> Markdown version of [/videos/2086-data-binning-and-understanding-histograms](https://www.wearedevelopers.com/videos/2086-data-binning-and-understanding-histograms). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Data binning and understanding histograms

Are default Prometheus buckets hiding your worst latency spikes? Discover how proper histogram binning and percentile tracking can expose critical P99 outliers and fix your alerting infrastructure.

- **Speakers:** [Michal Bojko](https://www.wearedevelopers.com/@michal-bojko)
- **Event:** World Congress 2026 Europe - Virtual Stage
- **Published:** July 2, 2026
- **Duration:** 30:35
- **URL:** https://www.wearedevelopers.com/videos/2086-data-binning-and-understanding-histograms

## Summary

In observability and performance monitoring, relying on average (mean) latency metrics frequently masks the reality of poor digital experiences, leaving critical latency tails unaddressed. Because delays translate directly to lost revenue and user abandonment, capturing outliers requires a shift from mean-based tracking to precision monitoring using percentiles like P95 and P99. Central to this precision is visual analytics: configuring histogram bin widths to expose performance realities dynamically, as "too few will conceal, hide data patterns [and] too many will introduce unnecessary noise."

Applying proper bin widths to specific infrastructure distributions—such as normal, right-skewed, or bimodal datasets—reveals architectural nuances like fast cache retrievals versus slow cold database reads. However, even capable monitoring tools can deceive engineers if misconfigured. Prometheus, for instance, calculates histogram quantiles via linear interpolation. Relying on default bucket sizes instead of bespoke, service-specific configurations often distorts the true P99 representation, triggering false positives, alert fatigue, or entirely overlooked incidents.

To construct an optimal alerting infrastructure, teams must actively probe their APIs and web services to establish actual data baselines. While statistical guidelines like the Sturgis or Freedman-Diaconis models offer structural hints, defining robust bucket boundaries requires empirical testing against service level objectives (SLOs). By splitting bimodal architectures into distinct metric streams and tying notifications strictly to tailored percentile thresholds, organizations can drastically lower alert lag, confidently accelerate root cause analysis, and ensure the reliability of the end-user experience.

**Keywords:** data binning strategies, latency histograms, P99 tail latency, prometheus bucket configuration, sturgis formula, freedman-diaconis rule, right-skewed distribution, bimodal metric tracking, service level objectives (SLO), alert lag reduction, observability false positives, prometheus linear interpolation, mean vs percentile monitoring, digital experience latency

## Chapters

1. **Business impact of slow API responses** (00:02) — How slow system responses affect revenue and why precise observability identifies costly performance drops.
1. **Average latency metrics mask real user problems** (02:49) — Why relying on mean metrics obscures the severe infrastructure delays experienced by long-tail users.
1. **Revealing true performance with percentile metrics** (04:42) — How utilizing P99 metrics efficiently exposes the actual experience of users impacted by service latency.
1. **Finding the correct histogram bin width** (05:20) — How few histogram bins hide internal data patterns while excessive bins introduce unnecessary visualization noise.
1. **Mathematical models for bin width selection** (09:33) — How the Sturges and Freedman-Diaconis formulas act as starting models for optimizing histogram sizing.
1. **Mean versus percentile alert scaling speed** (12:00) — Why percentile-based performance alerts trigger much faster operational responses compared to standard mean-based alerting setups.
1. **Identifying latency distribution types in traffic** (14:06) — Simulating normal, right-skewed, and bimodal data patterns to correctly design realistic service level objectives.
1. **Configuring Prometheus histogram buckets for accuracy** (19:35) — How linear interpolation and default bucket settings inside Prometheus can lead to inaccurate latency representations.
1. **Probing internal APIs for optimal bucket boundaries** (21:22) — How profiling disparate APIs establishes proper boundary values for active operational latency alerts.
1. **Consequences of inaccurate percentile reporting on alerts** (27:05) — The damaging impact of misconfigured metric buckets on general alert fatigue and service health stability.
1. **Key takeaways for monitoring service latency** (28:10) — Why site reliability engineers must continuously probe APIs and customize buckets to capture real user experiences.

## Related Moments

- [Measuring database latency through high percentiles](https://www.wearedevelopers.com/videos/806-leveraging-real-time-data-in-fsis) (from "Leveraging Real time data in FSIs")
- [Evaluating request latency using metric distribution timing histograms](https://www.wearedevelopers.com/videos/1599-keycloak-case-study-making-users-happy-with-service-level-indicators-and-observability) (from "Keycloak case study: Making users happy with service level indicators and observability")
- [Introduction to metrics and observability challenges in monitoring](https://www.wearedevelopers.com/videos/57-all-your-telemetry-data-from-any-source-in-one-place) (from "All your telemetry data from any source in one place")
- [Measuring network latency and combating performance metric fatigue](https://www.wearedevelopers.com/videos/1772-wearedevelopers-live-guten-tag-web-standards-ai-and-more) (from "WeAreDevelopers LIVE – Guten TAG, Web Standards, AI and more")
- [Analyzing telemetry trace results for millisecond response latency](https://www.wearedevelopers.com/videos/1167-maximising-cassandra-s-potential-tips-on-schema-queries-parallel-access-and-reactive-programming) (from "Maximising Cassandra's Potential: Tips on Schema, Queries, Parallel Access, and Reactive Programming")
- [Debugging performance bottlenecks with developer tools](https://www.wearedevelopers.com/videos/817-the-illusion-of-a-performant-web-application) (from "The Illusion of a Performant Web Application")

## Related Articles

- [MLops – Deploying, Maintaining And Evolving Machine Learning Models in Production](https://www.wearedevelopers.com/magazine/115-mlops-deploying-maintaining-and-evolving-machine-learning-models-in-production)
- [Making Data Warehouses Fast: A Developer’s Story](https://www.wearedevelopers.com/magazine/107-making-data-warehouses-fast-a-developer-s-story)
- [Dev Digest 139 - Soft and hard queries](https://www.wearedevelopers.com/magazine/487-dev-digest-139-soft-and-hard-queries)
- [Dev Digest 119 - ❤️ === ❤️](https://www.wearedevelopers.com/magazine/454-dev-digest-119)

## Related Jobs

- [Data Scientist](https://www.wearedevelopers.com/jobs/ext/1351648-data-scientist) at **Almedia**
- [Principal Software Engineer, Database Infrastructure](https://www.wearedevelopers.com/jobs/ext/1465908-principal-software-engineer-database-infrastructure) at **GitHub**
- [Devops Engineer](https://www.wearedevelopers.com/jobs/ext/1940926-devops-engineer) at **Bitpanda**
- [Staff Business Intelligence Engineer](https://www.wearedevelopers.com/jobs/ext/626164-staff-business-intelligence-engineer) at **Twilio**
- [Staff Software Engineer, Database Infrastructure](https://www.wearedevelopers.com/jobs/ext/1470125-staff-software-engineer-database-infrastructure) at **GitHub**
- [Staff, Business Intelligence Engineer](https://www.wearedevelopers.com/jobs/ext/1401813-staff-business-intelligence-engineer) at **Twilio**