Beyond Uptime: Redesigning Infrastructure Metrics for Resilient, Human-Centered Systems

When a therapy platform goes down, it's not just a technical incident—it's a moment when someone seeking help may be left waiting, frustrated, or disconnected from their support network. Yet most infrastructure teams still measure success by a single number: uptime. While 99.9% availability sounds impressive, it masks the real experiences of users: slow page loads during peak hours, intermittent errors that break workflows, and maintenance windows that disrupt care. This guide argues for a fundamental redesign of infrastructure metrics—one that centers human outcomes, not just system availability. We'll explore why uptime is insufficient, what frameworks can replace it, and how to implement a metrics strategy that truly serves both operators and the people who depend on your systems.

Why Uptime Alone Fails Users and Operators

Uptime is a binary measurement: a system is either up or down. But for users, the experience is rarely binary. A therapy appointment platform might be technically 'up' yet deliver a terrible experience—pages loading in 10 seconds, video calls dropping, or forms failing to submit. These partial failures erode trust and disrupt care, yet they don't register in uptime dashboards. Operators, meanwhile, are incentivized to keep the system 'green' at all costs, sometimes postponing necessary maintenance or deploying risky patches to avoid a downtime blip. This creates a culture of fear and short-term thinking, where the true health of the system is hidden behind a single, deceptive number.

The Hidden Costs of Uptime Obsession

Teams that prioritize uptime above all else often find themselves in a reactive cycle. When something breaks, they scramble to restore service, but they rarely investigate the underlying causes of poor performance. Over time, technical debt accumulates, user complaints rise, and burnout among operators increases. In therapy contexts, the stakes are higher: a slow system can mean a client misses a critical session, or a clinician cannot access notes before an appointment. Uptime metrics simply don't capture these human costs. They also ignore the fact that some downtime is acceptable—even beneficial—if it allows for safe deployments or planned improvements. The question should not be 'Is the system up?' but 'Is the system delivering value to its users?'

Moreover, uptime is often measured at the infrastructure level—server availability, network connectivity—but not at the user-facing service level. A database might be running, but if the application layer is misconfigured, users still see errors. This disconnect means that teams celebrate 'five nines' while users experience daily frustration. To truly understand system health, we need metrics that reflect the end-to-end user journey, from login to task completion. This shift requires a cultural change: moving from a mindset of 'keeping the lights on' to one of 'enabling human flourishing.'

Core Frameworks for Human-Centered Metrics

Several frameworks have emerged to help teams move beyond uptime. The most well-known is the Google SRE approach, which uses Service Level Indicators (SLIs), Service Level Objectives (SLOs), and Error Budgets. But these concepts can be adapted for any organization, especially those in sensitive domains like therapy. The key is to define SLIs that measure what matters to users—response time, error rate, throughput—and set SLOs that reflect acceptable thresholds. Error budgets then allow teams to balance reliability with innovation: if you have remaining error budget, you can deploy new features; if you've exhausted it, you focus on stability.

Defining Meaningful SLIs for Therapy Platforms

For a therapy platform, relevant SLIs might include: time to book an appointment (should be under 30 seconds), video call connection success rate (target 99.5%), and note synchronization latency (under 2 seconds). These indicators directly impact the therapeutic relationship. A client who struggles to book may feel abandoned; a clinician whose notes don't sync may miss critical information. By measuring these user-facing metrics, teams can prioritize improvements that actually matter. It's also important to consider 'soft' indicators like user satisfaction scores or task completion rates, which can be gathered through surveys or in-app feedback.

Error Budgets as a Tool for Decision-Making

Error budgets transform reliability from a binary goal into a manageable resource. If your SLO for video call success rate is 99.5%, you have a 0.5% error budget per month. This budget can be 'spent' on risky deployments, experiments, or maintenance. When the budget is low, teams must focus on stability. This framework encourages honest conversations about trade-offs: a planned maintenance window might use up error budget, but it could prevent a future outage. In therapy settings, error budgets help clinicians and engineers collaborate—clinicians can understand that some downtime is necessary for long-term reliability, and engineers can justify investments in resilience.

Another framework is the 'Four Golden Signals' from Google: latency, traffic, errors, and saturation. These provide a balanced view of system health. For therapy platforms, latency might include page load times and API response times; traffic includes concurrent users and session counts; errors include HTTP 5xx and application-level failures; saturation covers CPU, memory, and database connections. Monitoring all four gives a richer picture than uptime alone. Teams can also add domain-specific signals, such as appointment completion rate or client dropout rate during video calls.

Step-by-Step Process to Redesign Your Metrics

Redesigning metrics is not a one-time project but an ongoing practice. Here is a step-by-step process that teams can follow, adapted from SRE best practices and tailored for human-centered systems.

Step 1: Map User Journeys and Identify Critical Paths

Start by documenting the key workflows your system supports. For a therapy platform, these might include: client registration, appointment booking, video session, note-taking, and billing. For each workflow, identify the steps that are most sensitive to failure. For example, during a video session, the connection quality is critical; if it drops, the session is disrupted. Map these critical paths and list the technical components involved (servers, APIs, databases, third-party services). This exercise helps you focus on what truly matters to users.

Step 2: Define SLIs for Each Critical Path

For each critical path, define one or more SLIs that capture user experience. Use the 'RED' method: Rate (requests per second), Errors (failed requests), and Duration (latency). For video sessions, you might track: connection success rate, average bitrate, and dropout frequency. For appointment booking, track: form submission success rate, average booking time, and calendar sync latency. Ensure SLIs are measurable with existing tools or by adding instrumentation. Start with a small set—five to ten SLIs—and refine over time.

Step 3: Set SLOs Based on User Expectations and Business Needs

SLOs should be ambitious but achievable. Engage stakeholders—clinicians, clients, product managers—to understand what level of reliability is acceptable. For example, clinicians might tolerate a 1% error rate on note saving but not a 5% rate. Use historical data to set realistic targets. It's okay to start with loose SLOs and tighten them as you improve. Document the rationale for each SLO so that future teams understand the trade-offs.

Step 4: Implement Monitoring and Alerting

Instrument your code to collect SLI data. Use tools like Prometheus, Grafana, or Datadog to visualize metrics. Set up alerts that fire when SLOs are at risk, not when individual metrics spike. For example, instead of alerting on a single slow request, alert when the error budget is 50% consumed. This reduces alert fatigue and focuses attention on systemic issues. Also, create dashboards that show user-facing metrics prominently, not just infrastructure health.

Step 5: Establish Error Budget Policies

Define how error budgets are used. For instance, if the error budget for video sessions is 0.5% per month, the team can deploy new features as long as the budget is not exhausted. When the budget is low, only stability work is allowed. Communicate this policy to the whole organization so that everyone understands why some deployments are delayed. In therapy contexts, involve clinical leadership in these decisions to ensure alignment with patient care priorities.

Step 6: Iterate and Review Regularly

Metrics are not static. Review SLIs and SLOs quarterly with stakeholders. Are the targets still relevant? Have user expectations changed? Are there new critical paths? Adjust as needed. Also, conduct post-incident reviews that focus on user impact, not just technical root cause. Ask: How many users were affected? What was the clinical impact? This reinforces the human-centered approach.

Tools, Stack, and Economics of Modern Monitoring

Choosing the right tools is essential for implementing human-centered metrics. The monitoring landscape offers many options, from open-source to commercial, each with trade-offs. Below is a comparison of three common approaches, with considerations for therapy platforms that must balance cost, privacy, and ease of use.

Privacy and Compliance Considerations

Therapy platforms handle sensitive health data, so monitoring tools must comply with regulations like HIPAA or GDPR. Ensure that any SaaS tool signs a Business Associate Agreement (BAA) and that data is encrypted in transit and at rest. Open-source tools can be self-hosted, giving full control over data, but require more effort to secure. Also, avoid logging detailed user information in metrics; aggregate where possible. For example, track error rates by endpoint, not by patient ID.

Cost Management

Monitoring costs can spiral if not managed. Set retention policies for metrics—keep high-resolution data for 30 days, then aggregate. Use sampling for high-cardinality metrics. For therapy platforms with variable traffic (e.g., more sessions in the evening), consider using serverless or auto-scaling to reduce idle costs. Regularly review your monitoring bill and trim unused dashboards or alerts. Remember that the goal is to gain insights, not to collect every possible data point.

Growth Mechanics: Scaling Metrics Without Losing Focus

As your platform grows, so does the complexity of your metrics. New features, more users, and additional services can lead to metric sprawl—hundreds of dashboards and alerts that nobody looks at. To avoid this, adopt a tiered approach to metrics. Define 'tier 1' metrics that are critical for user experience and business health; these are monitored 24/7 and have strict SLOs. 'Tier 2' metrics are important but can tolerate higher thresholds; 'tier 3' are informational. This prioritization helps teams focus on what matters.

Using Metrics to Drive Product Decisions

Human-centered metrics should inform not just operations but also product development. For example, if the SLO for appointment booking is consistently met, but user satisfaction is low, the issue might be in the user interface, not the infrastructure. Combine quantitative metrics with qualitative feedback. In therapy platforms, consider running periodic surveys or interviews with clinicians and clients to understand their pain points. Use metrics as a starting point for conversation, not as the sole truth.

Building a Culture of Reliability

Metrics alone won't change behavior. Teams need to embrace a culture where reliability is everyone's responsibility. This starts with leadership: executives should ask about SLOs and error budgets, not just uptime. Celebrate improvements in user-facing metrics, not just system availability. Conduct blameless post-mortems that focus on learning. In therapy settings, involve clinicians in incident reviews—they can provide context on how outages affect care. Over time, this culture shift leads to more resilient systems and better outcomes for users.

Risks, Pitfalls, and Mitigations

Transitioning to human-centered metrics is not without challenges. Here are common pitfalls and how to avoid them.

Pitfall 1: Defining Too Many SLOs

Teams sometimes try to define SLOs for every possible metric, leading to complexity and alert fatigue. Mitigation: Start with 3–5 critical SLOs that cover the most important user journeys. Expand only after the team is comfortable with the process. Remember that an SLO is a commitment—each one requires monitoring, alerting, and review.

Pitfall 2: Setting Unrealistic Targets

Setting SLOs too high (e.g., 99.99% for everything) can be expensive and demoralizing. Mitigation: Use historical data to set achievable targets. Involve stakeholders to understand what is truly acceptable. It's better to have a realistic SLO that is consistently met than an aspirational one that is constantly breached.

Pitfall 3: Ignoring the Human Element

Metrics can become a substitute for empathy. Teams might optimize for a metric (e.g., low latency) at the expense of user experience (e.g., by caching stale data). Mitigation: Always pair quantitative metrics with qualitative feedback. Use metrics to identify areas for investigation, not as the final verdict. In therapy contexts, remember that the ultimate goal is to support healing, not to achieve a perfect dashboard.

Pitfall 4: Neglecting Technical Debt

Error budgets can be used to justify risky changes, but if the underlying system is fragile, even small changes can cause outages. Mitigation: Use error budgets as a signal, but also invest in resilience engineering—chaos engineering, redundancy, and regular maintenance. Allocate a portion of the error budget specifically for stability work.

Decision Checklist and Mini-FAQ

Before implementing a new metrics strategy, consider the following checklist to ensure you're on the right track.

• Have you identified the top 3 user journeys that are most critical?
• Do you have SLIs that directly measure user experience for each journey?
• Are your SLOs set with input from stakeholders (clinicians, clients, product)?
• Do you have a process for reviewing and adjusting SLOs quarterly?
• Is your monitoring tooling compliant with relevant privacy regulations?
• Have you established an error budget policy that balances reliability and innovation?
• Are alerts based on error budget consumption, not raw metric spikes?
• Do you conduct post-incident reviews that focus on user impact?

Frequently Asked Questions

Q: How do we convince leadership to move away from uptime? A: Present data showing that uptime doesn't correlate with user satisfaction. Use examples from your own platform where the system was 'up' but users were unhappy. Propose a pilot with a few SLOs and measure the impact on user experience and team morale.

Q: What if we don't have the resources to implement complex monitoring? A: Start small. Use your existing logging and error tracking to define one or two SLIs. For example, track the error rate of your most critical API endpoint. Set a simple SLO and alert when it's breached. You can expand gradually.

Q: How do we handle third-party dependencies in our SLOs? A: Include them in your SLIs as long as they affect user experience. For example, if you use a third-party video API, track its error rate. However, you may not be able to control its reliability. In such cases, consider building fallback mechanisms or communicating limitations to users.

Q: Is it okay to have different SLOs for different user segments? A: Yes. For example, you might have a stricter SLO for paying clients versus free users. Just be transparent about it and ensure that the differences are justified by business needs.

Synthesis and Next Actions

Moving beyond uptime is not just a technical change—it's a philosophical one. It means acknowledging that systems exist to serve people, and that metrics should reflect that purpose. By adopting frameworks like SLOs and error budgets, teams can make better decisions, reduce burnout, and ultimately deliver more reliable and compassionate services. For therapy platforms, this approach is especially critical: every second of downtime or slow response can affect someone's mental health journey. Start small: pick one user journey, define one SLI, set one SLO, and see how it changes your team's focus. From there, iterate and expand. The goal is not perfection, but continuous improvement toward a system that truly supports human well-being.

Remember that this is general information only, not professional advice. For specific guidance on compliance or clinical impact, consult a qualified professional.