaries do not support Ruby at this time.
+3. Schedule the running of the CodeBuild project with an AWS Event Bridge
+   scheduled rule.
+4. Use AWS CloudWatch Metric Alarm to alarm when a smoke test fails or when
+   they stop running. The alarm will send to an appropriate SNS topic sending
+   to either Slack or PagerDuty as required.
+5. The smoke test infrastructure will exist in the same AWS account as the
+   environment it is testing. This aligns with the strategy of self-contained
+   accounts and environments.
+
+## Consequences
+
+1. Using Ruby and Capybara aligns with the existing team technology choices.
+2. The tests run on infrastructure provisioned by AWS. This should improve
+   reliability and reduce the maintenance burden over running on team
+   maintained virtual machines.
+3. Alerting remains consistent with existing solutions and can be modified to
+   alert the appropriate channel from Slack to PagerDuty.
+4. Engineers will need to access the production AWS account to begin debugging
+   smoke test failures in production however this access would be necessary to
+   debug the underlying cause anyway.