Testing microservices — how do you balance service isolation vs full integration tests?

We have 8 microservices. Testing strategy question: where do you draw the line between testing each service in isolation vs testing across service boundaries? Current state: - Each service has unit + integration tests (tests the service against its real DB, with other services mocked) - We have ~20 E2E tests that spin up all 8 services in Docker Compose and test full user journeys Problems: - Service mocks drift from real service behaviour - The 20 E2E tests take 18 minutes and are fragile Is contract testing (e.g. Pact) the answer here? Or is there a simpler strategy that doesn't require maintaining contracts for every service pair?