Software discovery and model normalization in Software Asset Management Professional turns raw installation records from scans into standard publisher, product, version and edition values. Reconciliation uses those values to match installs against the licenses you hold. Records that no rule can match keep a lower normalization status until a user works them by hand.
How discovered software enters the normalization queue
The process starts with a scan. ServiceNow Discovery, or Microsoft SCCM where desktops are managed through it, records each software installation it finds. For the scan side in more detail, see ServiceNow Discovery.
Each installation lands in the Software Installations table. Discovery then matches the installation’s publisher, display name and version, which form its primary key, against the Software Discovery Model table, reusing a match or creating and linking a new model. Normalization runs after that step, so the quality of the scan data limits what normalization can achieve. For the wider concept behind the discovery model, see software model explained.
For a first-time SCCM integration, request the Service Graph Connector for Microsoft SCCM from the ServiceNow Store, because the older Microsoft SCCM integration plugins are being deprecated.
What normalization compares and what it writes
Normalization takes four discovered values, publisher, product, version and edition, and compares them with normalized equivalents in the Software Asset Management Content Library. A match writes a clean value into the matching column on the Software Discovery Model table. A discovered publisher string such as Microsoft Corp normalizes to Microsoft.
Those normalized columns feed three later steps: reconciling the entitlements you purchased, calculating license position, and managing software lifecycle. If one publisher appears under several names, reconciliation has a harder time matching those installs to the right entitlements. For the purchase side of that reconciliation, see software entitlement explained.
Four kinds of rule do the work, and they differ mainly in how many discovery models each one covers.
| Rule type | What it covers | Practical effect |
|---|---|---|
| Product Map and Package Map | One rule for each distinct discovery model | Precise, but numerous. A thousand distinct models can need a thousand rules |
| Pattern normalization | One rule for a shared data pattern | A name starting with a fixed product string maps to one normalized product |
| Enhanced pattern normalization | Variations in product, edition and version names | Conditional mapping, where the presence or absence of specific terms decides inclusion |
| Wide-net normalization | Models whose discovered values look alike | Applies to products that the ServiceNow Content Service marks as eligible |
One pattern can treat anything starting with Microsoft Office 2019 as Microsoft Office, and an enhanced pattern can treat standard, STD and stand as one edition.
Content is refreshed on a schedule. A daily Central Data Service job pulls the content tables. The daily normalization job applies content rules to discovery models, and the weekly job rechecks machine learning predictions against the latest rules. That same daily job also assigns each model its normalization status.
Reading the six normalization statuses
Each discovery model carries one of six statuses. The status shows how much of the record the platform could resolve and whether a person needs to step in.
| Status | What it means | What you can do |
|---|---|---|
| New | Not yet through normalization | Wait for the scheduled job |
| Normalized | Publisher, product and version matched; no fields are editable | Revert if the result is wrong |
| Partially Normalized | Publisher and product matched; version is missing | Add the version by hand |
| Publisher Normalized | The publisher alone matched | Add the product and version by hand |
| Match Not Found | None of the three key fields matched a rule | Normalize by hand, or revert |
| Manually Normalized | Values were entered by a person | Review suggestions; cannot be reverted |
Two details change how the table reads. A discovered publisher and product for a product type of Not Licensable, Child, Driver or Patch count as Normalized even without a version. Normalized records also receive a full version, while partially and publisher normalized records do not.
Match Not Found is the status to watch. It occurs when no normalization rule exists for the software, and custom software built for your own organization is a typical example.
Working the records that did not normalize
Records in Partially Normalized, Publisher Normalized or Match Not Found need a person to finish them. In the Software Asset Workspace, go to License operations, then Discovery models. From there:
- Open a discovery model with one of the three statuses.
- Open the ellipsis icon on the right side of the page.
- Select Normalize and enter the correct values.
- Confirm the status now reads Manually Normalized.
The sam_admin and sam_user roles can run this procedure. The classic interface offers the same task on the Software Discovery Models form.
Enter values carefully. A manually normalized record cannot be reverted with the revert action, so check them against the software’s actual publisher and product name before saving.
Reverting a normalization
Reverting is the way back for a record normalized incorrectly. Open the discovery model in the workspace and select Revert Normalization. The normalized fields reset to their original values, and rules tied to that model are deactivated. The record then shows Match Not Found, and the next run can try again.
Three statuses can be reverted: Normalized, Partially Normalized and Publisher Normalized. Manually Normalized records and records normalized by pattern rules cannot be reverted.
Reviewing normalization suggestions
During the weekly normalization process, the platform compares your manual values with Content Service values for publisher, product, version, edition, platform and language. A suggestion is created for each manually normalized model whose values are compared with Content Service values. If any field differs, the suggestion waits for review. If nothing differs and the auto-accept property is on (it is on by default), the suggestion is accepted automatically.
Reviewing a suggestion is a short decision:
- Accept copies the suggested values onto the discovery model, changes the status to Normalized, and sets the normalization date to the acceptance date.
- Reject keeps your manual values. The status stays Manually Normalized and the suggestion is recorded as Rejected.
The property com.snc.samp.auto_accept_normalization_suggestion is enabled by default. With it on, a suggestion with no differences from your manual values is accepted automatically. Domains in a domain-separated environment can override the property. A caution icon next to the Normalization status field marks a model that has a normalization suggestion record, and clicking it opens that record. The sam_admin role accepts or rejects suggestions, and sam_user can view them.
Machine learning as an opt-in second pass
Machine learning normalization supplements the content rules. Activate the Software Asset Management – Machine Learning Normalization plugin (com.sn_sam_ml_normalization) and confirm that the property Enable ML Normalization for discovered software (com.snc.samp.enable.ml_normalization) is selected. Disabling that property opts you out, and normalization then runs against the content rules alone.
The model predicts version, full version and edition. Partially normalized models are picked up by a separate job, SAM-Normalize discovery models using machine learning. Results appear in three columns on the Software Discovery Model table: ML prediction values, ML model version and ML normalization status, which can read ML normalized, Reverted or Content overridden.
Content values take precedence. The weekly content job retries machine learning records against the latest rules and replaces any prediction that differs. From the Brazil release, this capability is available to regulated markets on the ServiceNow Protected Platform in Singapore and Australia.
Tracking progress on the normalization and content dashboard
The Normalization and content dashboard sits under Software asset analytics in the Software Asset Workspace and is integrated with Performance Analytics. It shows the percentage of normalized installs, which you can filter by publisher or product. A donut chart breaks installs down by normalization status, and selecting a status lists its discovery models. A Central Data Service Download Status list shows, for each content table, the current count, expected count, last updated date and next scheduled pull.
The widgets fill after the daily normalization jobs have run successfully. The Central Data Service list is the quickest check on the content feed, and a current count well below the expected count suggests it may be incomplete.
Duplicate installs and reconciliation
A duplicate software installation appears when the same publisher, product, version and edition are discovered through more than one source. Duplicates matter because reconciliation counts installs, so a double-counted install can overstate usage.
By default, IBM software installations are prioritized over the same installation discovered from another source. New duplicate entries start with the Active flag set to false. The scheduled job SAM – Deduplicate Install Table then marks one record active and the other inactive, leaving one active record per install. For the license work these records feed, see software license optimization.
Frequently asked questions
Why does a record still show Match Not Found after the daily jobs have run? No rule matched any of its three key fields. Custom software is a common reason. The record stays in that status until a person normalizes it by hand, or until a content update adds a matching rule that a later daily run can apply.
Should I accept a suggestion or keep my own values? Compare the suggested values with the ones you entered. Accepting copies the content values onto the record and sets it to Normalized. Rejecting keeps your values, and the record remains Manually Normalized.
Can I undo a normalization? Yes, for records in the Normalized, Partially Normalized or Publisher Normalized statuses, through Revert Normalization. Manually Normalized records and records normalized by pattern rules cannot be reverted.
Does machine learning replace the content rules? No. Machine learning is opt-in and predicts version, full version and edition. Content values take precedence, so a prediction is overwritten when the content values differ.
For the wider Software Asset Management Professional picture, the hub page for Software Asset Management Professional covers the rest of the program.