Predictive intelligence scoring for similar incidents is not working as expected.Issue <!-- /*NS Branding Styles*/ --> .ns-kb-css-body-editor-container { p { font-size: 12pt; font-family: Lato; color: var(--now-color--text-primary, #000000); } span { font-size: 12pt; font-family: Lato; color: var(--now-color--text-primary, #000000); } h2 { font-size: 24pt; font-family: Lato; color: var(--now-color--text-primary, black); } h3 { font-size: 18pt; font-family: Lato; color: var(--now-color--text-primary, black); } h4 { font-size: 14pt; font-family: Lato; color: var(--now-color--text-primary, black); } a { font-size: 12pt; font-family: Lato; color: var(--now-color--link-primary, #00718F); } a:hover { font-size: 12pt; color: var(--now-color--link-primary, #024F69); } a:target { font-size: 12pt; color: var(--now-color--link-primary, #032D42); } a:visited { font-size: 12pt; color: var(--now-color--link-primary, #00718f); } ul { font-size: 12pt; font-family: Lato; } li { font-size: 12pt; font-family: Lato; } img { display: ; max-width: ; width: ; height: ; } } Predictive intelligence scoring for similar incidents is not working as expected. The similarity solutions are flagging the incidents as similar only when the details (short description and description) of the incident are exactly same or extremely close. When entering similar details which are not exactly similar or identical details, the scoring mechanism is not identifying the incidents as similar.In certain cases even when there are incidents in the training sample set, using the same details as those incidents does not flag the incident as similar. Release<!-- /*NS Branding Styles*/ --> .ns-kb-css-body-editor-container { p { font-size: 12pt; font-family: Lato; color: var(--now-color--text-primary, #000000); } span { font-size: 12pt; font-family: Lato; color: var(--now-color--text-primary, #000000); } h2 { font-size: 24pt; font-family: Lato; color: var(--now-color--text-primary, black); } h3 { font-size: 18pt; font-family: Lato; color: var(--now-color--text-primary, black); } h4 { font-size: 14pt; font-family: Lato; color: var(--now-color--text-primary, black); } a { font-size: 12pt; font-family: Lato; color: var(--now-color--link-primary, #00718F); } a:hover { font-size: 12pt; color: var(--now-color--link-primary, #024F69); } a:target { font-size: 12pt; color: var(--now-color--link-primary, #032D42); } a:visited { font-size: 12pt; color: var(--now-color--link-primary, #00718f); } ul { font-size: 12pt; font-family: Lato; } li { font-size: 12pt; font-family: Lato; } img { display: ; max-width: ; width: ; height: ; } } All Resolution<!-- /*NS Branding Styles*/ --> .ns-kb-css-body-editor-container { p { font-size: 12pt; font-family: Lato; color: var(--now-color--text-primary, #000000); } span { font-size: 12pt; font-family: Lato; color: var(--now-color--text-primary, #000000); } h2 { font-size: 24pt; font-family: Lato; color: var(--now-color--text-primary, black); } h3 { font-size: 18pt; font-family: Lato; color: var(--now-color--text-primary, black); } h4 { font-size: 14pt; font-family: Lato; color: var(--now-color--text-primary, black); } a { font-size: 12pt; font-family: Lato; color: var(--now-color--link-primary, #00718F); } a:hover { font-size: 12pt; color: var(--now-color--link-primary, #024F69); } a:target { font-size: 12pt; color: var(--now-color--link-primary, #032D42); } a:visited { font-size: 12pt; color: var(--now-color--link-primary, #00718f); } ul { font-size: 12pt; font-family: Lato; } li { font-size: 12pt; font-family: Lato; } img { display: ; max-width: ; width: ; height: ; } } The behaviour observed is expectedThere are several compounding factors at play.1. Factor 1 — Insufficient and Unrepresentative Training Data on Sub-Prod (Primary Cause)The most important variable is that this is a Sub-Prod environment. UAT instances typically contain a fraction of the data volume present in production. The similarity model's perceived "strictness" is directly tied to the size and diversity ofthe training dataset.When training data is sparse, the percentile score distribution compresses. Pairs that would score 80+ in a production environment with thousands of diverse incidents may only score 55-65 in a Sub-prod environment with a few hundred records —because the percentile is calculated relative to the entire training distribution, and a small dataset does not provide enough contrast between "somewhat similar" and "highly similar" pairs.Short phrases with 3-4 meaningful tokens (e.g., "Myapp not responding") are particularly sensitive to this: after stop-word removal, the pre-trained model has minimal text to compute semantic distance from, and the resulting scoresare compressed toward the lower range of the distribution.you should check the record count displayed on the Similarity Definition form next to the Table/Filter fields. If fewer than a few hundred records are in the training pool, this is a strong confirmation.-> For the solution number of matching records is too lowRecommendation :-The preferred number of records for training a solution is between 30,000 records and 300,000. If you submit more than 300,000 records, the most recent 300,000 records are used to train the solution. Use only authentic records from the database.refer below document for more detailshttps://www.servicenow.com/docs/r/it-operations-management/event-management/word-collection-similarity-solution.html?contentId=l3xqXBXafUdz54zrBX__Jw2. Factor 2 — Threshold of 75 is Too High for This Training DatasetGiven a compressed score distribution from sparse data, the threshold of 75 effectively cuts off everything that is not a near-literal match. The platform documentation explicitly states: "The lower the threshold value, the lower theprecision and the higher the coverage."Lowering the threshold to 50 or 60 is the fastest and most reversible diagnostic step. If semantically similar incidents suddenly appear in results after lowering it, the root cause is confirmed as the training data density issue combined with the threshold being too restrictive.We have lowered the threshold to 50 and getting the results as expected.Docs to refer :-https://www.servicenow.com/docs/r/zurich/intelligent-experiences/predictive-intelligence/create-similarity-solution.htmlhttps://www.servicenow.com/docs/r/zurich/intelligent-experiences/predictive-intelligence/update-similarity-threshold.htmlhttps://www.servicenow.com/docs/r/zurich/intelligent-experiences/predictive-intelligence/review-similarity-examples.html