Start Brownfield Network Automation With the Data You Have

|

Oct 22, 2026

Most network teams don’t have the luxury of starting with a clean slate. Their infrastructure has evolved through years of hardware refreshes, acquisitions, emergency fixes, and changing operational requirements.

Along the way, information about the network has ended up in monitoring systems, network controllers, configuration backups, spreadsheets, and sometimes the heads of a few experienced engineers.

One of the most common mistakes in network automation is trying to build everything before you start. Teams assume they need a complete source of truth and a perfect data model before they can automate anything. They spend months designing schemas, collecting attributes, and resolving inconsistencies without delivering much operational value.

A more practical approach is to start small, identify a few quick wins, and use those successes to build momentum. You can begin automating with a limited dataset and gradually improve your source of truth as your requirements evolve.

Start with a clearly defined scope

The source of truth for a brownfield network needs a manageable starting point, even if your long-term ambition is to cover the entire infrastructure.

If you manage hundreds of sites across several countries, onboarding everything at once could become an enormous project. Instead, choose a part of the network you can clearly define and realistically complete.

That might be:

  • All network devices at one campus
  • The switching infrastructure across a group of branch offices
  • One data center fabric
  • All devices managed by a particular operational team

Be equally explicit about what’s excluded. If your initial scope covers campus switching, you might leave out provider-managed routers, wireless controllers, or passive cabling until a later phase.

Choose an initial use case that delivers value without introducing unnecessary risk. Read-only workflows are often a good starting point because they allow you to demonstrate the benefits of automation without making changes to production devices.

Automated configuration backups are one example. With a reliable device inventory, connection details, and appropriate credentials, you can automate a useful task without changing device configurations.

Early successes help demonstrate the value of automation and build confidence across the team. As your data becomes more complete and reliable, you can introduce higher-value workflows that require more effort and carry greater operational risk.

Build a complete foundation before adding detail

Once you’ve defined the scope, focus on getting the essential information for every device into your source of truth.

The priority at this stage is broad coverage with a limited set of reliable attributes. You can add more detail as your automation requirements evolve.

Suppose you’ve identified 200 switches and routers in your initial environment. You could spend weeks collecting dozens of attributes for the first 20 devices. Or you could establish a reliable inventory of all 200 using a small set of fields.

The second approach gets you to something useful much faster.

For a basic device inventory, you might start with:

  • Hostname or another unique identifier
  • Management IP address
  • Serial number
  • Device platform or model
  • Site or location
  • Operating system

The exact fields depend on your first use case. If you’re automating configuration backups, you’ll need enough information to select the correct connection method and authenticate each device.

Some modeling decisions deserve more attention upfront, particularly how you identify objects and represent their relationships.

Take a stack of switches. Depending on the platform, the stack may have one management address while containing multiple physical devices, each with its own serial number. If your model assumes every management IP represents exactly one physical device, you’ll run into trouble.

The same applies to chassis, controllers, line cards, and other components that don’t fit neatly into a one-device-one-record structure.

Model relationships explicitly rather than hiding them in text fields. A device’s relationship to a site, for example, should be represented as a relationship to a site object rather than an arbitrary string someone types into a location field.

When you model relationships across your infrastructure, you can query connected resources, validate dependencies, and make changes with a better understanding of what else may be affected.

Once the foundation is in place, add attributes as new workflows require them. For instance, if your next objective is to generate configurations based on device function, you might introduce what I call the three golden attributes:

  • Role: Access switch, distribution switch, core router
  • Type: The platform or implementation
  • Status: Provisioning, active, maintenance, decommissioned

I’ve written previously about how role, type, and status can make legacy infrastructure easier to automate. These attributes help you group and manage resources according to their function without depending on complicated naming conventions.

Add each attribute deliberately. Define its meaning, determine which objects require it, and establish how its value will be maintained. Every field you add creates another piece of information that can become inaccurate.

A flexible data model makes incremental development easier. With Infrahub, for example, you can define the object types, attributes, and relationships your environment needs, then evolve the schema as your automation requirements grow.

Use existing systems to discover and validate your network

You probably already have much of the information needed to populate your initial source of truth. It’s just scattered across multiple systems.

Your monitoring platform might have a reasonably complete list of devices and management addresses. Network controllers may know device models and software versions. Configuration backups can reveal interface configurations and other operational details.

Start by extracting the information you need from these existing sources. You don’t need an elaborate integration pipeline immediately. A CSV export, a Python script, or a spreadsheet may be enough to assemble and review your initial dataset.

Existing systems often disagree, so review the data carefully before importing it. A monitoring platform might still contain decommissioned devices. A backup tool might use a different hostname for the same device. Two records might refer to the same physical switch, or an IP address might have been reassigned.

Before importing the data, work through a few basic checks:

  1. Identify duplicates. Determine whether multiple records represent the same physical or logical object.
  2. Check identifiers. Verify that hostnames, serial numbers, and management addresses are associated with the correct devices.
  3. Normalize values. Use consistent naming, platform identifiers, and location references.
  4. Check coverage. Compare your dataset against the defined scope to find missing devices.
  5. Separate facts from assumptions. Be careful with fields whose values may reflect historical configurations or outdated documentation.

For the initial import, favor attributes you can verify with confidence. A serial number retrieved from a device is generally straightforward to verify. Whether a particular VLAN should be configured on an interface requires a different kind of judgment.

Discovery and synchronization tools can also reduce manual effort. Our brownfield network automation demonstration using Slurp’it and Infrahub shows an approach to bringing discovered infrastructure information into a structured data platform. If you already maintain network data in NetBox, we also have a quick demonstration of syncing NetBox data to Infrahub.

You don’t have to abandon your existing tools to get started. In many cases, it makes sense to keep using them for the jobs they already perform well while building a consistent data foundation for automation.

Reconcile observed state with intended state

The data you’ve collected describes what exists in your network today. But a source of truth for automation also needs to describe how you intend the network to operate. Those two states won’t always match, especially in a brownfield environment.

Imagine an access switch with an extra VLAN configured on one of its trunk ports.
Your discovery process can identify that VLAN, but it can’t determine whether the configuration is correct.

Maybe the VLAN supports a legitimate service that was never documented. Maybe it was added during an emergency change and should have been removed afterward. Or maybe it’s a mistake. Automatically importing that configuration as the intended state would effectively approve it without investigation.

This is why it’s important to distinguish between observed state and intended state.

Observed state tells you what’s actually deployed and running. Intended state describes the approved design and configuration that automation should work toward. Both are valuable, and the differences between them often reveal problems that need attention.

As you establish your source of truth, audit the data in two ways.

First, check the intended data against your own standards.

For example:

  • Do device names follow the naming convention?
  • Does every active device belong to a valid site?
  • Are required attributes populated?
  • Are relationships between devices and other infrastructure objects valid?
  • Do assigned roles and platforms comply with your design rules?

These checks help establish internal consistency before automation starts relying on the data.

Second, compare intended state against observed state.

This is how you detect configuration drift, undocumented changes, and differences between your design and production.

Not every discrepancy needs to be fixed immediately. Some represent legitimate exceptions, while others require scheduled remediation. The important thing is to make discrepancies visible, investigate them, and track them through resolution.

Over time, you want the source of truth to become authoritative for changes. New configurations are generated or validated against approved intent, while observations from production help verify that the network matches that intent.

This shift is fundamental to reliable automation.

It’s also why an asset inventory alone isn’t enough for infrastructure automation. Knowing which devices you own is useful, but automation also needs to understand how those devices are supposed to operate and relate to the rest of your infrastructure.

Make data maintenance part of normal operations

Populating your source of truth is a milestone. Keeping it accurate is an ongoing operational responsibility.

If engineers continue adding devices, changing configurations, and decommissioning equipment without updating the source of truth, you’ll eventually end up with another outdated inventory.

The best way to prevent that is to make data maintenance part of the workflows your team already uses.

Think about the lifecycle of a network device. When a new switch is provisioned, its record should be created or updated as part of that process. When it’s assigned to a site, the relationship should be recorded. When its role changes, the relevant attributes should be updated. And when the switch is decommissioned, its status should reflect that change.

Some updates can be automated. Others require human input or approval. The approach depends on which system or team is authoritative for the data.

A discovery tool, for example, may be a reliable source for the operating system version currently running on a device. But the approved target version might be defined by your engineering standards.

For each important attribute, establish where its value comes from, who’s responsible for it, and how it stays current. Clear ownership becomes increasingly important as more workflows depend on the data.

You should also determine how changes to the source of truth itself are reviewed. Changing the intended role of a switch from access to distribution could affect configuration generation, policy validation, and other automation. That kind of data change deserves appropriate controls.

In Infrahub, branching and Proposed Changes allow teams to work on infrastructure data changes, validate them, and review them before merging them into the main dataset. The same principles behind software change management can therefore apply to changes in your infrastructure’s intended state.

Finally, keep checking production against intent. Even with well-established processes, changes can happen outside your automation workflows. Regular comparisons help identify drift before inaccurate assumptions spread into other systems.

Build a source of truth that becomes more valuable over time

A brownfield network will always have some history attached to it. You don’t need to eliminate every inconsistency before you can start automating.

You need a clear starting scope, a reliable set of essential data, and a process for improving that data as your requirements evolve.

Begin by covering an entire, manageable part of the network. Model the important identities and relationships correctly. Use existing systems to discover what you have, then review that information before importing it.

From there, progressively reconcile production reality with approved intent and make data maintenance part of your normal operations.

As you expand into more environments and more advanced workflows, the source of truth becomes more useful because it captures a growing understanding of how your infrastructure is designed to work.

Start with a manageable project, demonstrate value through low-risk automation, and build on those early successes. Each new workflow gives you a reason to improve your data and extend your automation capabilities.

Over time, those incremental improvements create the foundation for more sophisticated automation you can trust.

divider

This post was inspired by a presentation on bootstrapping a source of truth in brownfield environments by network automation consultant Constantin Vurli.

Damien Garros

Damien Garros | Strategic innovator in infrastructure automation and data management with deep expertise in networking, observability, and open source development. Known for pioneering ideas and pushing industry boundaries through novel architectural approaches. Loves to challenge himself—and the status quo. Co-founder and CEO at OpsMill, makers of Infrahub.

REQUEST A DEMO

Infrahub by OpsMill logo

See what Infrahub can do for you

✔ Get a personal tour of Infrahub Enterprise

✔ Learn how we can support your infrastructure automation goals

✔ Ask questions and get advice from our automation experts

By submitting this form, I confirm that I have read and agree to OpsMill’s privacy policy.

Fantastic! 🙌

Check your email for a message from our team.

From there, you can pick a demo time that’s convenient for you and invite any colleagues who you want to attend.

We’re looking forward to hearing about your automation goals and exploring how Infrahub can help you meet them.

Get Ottermatic

Join hundreds of your industry peers exploring trends, tips, and talk in infrastructure automation.

Activate your subscription.

blue OpsMill otter