Masthead Data x Livesport
Case Study: How Masthead monitors 46 481 data metrics for Livesport
Key results
Livesport monitors 46,481 business-critical metrics with Masthead's ML anomaly detection, inside BigQuery, with no hand-written thresholds. About 0.85% of data points are flagged, so alerts stay actionable. Masthead has caught regional tracking outages that global dashboards would have hidden, and makes it actionable for Livesport team despite an immense scale that no human can comprehend.
Dimension
Telemetry & scope
Customer
Livesport publishers and networks
Scale
46,481 active time-series rules across 7 custom data scans
Monitored assets
5 BigQuery views covering 158 domains and 74 platform slices
Throughput
1,419,727 evaluations per rolling 30-day window (about 47,000 per day)
Signal-to-noise
0.85% anomaly rate
About Livesport
Livesport is the Prague-based technology company behind Flashscore, the world's largest live-score platform. Its portfolio reached a record 155+ million monthly active users and 400+ million app downloads in October 2025.
Fact
Detail
Brands
Flashscore, BeSoccer, Soccerway, Tribal Football, Diretta.it and local domains
Flashscore reach
125+ million monthly users across 70+ websites, 38 sports, 50+ countries
Audience mix
Football over 65% of users, tennis 10%, basketball 9%
Top markets
Brazil, Italy, France, UK, Poland, Germany; fast growth in Nigeria and the US
Peak load
48.5 million users and 2.58 billion views in one day (El Clásico, 26 October 2025)
Data stack
Google BigQuery, Dataform, Tableau
Revenue depends on engagement events that are easy to lose silently: advertising impressions, affiliate clicks and live-odds interactions with betting partners. Each is tracked per domain, platform and feature. A tracking tag that breaks on one app version in one country costs real money, yet barely moves the global totals.
Traffic is also unusually spiky. A single match can bring tens of millions of users in hours, then drop away. Any data-quality system has to tell a Champions League night apart from a real outage.
About Masthead
Masthead Data is a data observability and FinOps platform built for Google BigQuery. It reads BigQuery logs and metadata instead of querying customer tables, so it needs no access to customer data and adds no SQL compute of its own. Out of the box it monitors freshness, volume, schema changes and pipeline errors across every connected project.
Capability
What it does
Used at Livesport
ML models learn each table's update pattern on a day 0 from the BQ audit logs
Yes, warehouse-wide
Custom data-quality scans
Masthead learns thresholds from the results of the customer's own scheduled scans and raises incidents
Yes: 7 scans, 46,481 rules
Lineage and impact analysis
Column-level lineage from BigQuery to BI tools shows which assets an incident affects
Yes
BigQuery shared dataset
Incidents and insights exported to BigQuery for custom usage in dashboards and custom alerting
Yes
MCP server and agent skills
Agents triage incidents, trace lineage and act on incident insights (Masthead for agents)
Yes
Alerts go to Slack, Microsoft Teams, Google Chat, PagerDuty or Jira.
The challenge
Livesport is in the media and publisher business, they needed to watch and carefully track tens of thousands of small, revenue-linked web and conversion metrics that global dashboards could not see. Another challenge is monitoring over 40K metrics can be an expensive exercise for BQ compute, the goal was to make it reasonable cost on BigQuery from a unit economics standpoint.
Extreme cardinality. Engagement is tracked as domain × platform × action. That produces 46,481 separate time series.
Blind spots in global aggregates. If affiliate tracking breaks on Android in Peru, or odds feeds drop in Japan, the loss disappears inside worldwide totals.
No workable static thresholds. Nobody can hand-tune min/max rules for 46,481 metrics. Weekend fixtures, off-season lulls and tournaments like the World Cup make fixed limits either noisy or blind.
Direct revenue exposure. Affiliate clicks and odds interactions drive monetization. A silent break means lost revenue and disputes at partner reconciliation.
Cost at scale. Running anomaly checks across this many metrics, with a SQL-first vendor or Google Dataplex, would cost more than Livesport's annual BigQuery operations.
Masthead's solution
Masthead runs ML time-series anomaly detection on Livesport's own data-quality scans, inside BigQuery, without exporting raw data or changing production workflows.
Curated views. Livesport exposes 5 BigQuery views, such as odds actions and top app actions, sliced by 158 domains and 74 platforms.
Scheduled scans. 7 custom data scans aggregate each metric daily, about 47,000 evaluations a day.
One model per metric. Masthead models each of the 46,481 rules as its own time series. It learns a prediction interval from history, weekly seasonality and variance.
Incidents, not noise. A value outside its interval, or a silent drop to zero or null, becomes an incident. Only 0.85% of data points are flagged.
Back into the warehouse. Results sync to Livesport's BigQuery. The team joins them with internal analytics, Masthead’s Slack alerting and its own severity dashboard, and inspects each rule's history in the Masthead UI.
On top of the export, Livesport built a severity-ranked dashboard with a seven-day view, so recurring patterns across many rules stand out. Masthead supplies the signal; Livesport adds the domain context.
Business value and what's next
Livesport moved from reactive debugging to automated coverage of 46,000+ revenue-linked slices, with no thresholds to maintain.
Zero-touch maintenance. 46,481 metrics are monitored with no hand-written min/max rule.
Protected monetization. Broken affiliate tags and odds-interaction drops surface before monthly invoicing and partner reconciliation.
Precision. About 1.4 million evaluations produce a 0.85% anomaly rate, low enough for engineers to act on.
Native to BigQuery. Results live in Livesport's warehouse and fit its existing Dataform and governance workflows.
Affordable scale. Log-based monitoring avoids the SQL compute that made alternatives cost more than Livesport's BigQuery operations.
Next, the teams plan to give Livesport's agents near-real-time incident metadata through Masthead's MCP server. Agents would split incidents into operational breaks to fix the same day and known data skews to record.
