Skip to main content

Product Managers

What is the role of Product in experimentation?

Product Managers typically have the role of experiment owner in the experimentation workflow. Responsibilities include:

  • watching training for new experiment owners here - the slide deck in the video is here
  • starting and completing the experiment brief - which aligns everyone supporting the experiment to the learning goals.
  • align on the design of the experiment by attending a desktop or mobile office hour and filing a data org Jira ticket.
  • implementing the experiment in Nimbus console
  • coordinating with the QA team for sign-off, QA Jira ticket
  • launching the experiment
  • working with the experiment team to identify a set of reviewers trained and authorized to review experiments in your feature area. the experiment team will train them.

Where do I start?

  • Watch the 5 minute video on experiment workflow for an overview of the experiment lifecycle to understand the stages / ordering. Use the Workflow Miro board as your source for current links and guidance.
  • You must watch the training video before writing your first experiment. The training for new experiment owners is here - the slide deck in the video is here.
  • Ask any questions in ask-experimenter (you will have questions the first few times). Please ask for a shadow in #ask-experimenter you first time(s) through and someone will hop on zoom to write with you. We are happy to help people through the learning curve.
  • Start by opening the Experiment Brief link. If it is your first Experiment Brief - there is an optional 5 minute "how to video".
  • Fill out your experiment brief. As the Product Manager you are responsible for clearly articulating the product goals and questions that need to be answered. Capture as much as you can. It will improve and evolve as you have conversations with different subject matter experts. The experiment brief is the authoritative source to capture the shared understanding.
  • Link the experiment brief to your experiment when you write it in the experiment tool at
  • Referencing one source (the experiment brief) for the verbose experiment background enables the subject matter experts (engineering, QA, experiment tooling, data) to efficiently support the effort and respond to changing requirements. Product Managers will keep editing the experiment brief throughout the lifecycle of the experiment. Capture the decisions as they are made back in your experiment brief. Decisions left in in slack channels, verbal discussions, various notes, jira tickets, etc - increase the likelihood of errors and wasted effort.
  • There Checklist section at the end of your Experiment Brief to help you through the steps of your experiment.

Key Resources for Product Managers

Rather than duplicate links that may later break as the process improves - links to the resources and training videos are in the Workflow Miro Board.

  • There are several weekly touchpoints Office Hours for in person assistance from subject matter experts in different areas.
  • If you are confused on any aspect - that's expected the first few experiments until you've gotten the rhythm. Ask in the #ask-experimenter Slack channel - it is an open community of support from several disciplines.

When shipping product changes: a guide on when to use what option

Experiment (Experimentation Program)RolloutHoldback
DefinitionExperiment with 2 or more branches (treatment groups). Insights gathered at the end of experiment timeframe. Can be scaled to Rollout.Delivery where you ship 1 change, and that change goes to all the users in the determined population.When you run an experiment (see Experiment column) at the same time as Rollout.
Population and timeframeA fixed set of population for a fixed set of timeframeA fixed set of population at each shipping phase (population size is variable - can be scaled up & down)Can have different population sizes between Rollout and experiment. Time frame of rollout and experiment can differ.
Measure causal DAU impactYesNoYes
See statistical significant insights on guardrail metricsYesNoYes
When to useLooking for a causal relationship between a product change and key business metrics like DAU and 2-week retention. Also when measuring the impact of a potential new feature.Looking to primarily reduce technical risks such as scalability when shipping a product changeLooking to make a product change available immediately to a wide population while monitoring product change impact to business metrics
Delays the speed of making the product change available to a wider population.No insights on key business metrics like DAU, as tooling is not set up to make control and treatment group comparisons.If a product change has detrimental impact to guardrail metrics and DAU, we may or may not be too late in rolling back the changes.
AdvantagesCan measure impact first, then if positive ship the changeCan scale up or down as appropriateFaster than blocking on impact measurements
Experiment brief neededYesNoYes

Example of how to construct a balanced launch plan

[What Velocity team practices today]

  • First, run a small experiment
    • Goal: Reduce business risk and decide whether to ship
    • Example: 5% of users*; observe DAU impact of 1% to 10%
  • Then, if results of small experiment is positive, ship via rollout + holdback
    • Goal: Measure impact at scale
    • Example: 50% of users, 25% get feature and the other 25% doesn’t; observe DAU impact of 3%-5%

*Note that: 5% can be whatever population you choose to target. Can be en-US, “infrequent users”, “users who haven’t imported bookmarks”, etc.