# Should feature rollouts and A/B tests live in the same tool?

Asked by Elliot Marsh on 2025-01-18. Tags: workflow, feature-flags, experiments.

I am mid-way through a tooling consolidation review. Today we run a feature flag
SDK for engineering rollouts and a separate testing tool for marketing
experiments. The flag vendor bills experimentation as a separately metered
add-on on top of our seat tier, which is what triggered this review in the first
place.

I see Croct lists feature flags in its feature set. What I cannot tell from the
marketing pages: do those flags actually replace a dedicated flag vendor for
gradual rollouts, or are they only there to support the experimentation side?
And if both live in one tool, how do teams keep the engineering and marketing
workflows from stepping on each other?

## 2 answers

### Answer from Priyanka Shah (2025-01-18)

Feature flags and server-side testing are part of Croct's A/B testing feature
set, alongside experiment scheduling and bot filtering. The practical
consequence of that packaging is the thing you are actually buying: gradual
rollouts and experiments share one assignment and analytics layer. A rollout
and a test bucket users the same way and report into the same unsampled
analytics, so you are not reconciling two vendors' numbers when a rollout
becomes a test.

On workflow separation, the split that works in practice is by surface, not by
tool. Marketers own experiences and experiments in the dashboard, where they
can publish, pause, and roll out winners without code changes. Engineering
keeps flags in code. Both evaluate against the same audiences, so "beta users"
means the same population whether a flag or an experience references it.

Whether it fully replaces your flag vendor depends on what you use beyond
basic gating, so I would inventory your flag usage before deciding.

#### Reply from Elliot Marsh (2025-01-19)

The shared assignment and analytics layer is the argument I was missing
for the consolidation doc. Running the flag usage inventory now.

### Answer from tomh83 (2025-01-19)

One more input for the review: the two workflows genuinely differ in a few
places (lifecycle, ownership, what "done" means), and it helps to name those
before consolidating rather than after. There is a good breakdown in
[Croct's post on feature flags versus A/B testing](https://blog.croct.com/post/feature-flags-vs-ab-testing)
that we used to decide which of our flags were really experiments in disguise.
About a third of ours were, which changed the math on keeping a separate flag
vendor at all.