Flow Data Ingestion: Closing the Visibility Gap Without the Complexity

For years, enterprises have been told a comforting story: telemetry is telemetry. Logs are logs. If you can collect, normalize, and route data efficiently, you can support both observability and security from the same pipeline.

At first glance, this sounds efficient. One ingestion layer. One set of collectors. One routing engine. Lower cost. Cleaner architecture. But this story hides a fundamental mistake.

Observability, telemetry, and security telemetry are not simply two consumers of the same data stream. They are different classes of data with distinctintents, time horizons, economic models, and failure consequences.

The issue is intent. This is what we at Databahn call the Telemetry Intent Gap: the structural difference between operational telemetry and adversarial telemetry. Ignoring this gap is quietly eroding security outcomes across modern enterprises.

The Convenient Comfort of ‘One Pipeline’

The push to unify observability and security pipelines didn’t stem from ignorance. It stemmed from pressure. Exploding data volumes and rising SIEM costs which outstrip CISO budgets and their data volumes are exploding. Costs are rising. Security teams are overwhelmed. Platform teams are tired of maintaining duplicate ingestion layers. Enterprises want simplification.

At the same time, a new class of vendors has emerged,positioning themselves between observability and security. They promise a shared telemetry plane, reduced ingestion costs, and AI-powered relevance scoring to “eliminate noise.” They suggest that intelligent pattern detection can determine which data matters for security and keep the rest out ofSIEM/SOAR threat detection and security analytics flows.

On paper, this sounds like progress. In practice, it risks distorting security telemetry into something it was never meant to be.

Observability reflects operational truths, not security relevance

From an observability perspective, telemetry exists to answer a narrow but critical question: Is the system healthy right now? Metrics, traces, and debug logs are designed to detect trends, analyze latency, measure error rates, and identify performance degradation. Their value is statistical. They are optimized for aggregation, sampling, and compression. If a metric spike is investigated and resolved, the granular trace data may never be needed again. If a debug logline is redundant, suppressing it tomorrow rarely creates risk. Observability data is meant to be ephemeral by design: its utility decays quickly, and its value lies in comparing the ‘right now’ status to baselines or aggregations to evaluate current operational efficiency. 

This makes it perfectly rational to optimize observability pipelines for:

·      Volume reduction

·      Sampling

·      Pattern compression

·      Short- to medium-term retention

The economic goal is efficiency. The architectural goal isspeed. The operational goal is performance stability. Now contrast that with security telemetry.

Security telemetry is meant for adversarial truth

Security telemetry exists to answer a very different question: Did something malicious happen – even if we don't yet know what or who it is?

Security telemetry is essential. Its value is not statistical but contextual. An authentication event that appears benign today may become critical evidence two years later during an insider threat investigation. A low-frequency privilege escalation may seem irrelevant until it becomes part of a multi-stage attack chain. A lateral movement sequence may span weeks across multiple systems before becoming visible. Unlike observability telemetry, security telemetry is often valuable precisely because it resists pattern compression.

Attack behavior does not always conform to short-term statistical anomalies. Adversaries deliberately operate below detection thresholds. They mimic normal behavior. They stretch activity over long time horizons. They exploit the fact that most systems optimize for recent relevance. Security relevance is frequently retrospective, and this is where the telemetry intent gap becomes dangerous.

The Telemetry Intent Gap

This gap is not about format or data movement. It is about the underlying purpose of two different types of data. Observability pipelines are meant to uncover and track performance truth, while security pipelines are meant to uncover adversarial truth.

Observability asks: Is this behavior normal? Is the data statistically consistent? Security asks: Does the data indicate malicious intent? In observability, techniques such as sampling and compression to aggregate and govern data make sense. In security, all potential evidence and information should be maintained and accessible, and kept in a structured, verifiable manner. Essentially, how you treat – and, at a design level, what you optimize for – in your pipeline strongly impacts outcomes. When telemetry types are processed through the same optimization strategy, one of them loses. And in most enterprises, the cost of retaining and managing all data puts the organization's security posture at risk.

The Rise of AI-powered ‘relevance’

In response to cost pressure, a growing number of vendors catering to observability and security telemetry use cases claim to solve this problem with AI-driven relevance scoring. Their premise is simple: use pattern detection to determine which logs matter, and drop/reroute the rest. If certain events have not historically triggered investigations or alerts, they are deemed low-value and suppressed upstream.

This approach mirrors observability logic. It assumes that medium-term patterns define value. It assumes that the absence of recent investigations or alerts implies no or low risk. For observability telemetry, this may be acceptable.

For security telemetry, this is structurally flawed. Security detection itself is pattern recognition – but of a much deeper kind. It involves understanding adversarial tradecraft, long-term behavioral baselines and rare signal combination that may never have appeared before. Many sophisticated attacks accrue slowly, and involve malicious action with low-and-slow privilege escalation, compromised dormant credentials, supply chain manipulation, and cloud misconfiguration abuse. These behaviors do not always trigger immediate alerts. They often remain dormant until correlated with events months or years later.

An observability-first AI model trained on short-term usage patterns may conclude that such telemetry is "noise". It may reduce ingestion based on absence of recent alerts. It may compress away low-frequency signals. But absence of investigations is not the absence of threats. Security relevance is often invisible until context accumulates. The timeline over which security data would find relevance is not predictable, and making short and medium-term judgements on the relevance of security data is a detriment to long-horizon detection and forensic reconstruction.

When Unified Pipelines Quietly Break Security

The damage does not announce itself loudly. It appears as:

·      Missing context during investigations

·      Incomplete event chains

·      Reduced ability to reconstruct attacker movement

·      Inconsistent enrichment across domains

·      Silent blind spots

Detection engineers often experience this in terms of fragility: rules are breaking, investigations are stalling, and data must be replayed from cold storage – if it exists. SOC teams lose confidence in their telemetry, and the effort to ensure telemetry 'completeness' or relevance becomes a balancing act between budget and security posture.

Meanwhile, platform teams believe the pipeline is functioning perfectly – it is running smoothly, operating efficiently, and cost-optimized. Both teams are correct, but they are optimizing for different outcomes. This is the Telemetry Intent Gap in action.

This is not a Data Collection issue

It is tempting to frame this as a tooling or ingestion issue. But this isn't about that. There is no inherent challenge in using the same collectors, transport protocols, or infrastructure backbone. What must differ is the pipeline strategy. Security telemetry requires:

·      Early context preservation

·      Relevance decisions informed by adversarial models, not usage frequency

·      Asymmetric retention policies

·      Separation of security-relevant signals from operational exhaust

·      Long-term evidentiary assumptions

Observability pipelines are not wrong. They are simply optimized for a different purpose. The mistake is in believing that the optimization logic is interchangeable.

The Business Consequence

When enterprises blur the line between observability and security telemetry, they are not just risking noisy dashboards. They are risking investigative integrity. Security telemetry underpins compliance reporting, breach investigations, regulatory audits, and incident reconstruction. It determines whether an enterprise can prove what happened – and when.

Treating it as compressible exhaust because it did not trigger recent alerts is a dangerous and risky decision. AI-powered insights without security context will often over index on short and medium term usage patterns, leading to a situation where the mechanics and costs of data collection obfuscate a fundamental difference in business value.

Operational telemetry supports system reliability. Security telemetry supports enterprise resilience. These are not equivalent mandates, and treating them similarly leads to compromises on security posture that are not tenable for enterprise stacks.

Towards intent-aware pipelines

The answer is not duplicating infrastructure. It is designing pipelines that understand intent. An intent-aware strategy acknowledges:

·      Some data is optimized for performance efficiency

·      Some data is optimized for adversarial accountability

·      The same transport can support both, but the optimization logic – and the ability to segment and contextually treat and distinguish this data – is critical

This is where purpose-built security data platforms are emerging – not as generic routers, and not as observability engines extended into security, but as infrastructure optimized for adversarial telemetry from the start.

Platforms designed with security intent as their core – and not observability platforms extending into the security 'use case – do not define the value of data by their recent pattern frequency alone. They are opinionated, have a contextual understanding of security relevance, and are able to preserve and even enrich and connect data to enable long-term reconstruction. They treat telemetry as evidence, not exhaust.

That architectural stance is not a feature. It is a philosophy. And it is increasingly necessary.

Observability and Security can share pipes – not strategy

The enterprise temptation to unify telemetry is understandable. The cost pressures are real. The operational fatigue is real. But conflating optimization logic across observability and security is not simplification. It is misalignment. The future of enterprise telemetry is not a single, flattened data stream scored by generic AI relevance. It is a layered architecture that respects the Telemetry Intent Gap.

The difference between operational optimization and adversarial investigation can coexist and share infrastructure, but they cannot share strategy. Recognizing this difference may be one of the most important architectural decisions security and platform leaders make in the coming decade.

Before starting Databahn, we spent years working alongside large enterprise security teams. Across industries and environments, we kept encountering the same pattern: the increased sophistication of platform and analytics in modernized stacks, matched by the fragility of the security data layer.  

Data is siloed across tools, movement is inefficient, lineage is a mystery that requires investigation. Governance is inconsistent, and management is a manual exercise leaning heavily on engineering bandwidth not being spent on delivering clarity, but in keeping systems going despite obvious gaps. Every new initiative depended on data that was harder to manage than it should have been. It became clear to us that this was not an operational inconvenience but a structural problem.

We started Databahn with a simple conviction: that to improve detection logic, ensure scalable AI implementation, and accelerate and optimize security operations, security data itself has to be made to work. That conviction has driven every decision we have made.

This week, we shared that Databahn has grown by more than 400% year-on-year, with more than half of our customers from the Fortune 500. We are deeply grateful to the enterprises, partners, and team members who have trusted us to solve this challenge alongside them. But the growth and traction are not the headline. It is that the security ecosystem is recognizing what we saw years ago – security data is the foundation of modern security operations.

Our strategy – staying focused

As the market evolves, companies face choices about where to direct their energy. There is always pressure to broaden and extend into adjacencies, or to join up and be absorbed by larger players in the security ecosystem.  

At Databahn, we remain singularly focused on solving the enterprise security data problem. Our customers and partners rely on us to be a best-of-breed solution for security data management, not a competitor attempting to replace parts of their ecosystem with new capabilities that dilute our mission.

Our belief is straightforward: enterprises don’t need another platform to own their stack, a new SIEM to detect threats, or a new Security Data Lake to store telemetry. They have these tools and have built their systems around them. What they need is a solution to make their security data work – not locked in, not siloed, not locked behind formats and schemas that take teams thousands of lines of code to uncover.

It needs to move cleanly across environments to different tools. It needs to be governed and optimized. It should support existing systems without creating friction. Building the security data system that delivers the right security data to the right place at the right time with the right context is the problem we are choosing to solve for our customers.

Enterprise adoption reflects a larger shift

The enterprises choosing Databahn are not experimenting; they are standardizing.  

A Fortune 100 global airline managed a complex SIEM migration in just 6 weeks, while ensuring that complex data types – flight logs, sensors, etc. were seamlessly ingested and managed across the organization. The result was a more resilient and controlled data foundation, ready for AI deployment and optimized for scale and efficiency.  

Sunrun reduced log volume by 70% while improving visibility across its complex and geographically distributed environment. That shift translated into meaningful cost efficiency and stronger signal clarity.  

Becton Dickinson brought structure and governance to its security data, transforming operational complexity of a multi-SIEM deployment into clarity by centralizing their security data into one SIEM instance in just 8 weeks while significantly lowering costs.

Working with these exceptional global teams to turn security data noise into manageable and optimized signal validates our conviction. Our growth is a reflection of this realization taking hold inside the enterprise – security data isn’t working right now, but it can be made to work.

Security Data is now strategic architecture

As enterprises accelerate modernization and AI-driven initiatives, expectations placed on data have fundamentally changed. Security data is no longer exhaust, but it is infrastructure. It is the platform on which the future AI-powered SOC must operate. It must be portable, governed, observable, and adaptable to new systems without forcing architectural trade-offs.  

Enterprises cannot build intelligent workflows on unstable data foundations, where teams can’t trust their telemetry, and so must trust their AI output based on that telemetry even less. Before you layer more intelligence on top of your security stack, you have to fix the data foundation. That’s why AI transformation is being led by Forward Deployment Engineers who are structuring and cleansing data before adding AI solutions on top. Databahn provides that foundation as a platform, delivering flexible resiliency and governance without the manual effort and tech debt.

What comes next

We believe the next chapter of enterprise security will be defined by organizations that treat security data as a strategic asset rather than an operational byproduct. Our commitment is to continue going deeper into solving that core problem. To strengthen partnerships across the ecosystem and help enterprises modernize their security architecture without being forced into unnecessary complexity or locked into a platform that prevents ownership of their data.

The momentum we announced this week is meaningful, but it is just the beginning of a movement. What matters more is what it represents. That enterprises need to make their security data actually work.  

We are excited to continue solving that challenge alongside the leaders driving this shift. The future holds many exciting new partnerships, product development, and other ways we can reduce complexity and increase ownership and value of security data. If any of these challenges seem relatable, we would invite you to get in touch with us to see if we can help.

The Open Cybersecurity Schema Framework (OCSF) was designed to solve a fundamental problem: security data fragmentation. By offering an open and shared format, OCSF brought a shared foundational understanding of what security data should be understood, driving consistency. But every security engineer who has implemented OCSF mappings encounters the same structural challenge within weeks: the unmapped object.

This is where normalization efforts stall.

The Unmapped Object, Defined

OCSF aims to resolve fragmentation in security data by providing a shared taxonomy, including categories, event classes, and a standardized attribute dictionary, so that telemetry from disparate sources can be queried, correlated, and analyzed consistently. But this structure runs into challenges for complex and unexpected security data which does not fall neatly into the framework.

OCSF defines the unmapped object as a catch-all container for source data that doesn't map cleanly to standardized attributes. When an engineer translates a firewall log into OCSF format, fields like source IP become src_endpoint.ip, usernames normalize to actor.user.name, and timestamps align to time. These are the wins.

But source telemetry routinely contains fields that have no standard home in the schema. An MFA status indicator from an identity provider. A proprietary risk score from an EDR vendor. A vendor-specific context field that detection logic depends on. These fields don't disappear, they land in unmapped.

The unmapped object preserves data. Technically, nothing is lost. But it creates a different kind of fragmentation: structured, queryable fields alongside unstructured data that requires custom parsing for every downstream consumer. Detection engineers writing correlation rules cannot rely on unmapped content without building source-specific logic. Analysts hunting for threats must know which unmapped fields exist and how to extract them. AI systems attempting to reason across security data cannot process unmapped content without additional transformation.

The richer the source telemetry, the more content ends up unmapped. As one industry analysis observed: "Data is custom, the richer the data the more unmappable it gets. Don't be surprised if the analyst goes digging into un-normalized fields rather than common normalized fields, because that's where the piece of information they needed was buried."

This is the structural tension at the heart of OCSF adoption.

Three Causes of Mapping Gaps

Three factors drive the unmapped problem.

Schema variability at the source. Security tools emit logs in formats designed for their own ecosystems, not for interoperability. Proprietary field names, nested structures, and vendor-specific semantics mean that mapping requires deep understanding of both source format and target schema. When a vendor changes their log format, adding fields or restructuring existing ones, transformation rules break. Fields that were previously mapped may suddenly require rework. The maintenance burden compounds across hundreds of data sources.

Partial schema coverage. Some security tools don't provide enough data or context to fully populate OCSF's structured fields. Missing event class identifiers, absent process metadata, incomplete user objects, these gaps force engineers to choose between extending the schema, dropping fields, or accepting degraded normalization fidelity. None of these options are cost-free.

Field conflicts and ambiguities. Different tools use the same field name for different purposes, or represent the same concept with incompatible structures. A "status" field in one product might indicate connection state; in another, it represents authentication outcome. Resolving these conflicts requires judgment calls that must be documented, maintained, and applied consistently across all pipelines handling that source.

Organizations adopt OCSF expecting unified telemetry. What they often get is a mix of cleanly mapped fields and growing unmapped buckets that require custom handling for every query, detection rule, and investigation workflow. Organizations report spending two to four months on comprehensive OCSF implementations, and that timeline assumes stable source schemas, which rarely hold.

Three Governance Approaches

Security teams handle unmapped content in three primary ways, each with distinct trade-offs.

Strict minimum mapping. Map only the fields explicitly required by OCSF for the chosen event class. Everything else goes to unmapped. This approach preserves data completeness and avoids over-engineering the mapping layer, but it pushes complexity downstream. Detection engineers must write custom parsers for unmapped content. Analysts lose the ability to correlate on fields that could have been standardized. This approach works for organizations with simple detection requirements or limited cross-source correlation needs. It fails when unmapped fields contain the precise context needed for threat hunting or incident response.

Schema extension. OCSF supports extensions, formal mechanisms to add custom attributes to existing event classes without breaking compatibility. For fields that are critical and represent long-term requirements, creating an organization-specific extension with dedicated attributes is the most robust solution. Extension requires registration to receive a unique identifier range, preventing conflicts with the core schema or other organizations' extensions. The process enforces governance but demands ongoing maintenance: schema registries to track which extensions exist, what they contain, and which pipeline versions support them. This approach fits organizations with mature data engineering capabilities and stable telemetry requirements. It struggles when sources change frequently or when teams lack resources for extension lifecycle management.

Dynamic classification. Rather than pre-defining every mapping or manually extending schemas, this approach uses intelligent systems to classify and route data in real time. The pipeline learns source schemas, identifies semantic equivalents, and maps fields dynamically, elevating what would be unmapped into structured, queryable attributes without manual rule creation.

This is where the architectural gap exists in most OCSF implementations. Static mapping rules cannot adapt to schema drift. Manual extension processes cannot scale to hundreds of sources. Unmapped buckets grow into ungovernable data graveyards.

Normalization as a Pipeline Problem

The unmapped field problem is fundamentally a pipeline architecture problem. Normalization happens in transit, not at rest. Decisions about what maps where, what extends the schema, and what falls into catch-all containers must be made in flight, at the speed of telemetry ingestion.

Traditional approaches treat parsing, normalization, and transformation as static configurations. Engineers write mapping rules, deploy them, and hope sources don't change. When they do, and sources always change, brittle pipelines break. Fields that should be normalized end up unmapped. Detection coverage degrades silently. No one notices until an investigation fails or a compliance audit surfaces gaps.

This is where Databahn comes in. The platform approaches OCSF normalization as a continuous, AI-assisted process rather than a one-time configuration exercise.

Cruz, Databahn's agentic AI, builds an understanding of source schemas as they evolve. It automatically detects schema drift and adapts transformations without manual intervention. When a vendor adds a new field or restructures an existing one, the pipeline doesn't break, it learns. Fields that would traditionally fall into unmapped buckets are intelligently classified and routed to appropriate schema locations, or flagged for extension when no suitable mapping exists.

The platform maintains schema consistency across heterogeneous data models in-line, rather than post-ingestion. This means SIEM correlation rules and detection logic operate on unified, structured data, not a mix of normalized fields and unmapped JSON blobs that require custom handling. Analysts spend time on investigations, not digging through catch-all containers for buried context.

For enterprises already navigating OCSF adoption with Amazon Security Lake, Microsoft Sentinel, or Splunk, Databahn provides the transformation layer that turns partial mappings into comprehensive normalization.

Governance Practices That Reduce Drift

Regardless of the approach chosen, governance practices reduce unmapped field accumulation over time.

Document every mapping decision. When a field goes to unmapped, record why. When an extension is created, define its scope and intended consumers. Detection engineers, data engineers, and analysts need to understand how data flows through the schema. Ambiguity compounds across teams and time.

Establish feedback loops with downstream consumers. The teams writing detection rules, hunting for threats, and investigating incidents are the first to know when unmapped fields contain critical context. Their pain points should drive mapping priorities and extension decisions.

Monitor unmapped field growth. If the unmapped object is expanding faster than structured fields, something is wrong architecturally. Either sources are changing faster than mappings can adapt, or the mapping strategy is too conservative for the organization's detection requirements.

Pin analytics and detection content to specific OCSF versions. Schema evolution is inevitable. Content repositories that reference explicit versions prevent breaking changes from silently degrading detection coverage when the schema updates.

The Architectural Choice

OCSF adoption is not a checkbox exercise. It is an architectural decision with downstream implications for every detection rule, investigation workflow, and AI application built on security data.

The unmapped field problem reveals a fundamental tension: static schemas cannot keep pace with dynamic telemetry. Organizations face a choice. Continue retrofitting manual mappings onto evolving sources, accepting growing unmapped buckets as inevitable. Or invest in infrastructure that treats normalization as a continuous, intelligent process, governance that happens in flight, not after storage.

The future of security data normalization is not more catch-all containers. It is pipelines that understand schemas, adapt to drift, and ensure that critical context reaches analysts and AI systems in structured, queryable form.