How much infrastructure are you burning while AI thinks?

As system data grows and budgets tighten, observability must justify its cost with business-aligned outcomes – and that’s a good thing.

This shift signals rising expectations, not diminishing faith. Enterprises still care about uptime, but they increasingly want to know whether their systems are delivering results that matter: seamless user experiences, stable revenue pipelines, reduced operational drag. That’s not a downgrade for observability – it’s a promotion. It means these tools are no longer seen as passive monitors, but as active participants in delivering business value.

Insights from Florian Ortner, SVP of Product Management for Observability at Dynatrace, during the company's London Innovate Roadshow reveal this transformation through the practical constraints and hard-won lessons that actually drive enterprise adoption – moving beyond vendor positioning toward operational reality.

The "use case available" framework

A concept emerging from enterprise observability practices cuts through conventional uptime metrics: "use case availability." This represents a fundamental departure from traditional service availability measures. Ortner explains:

We talk internally about whether something is 'use case available.' It's not just that the cloud service is running, but whether it performs adequately, has the right usability, and supports the real-world outcomes our customers care about.

This distinction reflects a broader shift in how enterprises evaluate system health. Traditional monitoring focused on whether services were operational, whereas use case availability examines whether they're delivering intended business value. For a financial services application, this might mean not just that the payment API responds, but that transaction completion rates meet business requirements under realistic load conditions.

The framework represents what Ortner describes as observability's evolution "closer to business outcomes, more sensitive to economic pressure, and increasingly shaped by cross-functional accountability." Rather than viewing observability as purely technical infrastructure, enterprises are demanding direct correlation between monitoring investments and operational outcomes.

While Dynatrace's keynote emphasized AI capabilities – causal AI for root cause analysis, predictive AI for anomaly detection, and generative AI for natural language interfaces – a sit-down conversation with Ortner provided crucial context around AI's practical limitations.

He shares a telling example involving GitHub Copilot's code suggestions during a recent demonstration: 

One of the Large Language Model's (LLM) recommendations was to just send the password in plain text. Then you don't have the problem. Another was to make the service public. Technically correct—completely wrong.

This example reflects broader industry experience. Ortner references Microsoft research examining LLMs' performance in Azure incident root cause analysis, noting that while LLMs could often identify likely causes, "the time to get there ranged from minutes to days, and the computational cost was substantial." The research concluded that models "want to be right. And they don't stop talking until they are."

The research found nuanced results – models could often identify likely root causes, but with unpredictable timeframes and substantial computational costs. For enterprise buyers facing quarterly budget reviews, this uncertainty translates to a stark question, as Ortner explains:

You're not just trying to answer 'what broke', but how quickly can you find it, and how much infrastructure are you burning while you try?

It's the difference between AI as operational enhancement and AI as budget liability. This represents a crucial reality check for enterprise buyers. AI augmentation shows genuine promise, but within clearly defined operational boundaries.

Economic pressures reshaping budget priorities

Ortner reveals significant budget pressure affecting observability investments. He describes dramatic evolution in customer spending on monitoring and related personnel:

When I started about 20 years ago, there was no cost issue because the data collected was so simple and scarce. But then more things were added, more telemetry signals, and other things were added.

Budget allocations have shifted from five percent to 30% of total infrastructure spending before enterprises pushed back. Ortner continues:

We have quite a few customers that are there, but they all say it's way too much. That's not a healthy ratio. It's gotta go down.

This economic reality is driving enterprises toward solutions that demonstrate clear operational value rather than comprehensive data collection. Customers are increasingly asking "how many alerts do I get? How many of them are real? Then how many war rooms do I have to create? How many people are in them? How long do they take?"

These questions show how observability ROI is being evaluated, moving beyond feature comparisons of what it can do, but what it tangibly improves for an organization.

Ortner describes Dynatrace's approach to product development as notably different from typical enterprise software companies. The organization conducts systematic customer engagement that goes well beyond standard feedback mechanisms, he explains:

We are inviting about 20 to 30 customers every year in strategic customer briefings to our R&D HQ, either in Linz or Vienna, Austria.

More significantly, engineers across Dynatrace's European development centers participate directly in customer conversations. This direct engineering-customer connection addresses a common enterprise software problem – the disconnect between product development and operational reality – by giving engineers the opportunity to better understand the customer pain-points and future goals. As Ortner emphasizes: 

If engineers only see problems through the lens of support, it kills the innovation spirit. They need to hear unfiltered feedback – sometimes harsh, but motivating.

Business integration

The most substantial development in observability concerns its expanding scope beyond traditional IT operations. Ortner describes an example of advanced implementations that now correlate external business events with system performance: 

Someone might say 'Nothing has changed, but now we have fewer people using our e-commerce site or our mobile app'. But the reason for that is an external event – it could be a marketing event or maybe a system change in your finance backend, where suddenly the price quotes are slightly higher and you are not competitive anymore.

This integration represents observability's evolution from reactive problem-solving to proactive business intelligence. Rather than simply detecting and resolving incidents, enterprises are using observability platforms to understand the relationship between technical performance and business metrics.

The implications extend to organizational structure. Ortner notes that mature organizations want assurance that systems are not just working, but work end-to-end, perform adequately and are right for usability – because "even if it's super fast, it's no good if you lose people across the use case."

Recent academic research validates Ortner's observations about AI limitations in incident management. Microsoft's study of LLMs in Azure incident analysis – which Ortner referenced during our conversation – examined whether LLMs could replicate or assist site reliability engineering teams when equipped with properly structured knowledge bases.

The research found nuanced results – models could often identify likely root causes, but with unpredictable timeframes and substantial computational costs. The study concluded that LLMs could become valuable triage tools, particularly when tailored to different operational roles, but only within clearly defined limits. The bigger the scope, the more unpredictable the performance.

This research underscores the gap between AI marketing promises and operational reality, supporting Ortner's emphasis on augmentation rather than autonomy.

Industry developments reveal observability platforms transitioning from operational tools to strategic business enablers. This shift demands different evaluation criteria – rather than focusing primarily on technical capabilities, procurement decisions increasingly center on operational efficiency, business correlation, and economic justification.

Ortner's experiences suggest that successful observability implementations require not just technical sophistication, but organizational alignment between engineering, operations, and business stakeholders. The "use case availability" concept is a prime example of this integration – measuring success based on business outcomes rather than purely technical metrics.

My take

What Dynatrace is seeing is that customers have several critical considerations – the platform's ability to correlate technical and business metrics, the economic efficiency of data collection and analysis, and vendor commitment to understanding operational context rather than simply delivering features.

Ortner's pragmatic perspective cuts through observability market hype to reveal a discipline undergoing fundamental transformation. The "use case availability" framework is particularly compelling because it represents a fundamental shift in how we think about infrastructure – from keeping systems running to understanding whether they're actually serving their intended purpose. This isn't just semantics – it's recognition that infrastructure work is analytical and strategic, not just operational maintenance.

This alignment between IT operations and business outcomes has been a challenge for businesses for decades, and observability could be an important way to deliver it.

Most importantly, Ortner's measured skepticism about AI capabilities provides essential grounding for enterprise buyers drowning in vendor promises about autonomous operations. His conclusion that "observability might be a prime candidate for AI support, but only if the ground rules are clear – and the model understands when to stop" should be required reading for any organization evaluating AI-enhanced observability platforms.

The economic pressures Ortner describes – customers pushing back against 30% infrastructure spending on monitoring – suggest the market is reaching a point where vendors must demonstrate clear operational value rather than comprehensive feature sets. This shift toward ROI-focused evaluation is the opportunity to see the difference between sustainable observability strategies and expensive monitoring sprawl.