95% of AI Pilots Fail — Here’s How to Be the 5%

Dallon Robinette
August 29, 2025

When MIT released research showing that 95% of enterprise AI pilots fail to deliver measurable business impact, it made headlines for a reason. After years of heavy investment in artificial intelligence, the vast majority of organizations still haven’t moved beyond pilots that promise much but deliver little.

This doesn’t mean AI itself is broken. In most cases, the technology performs as intended. What fails is the ability to take those pilots out of the lab and into the organization in a way that creates measurable outcomes. That’s the real lesson of the MIT report, and it should reshape how leaders think about their AI strategies going forward.

Why So Many Pilots Stumble

Pilots fail for many reasons that have little to do with underlying algorithms. The technology often performs exactly as intended. The real challenge lies in how organizations prepare for, prioritize, and ultimately adopt it.

Some of the most common pitfalls include:

Data readiness is overlooked: AI can’t deliver without a clean, integrated foundation of metrics, logs, and event data. Many pilots fail before they even begin because organizations can’t ingest, normalize, and correlate data at scale. This problem is especially acute in network and infrastructure operations, where the sheer volume and variety of metrics, logs, and events is overwhelming. Traditional LLMs weren’t designed for this type of data.
Pilots that never scale: AI projects often start with a narrowly defined scope. They solve a specific problem in a controlled environment, but there’s no roadmap for scaling that solution across business units, geographies, or functions. The result is innovation stuck in a lab.
Misaligned Investment: Budgets flow disproportionately toward highly visible projects like customer-facing chatbots, sales enablement, or marketing personalization, while the real treasure sits in less glamorous areas. Automating claims, streamlining procurement, or accelerating finance operations may not make headlines, but they consistently deliver stronger ROI.
Governance Gaps: Without clear policies on risk, compliance, and accountability, organizations hesitate to expand beyond pilots. The lack of governance creates hesitation at the exact moment when momentum should build.
Underestimating integration: Finally, the most challenging part: weaving AI into day-to-day workflows. AI is not plug-and-play. It requires process redesign, training, and cultural adoption. Without these, employees default to old ways of working, leaving the technology underutilized.

Each of these challenges compounds the others. That’s why many organizations end up with proofs of concept that demonstrate promise but never deliver sustained business value.

Data Readiness: The Foundation Most AI Pilots Miss

One of the biggest reasons AI initiatives fail is that they start on shaky ground. AI only delivers value if the underlying data is accurate, integrated, and available at scale. Yet for most organizations, data is fragmented across silos, inconsistent in quality, and challenging to unify.

This is especially true in network and infrastructure operations. The raw material for insight comes in the form of metrics, logs, and events — massive streams of telemetry that traditional LLMs were never built to handle. Without a way to ingest and correlate this data, AI initiatives can’t progress beyond surface-level pilots.

For many enterprises, this is the unspoken roadblock. No matter how sophisticated the model, without trusted data to fuel it, AI cannot succeed.

The Cost of Staying in Pilot Mode

The risks of stalled AI adoption extend well beyond wasted budgets. Over time, organizations face three compounding challenges:

Pilot Fatigue: Teams lose energy when project after project fails to translate into real change. The appetite for innovation erodes.
Shadow IT: Employees, eager to boost their productivity, adopt consumer AI tools like ChatGPT on their own. While this demand signals opportunity, it also introduces security, compliance, and data leakage risks.
Competitive Drift: The minority of organizations that successfully adopt new practices are already building efficiency, resilience, and differentiation. Every year spent stuck in pilot mode makes it harder to catch up.

This is why the 95% failure statistic matters so much: not because it proves AI doesn’t work, but because it shows how few organizations are positioned to capture its value.

What Successful Organizations Do Differently

The 5% of organizations that break through share a common set of behaviors. They treat AI not as a science experiment, but as a strategic capability.

They start with clear outcomes: Rather than chasing novelty, they begin with business problems that matter — in our case, reducing downtime, accelerating troubleshooting, or cutting operational costs — and apply AI as a lever to achieve them.
They plan for adoption: Success requires more than a working model. That means thinking about integration into workflows, change management, and training from day one rather than as an afterthought once the model is built.
They invest in data readiness: They ensure data is accessible, trustworthy, and aligned with the problems they want to solve. Without that foundation, scaling is impossible.
They leverage partnerships: The companies that succeed most consistently are not trying to reinvent the wheel. They partner with vendors who bring proven platforms and expertise, freeing internal teams to focus on areas where the business is truly differentiated.

These aren’t isolated practices. They add up to an approach that treats AI as a business capability rather than an experiment. That mindset makes all the difference.

The Reality of the “Learning Gap”

MIT researchers described the adoption barrier as a “learning gap”: the distance between what AI can technically achieve and how organizations adapt to use it effectively.

In practice, the learning gap often looks like this:

Business leaders hear promises of transformational impact and expect results in months.
Operational teams are tasked with delivering, but lack the processes, training, or governance to put AI into practice.
Employees experiment with new tools but fall back on familiar workflows when adoption feels disruptive.
Momentum stalls, projects are shelved, and enthusiasm gives way to skepticism.

The gap has nothing to do with intelligence and everything to do with alignment. And closing it requires more than technology. It requires an approach that blends innovation with integration, pairing AI capability with organizational readiness.

How Selector Helps Close the Gap

This is where Selector is different. Our platform was designed to ingest virtually any type of data, from virtually any source — whether structured or unstructured, real-time or historical. By normalizing and enriching these feeds, Selector turns messy operational data into a clean, trusted foundation for AI. It’s the heavy lifting most organizations struggle with, and it’s what enables everything else: correlation, root cause analysis, and measurable outcomes.

Native understanding of raw operational data: Selector was built to ingest, normalize, and analyze the massive volumes of metrics, logs, and events that define enterprise networks and infrastructure. Where most AI tools struggle with unstructured data, we do the heavy lifting natively. This is the foundation for every outcome we deliver.
From pilot to impact: Our customers don’t stall at the proof-of-concept stage. They expand. That’s why we’ve achieved 170% net revenue retention, because once organizations start with Selector, they keep growing with us.
AI that integrates naturally: We meet teams where they already work with the tools they already use, surfacing insights inside collaboration tools like Slack or Teams instead of creating another siloed dashboard.
Outcomes that matter: We focus on metrics business leaders care about, like improved uptime, faster troubleshooting, and lower operational costs. AI adoption succeeds when its impact is evident in business performance.
Trusted partnership: Our role doesn’t end with technology. We guide organizations through the process and cultural change required to turn adoption into a competitive advantage.

This combination of platform and expertise enables our customers to avoid the 95% trap and realize the full potential of AI.

A Smarter Path Forward

Every major technology shift follows a similar arc: initial hype, a period of disappointment, and then eventual maturity as organizations learn how to use it effectively. AI is no different.

The MIT study would not be read as a reason to retreat from AI investment. It should be seen as a signal to invest differently:

Prioritize outcomes over experiments.
Focus on adoption, not just innovation.
Build the right foundation for data and governance.
Partner with organizations that know how to move from pilot to impact. ‘

The companies that do this are already separating themselves from the pack.

Becoming the 5%

The MIT report highlighted a sober reality: most AI pilots fail. But AI itself isn’t failing. In fact, it is only getting started. The challenge is adoption, and the organizations that solve adoption will be the ones that define the next decade of enterprise performance. The MIT report also points toward a clear opportunity. Organizations that close the adoption gap will stop struggling with stalled experiments and start seeing lasting impact.

At Selector, we help enterprises make that leap, not by adding another pilot, but by turning AI into a capability that drives measurable business outcomes. By starting with raw operational data from any source, embedding AI into workflows, and focusing relentlessly on measurable outcomes, we enable organizations to realize the full value of their AI investment.

Learn more about how Selector’s AIOps platform can transform your IT operations.

To stay up-to-date with the latest news and blog posts from Selector, follow us on LinkedIn or X and subscribe to our YouTube channel.

More on our blog

April 10, 2026
Bob Slevin

AIOps, Observability

Beyond the Dashboard: Selector’s Patented Approach to Conversational Observability

For years, IT operations teams have been trapped in a frustrating paradox: the data they need to solve critical issues is right at their fingertips, yet entirely out of reach. Accessing it requires engineers to master complex, platform-specific query languages, dig through endless layers of dashboards, and hunt for the exact visualization that holds the answer. Under the intense pressures of modern speed, scale, and complexity, this rigid model is breaking down. At Selector, we recognized a fundamental opportunity to change how teams interact with their data. Our recently published U.S. patent application (US20250278401A1, filed March 2, 2024, and published September 4, 2025), titled “Dashboard metadata as training data for natural language querying,” outlines a transformative solution. By utilizing dashboard metadata, aliases, and user interaction data as training material, we empower operators to bypass structured queries entirely and obtain infrastructure insights using plain, natural language. The Core Innovation: Dashboards as Operational Intelligence Historically, dashboards have been viewed as the final destination for data—a static visualization tool. Selector flips this paradigm. We treat the dashboard as a rich source of operational intelligence and a single version of the truth. The patent details how our platform uses existing dashboard metadata to build dynamic alias datasets. These datasets effectively map natural language phrases to the correct operational context. Because the system leverages the existing organization, labeling, and usage patterns already established in an environment, it doesn’t have to learn from scratch with every user request. It already speaks your network’s language. Changing the Operator Experience This approach fundamentally redefines the operator experience. Instead of forcing an engineer to “think like a query engine,” Selector allows them to simply “think like an operator.” When an issue arises, a user can ask, “Why is packet loss increasing in the west region?” without needing to hunt through widgets or write complex syntax. The system instantly interprets the natural language request, identifies the necessary context, generates the underlying database queries, and returns real-time (or near-real-time) performance data. Capturing “Tribal Knowledge” This innovation goes far beyond a UI upgrade; it represents a major shift in how operational knowledge is institutionalized. Most operations centers rely heavily on “tribal knowledge”—the unwritten expertise of senior staff who inherently know which metrics matter, which dashboards to check, and what specific terms mean in their unique environment. Selector’s patented method converts this implicit expertise into durable training data. As users interact with the system, their natural language inputs continuously augment the alias dataset. The model aligns itself with the customer’s actual domain language, growing smarter and more accurate over time. Scaling Operations and Lowering the Skill Barrier For teams tasked with managing unprecedented scale, this adaptive approach is a game-changer. Traditional natural-language-to-query systems often fail because they require constant manual labeling and retraining whenever new terminology emerges. Selector’s patent directly solves this inefficiency. Our adaptive method automatically updates the alias dataset based on dashboard metadata and user language, even extrapolating new query templates before they are explicitly encountered. This drastically reduces the need for manual labeling while driving high relevance in highly specific, domain-heavy environments. The operational benefits are immediate and measurable: The Architectural Foundation Crucially, this conversational layer doesn’t exist in a vacuum—it is built on a powerful architectural foundation. The patent describes an operations management system capable of ingesting, normalizing, and labeling heterogeneous operational data from multiple sources before generating and executing queries against it. For AIOps and observability, this highlights a foundational truth: natural language querying is only effective when it rests atop a platform that is already proficient in data correlation, normalization, and contextual retrieval. The Future is Conversational Ultimately, this isn’t just about making dashboards easier to use. It is about transforming the relationship between humans and operational systems. Dashboards are evolving from passive displays into active learning agents. By moving operational data beyond static visualization and into the realm of conversational access, Selector lets the system learn the operator’s language—rather than forcing the operator to learn the system’s. We are delivering on the ultimate promise of AIOps: turning your operational data into a resource you can converse with, rather than just a dashboard you have to search. Stay Connected Selector is helping organizations move beyond legacy complexity toward clarity, intelligence, and control. Stay ahead of what’s next in observability and AI for network operations:

April 3, 2026
Dallon Robinette

AIOps, Observability

The Business Case for AI-Driven Observability in Network Operations

Modern network operations generate an extraordinary amount of telemetry. Metrics, logs, events, topology data, cloud signals, and service context all contribute to a richer picture of system behavior. As environments expand across cloud, data center, edge, and SaaS, the opportunity for operations teams is clear: when that telemetry is unified and understood in context, it becomes a powerful source of resilience, efficiency, and business insight. That is why AI-driven observability has become such an important priority for IT and operations leaders. Its value comes from helping teams move through complex environments with greater clarity. Correlated signals, contextual awareness, and shared operational understanding help teams identify issues faster, coordinate more effectively, and resolve incidents with greater confidence. For business leaders, the conversation is increasingly practical. They want to understand how observability investments contribute to uptime, team productivity, operational scale, and service quality. AI-driven observability answers that question by connecting technical insight to measurable operational outcomes. AI-Driven Observability Creates Shared Operational Context One of the most valuable outcomes in modern operations is shared context. Network, infrastructure, cloud, and application teams all work with data that reflects real conditions in the environment. When that information is connected across domains, teams can align quickly around what is happening, what is affected, and where to focus first. Previous articles we’ve written point to this operational need consistently. Full-stack visibility, event correlation, data harmonization, and contextual intelligence all support the same outcome: helping teams see systems as interconnected environments. This gives engineers a clearer path from telemetry to understanding, and it helps leaders create more consistent operational workflows across distributed environments. Shared context also improves collaboration during incidents. A unified operational view helps teams work from the same narrative, which supports faster triage, clearer ownership, and smoother coordination across functions. In high-pressure moments, that alignment has direct business value because it reduces confusion, accelerates decisions, and supports service continuity. Business Value Begins With Faster Understanding In many organizations, the most important operational gain comes from shortening the path to understanding. When engineers have access to correlated, context-rich insight, they can move quickly from detection to investigation and from investigation to action. That acceleration matters because every operational delay carries a cost. Teams invest time in triage, collaboration, handoffs, and escalation. Business services may experience degraded performance. Internal teams lose productivity. Customer-facing systems carry increased risk. AI-driven observability supports a more efficient operating model by helping teams understand relationships across signals and by surfacing the context needed to act earlier in the incident lifecycle. This is one of the clearest ways to express the value of AI-driven observability to executive audiences. Faster understanding improves incident response, strengthens operational discipline, and helps organizations sustain service quality as complexity grows. The Metrics That Show Real Value A strong business case becomes much easier to communicate when it is anchored in a focused set of operational metrics. MTTR Mean Time to Resolution remains one of the clearest indicators of operational effectiveness. AI-driven observability contributes to MTTR improvement by helping teams identify likely cause, affected services, and relevant context earlier in the process. This supports a faster path to action and a more efficient incident lifecycle. Time to Identify Early understanding shapes the rest of the response cycle. A clear view of correlated events, dependencies, and service impact helps teams assign ownership quickly and move forward with confidence. Incident and Ticket Volume Correlated incident management supports a more focused operating model. When related events are grouped into context-rich incidents, teams can work from a smaller number of more meaningful operational objects. That improves efficiency and helps reduce cognitive load across NOC and operations teams. Escalation Patterns High-quality context supports better decision-making at every level of the organization. It allows frontline teams to act with stronger situational awareness and helps senior engineers focus their expertise where it can create the greatest impact. This contributes to healthier team capacity and more scalable operations. Operational Toil Operations leaders increasingly care about the amount of repetitive manual work surrounding incidents: reviewing duplicate alerts, switching across tools, reconstructing timelines, and coordinating repeated handoffs. AI-driven observability supports a cleaner, more streamlined workflow that improves engineer productivity and creates a better day-to-day operating experience. Translating Operational Gains Into Executive Language Executive stakeholders respond most strongly when technical improvements are connected to business outcomes. AI-driven observability lends itself well to that conversation because the operational gains are tangible. Time saved during triage translates into labor efficiency. Faster resolution supports uptime and service quality. More focused incidents help teams scale their efforts across larger, more distributed environments. Better context strengthens planning, prioritization, and cross-team coordination. These outcomes support resilience while also contributing to cost discipline and organizational agility. This is especially important in hybrid operations, where service performance depends on relationships across infrastructure, network paths, providers, and applications. In these environments, observability creates value by helping organizations understand system behavior holistically and act with a stronger operational foundation. AI-Driven Observability Supports Resilient Growth As digital environments grow, the need for clarity grows with them. More services, more interdependencies, and more distributed architectures all increase the importance of context-rich operational intelligence. AI-driven observability helps organizations meet that complexity with a model that supports resilience and scale. Data harmonization, event intelligence, natural language access, intelligent incident management, and agentic workflows all contribute to a future where operational teams can work with greater speed, confidence, and precision. That progression begins with observability that understands relationships across the environment and delivers insights in a form teams can use immediately. A Simple Framework for Proving Value For teams building the business case internally, the clearest approach is often the simplest. Start by establishing a baseline for incident response, escalation patterns, and operational effort. Track the time spent identifying issues, coordinating across teams, and resolving events. Then measure how AI-driven observability improves those workflows through richer context, better alignment, and faster understanding. From there, tie those improvements to the outcomes leadership cares about most: service reliability, productivity, operational scale, and customer experience. This gives observability

March 27, 2026
Dallon Robinette

AIOps

Solving the Ticket Noise Problem: What We Learned from Our ServiceNow Webinar

On March 18th, we hosted a session focused on a challenge that continues to undermine even the most mature IT operations teams: ticket noise. It’s easy to dismiss noise as just “too many alerts”. But as we explored in the webinar, the real issue runs deeper. Ticket noise is a symptom of something more fundamental — a lack of correlation, context, and shared visibility across the stack. If you weren’t able to attend, this blog walks through the key ideas, examples, and takeaways. And if any of this feels familiar, it’s worth watching the full session. View “Solving the Ticket Noise Problem: Bringing Intelligence to ServiceNow”. The Hidden Cost of Tickets Most organizations don’t struggle because they lack monitoring. In fact, the opposite is true — they have too much of it. Over time, teams adopt specialized tools for every layer of the environment: Each tool does its job well within its domain, but incidents don’t respect those boundaries. As discussed in the webinar, what emerges is a fragmented operational model: The result is a familiar pattern: alert storms, duplicated effort, and delayed resolution. To put things more bluntly, it becomes a data correlation problem rather than a monitoring problem. Why Traditional ITSM Workflows Break Down Platforms like ServiceNow are central to how organizations manage incidents, but they are only as effective as the data they receive. When upstream systems generate noisy, uncorrelated alerts, ServiceNow becomes a reflection of that chaos: In the webinar, we walked through a scenario that highlights this breakdown. A single configuration change triggered an outage, resulting in dozens of alerts across different tools and teams. Each team began investigating independently, without a shared understanding of the root cause. What should have been a single incident turned into a multi-team firefight, slowing resolution and increasing operational risk. Rethinking the Model: From Alerts to Event Intelligence The core idea behind Selector’s approach is simple but powerful: Don’t manage alerts. Understand Events. Instead of treating ever alert as a separate signal, Selector ingests telemetry across the entire stack — network, infrastructure, application, and cloud — and builds a correlated model of what’s actually happening. This shift fundamentally changes how incidents are handled: This is what we refer to as event intelligence — the ability to move from raw signals to actionable insight. What This Looks Like Inside ServiceNow One of the most important aspects we covered in the webinar is how this intelligence translates into real operational workflows. Selector integrates directly with ServiceNow, but not in the traditional “forward alerts as tickets” sense. Instead, it transforms the structure and quality of what enters the system. Fewer Tickets, Higher Signal Rather than flooding ServiceNow with every alert, Selector creates correlated incidents. In one example shared during the session, a large-scale outage generated thousands of alerts in a traditional tool. Selector reduced that to just a handful of meaningful incidents, with each tied to a clear root cause. This dramatically reduces the cognitive load on engineers and allows teams to focus on resolution instead of triage. Bi-Directional Intelligence Another key differentiator is the bi-directional integration between Selector and ServiceNow. Selector doesn’t just push tickets into ServiceNow, but instead continuously exchanges information: This ensures that both systems remain aligned and eliminates the fragmentation that often occurs between monitoring and ITSM. It also enables smarter workflows, such as: From Basic Tickets to Actionable Context Perhaps the most meaningful shift is in the quality of each ticket. Traditional tickets often require engineers to begin their investigation from scratch. Selector changes that by embedding context directly into the incident: In effect, Selector elevates tickets from simple notifications to decision-ready artifacts, reducing the need for manual investigation and accelerating time to resolution. Real-World Examples To make this tangible, we walked through several real-world scenarios during the webinar. In one case, a failure in a network interface caused cascading issues across multiple access points. Without correlation, this would appear as a series of unrelated device failures. With Selector, the system identified the failing interface as the root cause and generated a single, context-rich incident, allowing the team to resolve the issue in under 30 minutes. In another example, a large SD-WAN outage impacted over 100 devices across dozens of sites. While other tools generated thousands of alerts, Selector reduced the situation to just a few actionable incidents. Engineers were able to coordinate quickly and focus on resolution rather than filtering out noise. These aren’t edge cases. These represent what happens when correlation is applied at scale. Why This Matters Now As environments become more distributed and complex, the cost of noise continues to rise. It’s not just about wasted time, but also: The organizations that succeed are the ones that move beyond monitoring and toward intelligent operations, where systems don’t just detect issues, but help explain and resolve them. The Takeaway Ticket noise isn’t solved by adding more filters, rules, or dashboards. It’s solved by changing how data is understood. By correlating events across the stack and delivering that intelligence into systems like ServiceNow, Selector enables teams to: Watch The Full Webinar This blog captures the core ideas, but the full session goes deeper into: Watch the full webinar on demand here. Stay Connected Selector is helping organizations move beyond legacy complexity toward clarity, intelligence, and control. Stay ahead of what’s next in observability and AI for network operations:

On this page