North Star Translation Layer
A system from Sri Batchu for aligning a large growth organisation around a single company North Star metric while preserving team-level accountability for metrics those teams can directly influence.
The problem it solves: a large growth team (Batchu managed 300+ at Instacart) works on diverse initiatives — checkout conversion, load time, onboarding, ads, retention. Each team needs a metric it can move. But if every team is goaled directly on the company North Star (e.g. monthly active orders), the connection between their local work and the company metric is too indirect to drive useful decisions.
How it works
-
Define the company North Star. At Instacart: monthly active orders (MAO). The North Star must satisfy two properties: clear linkage to business value, and intuitive enough for every person in the company to understand.
-
Give each team a local metric it can directly influence. Checkout team → checkout conversion rate. Load time team → median page load. Onboarding team → first-order rate.
-
Build a translation layer — jointly owned by finance and data — that converts each team’s local metric into expected MAO impact. ‘If you get one extra weekly order from the same customer because of your checkout improvement, it would have X impact on the company’s MAO.’ Update translations every six months based on new regression data.
-
Use the translation layer for cross-team resource allocation and planning, not for day-to-day team work. Teams run their sprints against their local metric; leadership allocates budget and engineering headcount based on MAO per dollar or per engineer.
The resource allocation benefit
The translation layer makes a previously intractable question tractable: ‘Should we add more engineering to the checkout team or the onboarding team?’ Without the translation layer, this requires qualitative judgement and political negotiation. With it, you compare MAO-per-engineer estimates across teams and allocate to where the translation is most favourable.
Marginal decisions remain hard
Batchu is explicit that the translation layer does not eliminate hard calls. ‘Marginal decisions are marginal for a reason. They are really hard things to decide.’ The framework reduces cognitive load for clear decisions; the hard cases still require judgement. Its value is reducing the set of decisions that require political or qualitative arbitration.