Building a Metric-First, High-Quality Data Culture at Voi

Data at Voi

From its founding, Voi’s leadership set a clear principle: every decision should be based on data. That culture is now built into every level, from central HQ in Stockholm to local market teams, and most office employees interact with data weekly.

Voi’s data team realised early on that metrics matter to end users, whether for decision-making in dashboards or for running hundreds of A/B tests to inform product decisions. Being “metric-first” has become an important philosophy for how the data team operates.

The challenge

Voi operates one of Europe’s most complex micromobility data setups, with over 130,000 scooters sending a ping every five seconds. That is billions of IoT, hardware, operations, CRM, and product data points.

From day one, Voi’s leadership pushed for decisions to be data-driven. Almost every office role uses data weekly, but scaling that culture came with important data quality-related challenges:

• Lacked the ability to proactively detect all kinds of issues to deliver reliable metrics
  
• Scattered data quality workflows and slow resolution times without clear issue resolution processes
  
• Alert fatigue and slow incident triage, especially during upstream outages like ingestion failures

“We’ve never had to sell data to our users. The challenge is keeping up with demand.” — Magnus Dahlbäck, Senior Director Data and Platform

A metrics-first approach with 90% end-user adoption

To eliminate KPI discrepancies and ensure reliability, Voi replaced dashboards with a metrics-as-a-product approach. Metrics are centrally defined in Steep and tied directly to the dbt models monitored in SYNQ, so freshness, volume, and quality are always visible.

With 250 Steep licenses and 90% monthly active usage, adoption is company-wide. If a metric fails a check, analysts see it in SYNQ immediately and can act before stakeholders use bad data, transforming the company’s ability to detect issues proactively.

“We don’t get the question anymore about why numbers differ between two dashboards. That problem is completely eliminated now.” — Axel Strandberg, Senior Data Analyst at Voi

Full-stack analysts with a centralised issue resolution workflow

Voi’s analysts are full-stack, owning everything from raw warehouse data to business advisory. Each analyst is embedded in a product team, giving them deep domain knowledge and a direct line to the engineers producing the data.

SYNQ gives analysts real-time visibility into the health of their domain’s models. By tagging Tier 1 assets in dbt, analysts can instantly see which issues impact the most critical products, trace lineage to downstream metrics, and keep stakeholders informed. These Tier 1 tags are carried through into SYNQ alerts in Slack, so analysts know immediately when a high-priority asset is affected.

“We wanted dbt to be the single source of truth for defining our quality standards. SYNQ’s tight dbt integration made that possible, so our monitoring is always aligned with how the models are built and owned.” — Magnus Dahlbäck, Senior Director Data and Platform at Voi

This structure, combined with SYNQ’s issue management workflow, has cut the time from problem detection to stakeholder communication from hours to minutes, even during upstream outages.

“Before, something could fail, and we had no central way of tracking if it was still ongoing or being worked on. Now in SYNQ, we can see all ongoing issues, who is handling them, and what has been resolved.” — Axel Strandberg, Senior Data Analyst at Voi

Data products with built-in quality controls

Data at Voi powers more than just dashboards. Some data products require stricter quality standards, including:

• A/B testing platform – Centralises test analysis in the data warehouse with standardised methodology. Hundreds of experiments run simultaneously, with SYNQ monitoring the underlying models to catch issues before results are published.
  
• Supply & Demand ML engine – Predicts demand to optimise scooter placement, battery swaps, maintenance, and repairs. These models are set to Tier 1 in dbt, so SYNQ applies stricter monitoring and alerting rules.

During a Google ingestion outage, SYNQ’s anomaly monitors detected the problem hours before dbt’s static freshness tests, allowing analysts to inform stakeholders and adjust operational plans in real time.

“That was proof for us. The dynamic monitoring picked it up faster, and we could act before the impact snowballed.” — Magnus Dahlbäck, Senior Director Data and Platform at Voi

SYNQ has also helped reduce the costs of monitoring some of these data products. In a recent example, freshness tests in dbt were running hourly to monitor marketing pipelines from Braze’s customer engagement platform. Moving to SYNQ’s freshness monitoring saved more than €2,000 per month.

Next up: AI at Voi

Voi’s data team already uses AI to speed up their work. Tools like Cursor and dbt Fusion help with coding, navigating the codebase, and scoping analytics work. LLMs summarise unstructured feedback and support experiment design.

For self-serve, Voi is using Magnowlia for natural language querying, underpinned by metrics and models monitored in SYNQ. The team is also using Pancake AI to get insights from rider feedback.

They are also early adopters of SYNQ’s Scout, using it to help triage incidents and identify root causes. By automatically surfacing likely causes and relevant upstream and downstream context, Scout can reduce the time analysts spend investigating and help prevent repeated back-and-forth between teams.

“Scout can take a lot of the guesswork out of figuring out what is broken and why. As we hit more complex issues, it will really speed us up.” — Axel Strandberg, Senior Data Analyst at Voi