Designed Amplitude’s first anomaly detection feature to help users spot meaningful changes in product metrics without manual guesswork. Reduced time-to-insight, improved trust in data, and empowered teams across 800+ customer accounts to act faster with confidence.
At Amplitude, I was part of the Growth team, a lean, high-impact group focused on rapidly validating ideas before committing them to the broader product roadmap. Our work helped de-risk big bets, prioritize high-impact opportunities, and inform long-term strategy across product teams.
In March 2020, we inherited the early anomaly detection and forecasting initiative from the analytics roadmap. With hiring frozen due to COVID-19, we were tasked with building a functional, validated foundation — fast. We collaborated directly with customers to prototype and refine the initial experience, laying the groundwork for future expansion by the core analytics team.
Amplitude’s users rely on the platform to monitor core metrics and understand their trends. However, the existing methods for evaluating metric fluctuations are often imprecise, leading to wasted time and resources investigating noise rather than meaningful insights. To solve this, I led the design of Amplitude’s first anomaly detection feature, developing an anomaly detection tool that helps customers distinguish significant metric changes from statistical noise. By identifying "what’s unusual in the data," users can focus on actionable insights and make informed decisions with confidence.
Anomaly detection was built to work across user segments, from data analysts to PMs and growth teams, without sacrificing flexibility or explainability. The feature offers three customizable modes to suit each segment:
The feature was beta-tested with 900 customer accounts, with 17% engaging directly through in-app prompts. Since launch, one-third of Amplitude’s 2,400 paying customers (over 800 teams) now actively use Anomaly Detection in their workflows. This strong adoption rate highlights the feature’s success in solving a critical user pain point and delivering measurable value at scale.
Monitoring core metrics shouldn’t require a math degree or constant vigilance. Yet, before this work, users had no automated way to detect meaningful shifts in usage or conversion, forcing teams to eyeball dashboards and manually compare week-over-week numbers to guess what changed and why. This led to:
We aimed to replace guesswork with clarity. By surfacing statistically significant changes and embedding them directly into the tools teams already used, we made it possible to catch issues earlier, validate hypotheses faster, and democratize data literacy across entire organizations.
Testing early iterations with customers allowed us to gather valuable insights that informed both design and product decisions. During these sessions, we explored how customers currently find this information, how frequently they use the "compare to past" feature, and their expectations for an anomaly detection tool. Customers consistently emphasized the importance of ease of use, trust in the underlying model driving the computations, and confidence in selecting parameters to configure their output settings effectively. These insights shaped our approach to building a tool that feels intuitive and reliable.
A key friction point for users was understanding the “model” behind their charts, specifically, how results were calculated and whether they’d chosen the “right” parameters. While users trusted Amplitude’s analysis engine, many felt uncertain about configuring the tool correctly, especially those working alongside a data scientist to understand their insights. "How do I trust Amplitude's tool more than my data scientist?"
To address this, we proposed leveraging Amplitude’s existing credibility by introducing a set of predefined “modes” — industry-aligned parameter presets designed for common use cases. These modes helped remove ambiguity and empower users to focus on insights, not setup.
This approach delivered value in several ways
One consistent insight across research: users value smart defaults. These predefined parameters, surfaced through mode selection, gave users confidence without requiring deep configuration. For forecasting, we began with an intentionally minimal state, allowing users to layer in complexity as needed. This maintained discoverability while avoiding cognitive overload.
We introduced three distinct modes to support a range of needs:
This tiered approach ensured that both novice and advanced users could extract meaningful insights without friction offering clarity by default and flexibility by design. Core features included:
To reduce visual noise while still supporting rich exploration, we implemented hover-based interactions for anomaly segments. By default, the UI displays up to 10 segments — enough to surface patterns without overwhelming the viewer. On hover, additional context like confidence bands and forecast parameters is revealed inline, allowing users to explore deeper without cluttering the interface.
We also reserved a dedicated color token from Amplitude’s design system solely for anomalous data points. This distinct visual treatment helped anomalous values stand out clearly in a product otherwise dominated by shades of blue — preserving clarity in a dense visual ecosystem.
Rather than a traditional on/off toggle, we implemented a button toggle to activate the feature. Early explorations using a toggle switch created inconsistency with Amplitude’s interface paradigms and introduced ambiguity about the feature’s state. The button model offered clearer behavior:
This solution minimized confusion, maintained platform consistency, and allowed for scalable interaction design in future enhancements.
The anomaly detection feature redefined how teams used Amplitude for proactive product monitoring. Instead of relying on slow, manual comparisons, users now had real-time insights delivered in context.
This release dramatically simplified metric monitoring while improving anomaly literacy across teams.
One of the biggest challenges was making the output of complex statistical models feel simple, usable, and credible to non-technical users. We had to carefully balance statistical accuracy with intuitive UX, ensuring that teams felt empowered, not confused.
It also required close partnership across disciplines: data science, product, and design had to align deeply on thresholds, terminology, and trust. That alignment not only shaped this feature, it helped shift the broader culture toward more accessible, explainable product intelligence.
Redesigned Credible’s pre-approval dashboard to reduce confusion, personalize next steps, and drive loan conversion. Improvements led to stronger engagement across buyer types, supporting both early-stage "dreamers" and ready-to-act borrowers.