Make the case for Meta AI in Instagram DMs, then set its goals and health metrics

Meta has brought Meta AI, its AI assistant, into Instagram direct messages. Evaluate that decision as a product manager: whether it is worth the investment, what it should aim to achieve, and how you would know it is working without hurting the product. ### Clarifying Questions - Is this a decision about whether to launch, or a review of a feature that is already live? - Does "investment" include the cost of serving every conversation at Instagram's scale, or only the cost of building the experience? - Is the assistant limited to one-on-one chats with it, or can people also bring it into conversations with friends? - Are teen accounts in scope, and under the same rules as adults? ### Part 1 — The investment case Is putting Meta AI into Instagram DMs a good investment for Meta? If you think it is, explain why. ```hint Why this surface Compare what DMs give an assistant that a standalone chatbot app does not, then set that against what every conversation costs Meta to serve and what time it might pull away from other parts of the app. ``` #### What This Part Should Cover - A rationale tied to Meta's position in messaging and AI rather than general enthusiasm for AI - The costs and risks of the bet, including serving cost, cannibalization, privacy and safety - A clear verdict and the conditions under which it holds ### Part 2 — Goals What goals would you set for Meta AI in Instagram DMs? ```hint Goals over time Ask how the goals should differ in the first months after launch and later on, and whose experience, besides the person talking to the assistant, they need to protect. ``` #### What This Part Should Cover - Goals that follow from the investment case in Part 1 - A sequence from early adoption to long-term value - What the product should explicitly not optimize for ### Part 3 — Metrics Which metrics would show that it is working, and which would you watch to keep the product healthy? ```hint Define success first The number of messages sent to the assistant can climb while its answers get worse. Decide what a successful exchange looks like before choosing what to count. ``` #### What This Part Should Cover - A north-star metric built on a precise definition of a successful exchange - Adoption, quality, habit and social metrics that show whether it is working - Guardrails for messaging between friends, safety, trust, overall app engagement and cost - How to measure the incremental effect, such as a holdout ### What a Strong Answer Covers - A verdict, goals and metrics that are consistent with one another - Clear separation of lasting usefulness from novelty - Attention to the private, social nature of DMs and to younger users - Awareness of what it costs to serve an AI assistant at Instagram's scale ### Follow-up Questions - Assistant usage in DMs is growing, but messages between friends per user are slipping in the same cohort. How do you decide whether that is a problem? - Leadership asks when Meta AI in DMs should start to earn money. Which options would you consider, and which would you rule out? - How would your goals and guardrails differ for teen accounts?