Latest blog and updates | Apex Configuration Group

ServiceNow Multi-Source Data: Precedence Starts Before Reconciliation

Written by Mariusz Staniszewski | Oct 15, 2026, 8:00:02 AM

Multi-source data precedence in ServiceNow has two halves, and most estates only design one. The half everyone knows is reconciliation, which decides whose value wins on each attribute of a configuration item (CI). The half that gets missed is identification. Before any source can win an attribute, every source has to agree which CI it is describing. If a source can't supply an identifier the others share, precedence never gets a chance to work. That source quietly builds its own copy of your estate, and nothing in the platform flags it.

Key Takeaways

  • Multi-source precedence has two halves. Identification decides which configuration item each source is describing, and reconciliation decides whose value wins on each attribute. Reconciliation never gets a chance to work if identification fails.
  • A source that identifies badly is never outranked. It creates new records, and each new record belongs entirely to that source.
  • Build a matrix of sources against classes recording which identifying attributes each source actually sends in real payloads. Any cell with no strong shared identifier is a duplicate factory.
  • Unrecognised duplicates never reach the de-duplication queue. Compare CI counts per class and per source with each source's own inventory every month and after every integration change.
  • Design attribute precedence from a second matrix of attributes against sources, naming one authoritative source and where it makes sense a fallback. An attribute that no reconciliation definition covers can be changed by any source in any order.
  • Register every source under its own name, give each source an owner and gate new sources through change with the identifier matrix, the attribute matrix and a replayed test payload.

Precedence starts with a shared identifier

The Identification and Reconciliation Engine processes every payload in the same order. It identifies the CI first. Only then does it apply reconciliation rules to decide which attributes the source may update.

So a source that identifies badly is never outranked. It simply creates new records, and each new record belongs entirely to that source.

The first design artefact for a multi-source configuration management database (CMDB) is a matrix of sources against classes. For each cell, record which identifying attributes that source actually sends. Serial number, hypervisor or cloud instance identifier, fully qualified name, hardware address: tick the ones that arrive populated in real payloads, rather than the ones the integration documentation promises.

Any cell where a source shares no strong identifier with your primary source is a duplicate factory. Deal with it before you write a single reconciliation rule.

When two sources can't agree which CI they're describing

On one estate of tens of millions of CIs, a cloud security platform was treated as the definitive source for cloud infrastructure. The plan was to let it create cloud virtual server CIs directly, alongside horizontal discovery.

The platform exposed no identifier for those servers that correlated with anything already held in the CMDB. Every server it reported failed to match, so the engine created a new CI. Discovery kept maintaining its own record of the same server.

These duplicates never reached the de-duplication queue. The engine had no shared value to compare, so as far as it could tell, they were different servers.

A cloud administrator spotted it. The CMDB held twice as many virtual servers as his own console showed. By then, change requests had been raised against the wrong CIs, the impact assessment ran against records with no relationships, and changes failed.

The fix was scoped to the class that couldn't identify. We switched off virtual server import from the security platform and replaced it with the Service Graph Connector for Amazon Web Services, which supplies the cloud provider's own instance identifiers. Every other class the security platform supplied carried on importing.

Count comparison catches what the de-duplication queue can't

The story above has a lesson for monitoring. A de-duplication task only exists when the engine recognises a duplicate. Unrecognised duplicates need a different check.

Compare CI counts per class, per source, against each source's own inventory. Run it monthly, and after every integration change.

  • A CMDB count well above the source's count points to duplicates the engine couldn't see.
  • A count well below it points to payloads rejected or skipped at identification.
  • A class where one source's CIs have no relationships at all is usually a parallel copy of the estate.

The comparison takes an afternoon to build. It's also the only control that would have caught the problem above before the changes failed.

Designing attribute precedence source by source

Once identification holds, reconciliation decides who owns what. Reconciliation definitions are set per class, name one or more sources and the attributes they may update, and carry a priority where the lowest number wins. Child classes inherit them.

Design these from a second matrix, attributes against sources. For each attribute that matters to change, incident or security work, name one authoritative source and, where it makes sense, a fallback.

Attribute typeUsually authoritativeWhy
Hardware and operating system detailHorizontal discovery or the cloud provider connectorRead directly from the device or the provider
Installed security agent statusThe endpoint security platformOnly it knows whether its agent is healthy
Support group and ownershipManual entry under change controlNo tool can discover accountability
Lifecycle and financial fieldsThe asset recordSet by procurement and disposal processes

Treat the table as a starting pattern, then test it against your own sources.

Then look for gaps. An attribute that no reconciliation definition covers can be updated by any source, in whatever order payloads arrive. On a busy estate, that attribute changes value several times a day and nobody can say why. The reconciliation rules article covers how definitions and priorities behave in more detail.

Stale sources and tied sources

Strict precedence has a failure mode. If the authoritative source stops reporting, its last value is locked in, and lower-priority sources that can see the change are blocked.

Data refresh rules handle this. They let a lower-priority source update an attribute once the higher-priority source's data is older than a period you set. Choose that period from how often the source actually reports, plus a margin.

Where two sources are equally trustworthy for an attribute, dynamic reconciliation lets you choose how the tie resolves, such as taking the most recently reported value. Use it deliberately and sparingly, because every tie you allow is an attribute nobody can explain from the rules alone.

Governing sources once they're live

Precedence decays when new sources arrive without a gate. Three controls keep it in place.

Register every source. Each integration writes under its own registered discovery source name. A source writing under a generic or borrowed name can't be given its own precedence, and it can't be traced when something goes wrong.

Give each source an owner. Write the integration owner into the RACI (responsible, accountable, consulted, informed) for the CMDB. That person answers for the source's count comparison and signs off changes to its transform.

Gate onboarding through change. A new source, or a new class from an existing source, needs three things before it writes to production: its row in the identifier matrix, its rows in the attribute matrix, and a replayed test payload showing it matches existing CIs. Track count variance per source as a key performance indicator (KPI) from day one.

Bad multi-source data doesn't stay in the CMDB. It surfaces in patch compliance figures and in integrations that never deliver the value promised.

How Apex helps

Our discovery and integration coverage review establishes what is missing from your estate view, where, and what that costs during an incident. It covers horizontal discovery and every third-party tool writing to your CMDB, including Service Graph Connectors. You get the identifier and attribute matrices for your actual sources, the count comparison, and a prioritised list of integrations to fix or switch off. We scope it with you at an initial consultation, based on your estate and requirements.

If the problem is already showing up as duplicates and unreliable attribute values, our data quality assessment measures where de-duplication, normalisation and reconciliation are failing, and why.

Book a CMDB diagnostic call or arrange a meeting with a consultant.

Frequently asked questions

Can a source be trusted for some classes and not others?

Yes, and it's common. Reconciliation definitions and integration mappings are set per class. A security platform can be authoritative for agent status on servers while another source creates the server CIs themselves.

Why don't unrecognised duplicates create de-duplication tasks?

A task is raised only when one payload matches more than one existing CI. If a source shares no identifying value with the existing record, there's no match at all. The engine creates a new CI and considers the job done.

What happens to attributes with no reconciliation definition?

Any source may update them. The value ends up being whatever the last payload sent. Find these gaps before they appear in a change or incident record.

How often should source counts be compared?

Monthly as a baseline, and immediately after any change to an integration, a transform or a source's own configuration. Most of the variance we find appears straight after a change.

Written by Mariusz Staniszewski, Apex Configuration Group.

Book a free call to discuss how Apex can help you on your journey to a better CMDB