Always-on Signals

Situation

Miro's key AI features (such as Sidekicks and Flows) were iterating rapidly, making traditional research sprints and static retrospectives obsolete. Experience quality data was fragmented across isolated in-product surveys, Slack channels, reports and decks. Leadership needed a continuous, real-time pulse on customer sentiment, so I decided to create a program to track satisfaction (CSAT) and product-market fit (PMF) directly linked to commercial impact and feature usage.

Task

Build and operationalize Always-on Signals — a standing, automated health measurement system. The primary goal was to triangulate live survey, usage, and revenue data from Snowflake into decision-ready views, replacing ad-hoc pulls with a system product teams would actively use.

Approach

Data was triangulated through automated pipelines from Snowflake, combining four core signals: sentiment (CSAT, PMF, overall NPS), value (ARR-weighted satisfaction by plan tier), engagement (active user tiers), and adoption/retention metrics. A joint-attitude framework mapped overlapping CSAT and PMF respondents into four distinct profile quadrants — AI Champions, Happy but Indifferent, Frustrated but Hooked, and Disappointed and Detached — to separate dependence from happiness.

Multiple linear regression was applied to coded open-text comments, isolating specific drivers and assigning a “CSAT uplift if solved” value to rank product fixes by business impact. The research itself was productized: continuous HTML dashboards and a dedicated web page, alongside human-curated weekly and bi-weekly updates to Slack — treating research as a living product with its own release cadence.

Result

The program earned high-scale operational adoption: it replaced traditional company dashboards and scaled to cover additional product lines, absorbing hundreds of thousands of CSAT, PMF, and relationship NPS responses across multiple product portfolios. It surfaced a critical strategic discovery — significant CSAT drops in specific key AI feature areas even while overall company sentiment looked stable — exposing enterprise revenue risk tied to low-satisfaction accounts. It also created a data-quality feedback loop, surfacing underlying instrumentation issues (like tag bleed between products) that fed directly into engineering fixes.

Methodology

Always-on Signals demonstrates construct definition in the wild, combining attitudinal metrics (CSAT/PMF/NPS) with behavioral telemetry (board creation, sharing, seven-day retention). Pairing linear regression with user-segment cuts (tenure, plan, role) extracts actionable statistical impact while preserving observational rigor — turning mixed-methods research into a core operational deliverable.