KPI as a Product: The Missing Principle in Data Mesh

This is Part 1 of the KPI as a Product series. Parts 2 and 3 develop the framework further and are forthcoming. Abstract Data mesh reframed how organizations manage analytical data by establishing four principles: domain-oriented decentralized ownership, data as a product, self-serve data infrastructure, and federated computational governance. The second principle — data as a product — has proven the most influential, transferring product-management discipline (ownership, contracts, service levels, discoverability) to datasets. This paper argues that the framework, as commonly implemented, stops one step short of the artifact that organizations actually steer by: the key performance indicator. In practice, product thinking is applied to the raw and refined data that feed a KPI, while the KPI itself — the computed, decision-facing metric — remains an unowned by-product, defined redundantly in dashboards, spreadsheets, and tribal memory. The result is a recurring enterprise failure mode: technically clean data products coexisting with contested, inconsistent metrics. As a proposed extension of Dehghani's framework, I name the missing element KPI as a Product (KaaP): the extension of data mesh's product discipline from the dataset to the metric. To ground the principle, I propose the Material–Measure–Indicator (MMI) hierarchy, which distinguishes raw material (source-aligned data products), measures (directly computed, rule-based quantifications), and indicators (interpretive claims built on measures), each with a distinct ownership obligation. I specify a KPI Product Contract comprising two accountable owners — a measure owner for the computation and an indicator sponsor for the proxy relationship — together with a versioned definition, declared inputs, quality service objectives, intended-use and known-misuse guidance, and change management. The contract is illustrated side-by-side with a conventional data contract to make the distinction concrete. A central diagnostic hypothesis follows: many failures attributed to "data quality" may not be measurement failures at all, but indicator failures — unowned inferential leaps from correct measures to unagreed meanings. The paper positions KaaP relative to adjacent work on semantic layers, aggregate data products, and medallion architecture, and outlines future work on AI-manufactured metrics, where deterministic KPI Product Contracts become the governance mechanism for non-deterministic metric generation. Working paper, version 1.2. Licensed CC BY 4.0.Correspondence: [email protected] · https://orcid.org/0000-0001-5887-7620