AI-Suggested Similar Cases

What it does

When an agent opens or creates a case, D365 Customer Service uses AI (text similarity models trained on historical case data) to surface other resolved cases that are similar to the current one. The suggestions appear in a panel in the case form, showing the most similar cases with their resolution details. The intent is to help agents find existing solutions without doing a manual knowledge base search.

Key facts

Similar case suggestions appear in the Productivity pane on the right side of the case form (same pane as Copilot and knowledge article suggestions)
Suggestions are generated by comparing the current case's description and subject against the text content of resolved cases in the environment
The AI model is trained on the organisation's own historical case data: it improves over time as more resolved cases accumulate
A minimum volume of resolved cases is required before suggestions become meaningful (Microsoft recommends at least several hundred; quality improves significantly with thousands)
Agents can click a suggested similar case to open it and review the resolution steps, then apply the same resolution to the current case
Administrators can configure which case fields are used as input for similarity matching (title, description, subject tree, custom fields)
The feature requires the Customer Service Intelligence add-on or is included with Customer Service Enterprise: verify with licence confirmation before scoping
Similar case suggestions are different from Copilot knowledge article suggestions (which search the knowledge base, not historical cases)

When to use / skip

Use for organisations with a substantial history of resolved cases where resolution patterns repeat, IT service desks, utilities faults, financial services complaint handling. The more resolved cases exist with quality descriptions and documented resolutions, the more useful the suggestions.

Skip or defer for new deployments with no historical case data, there's nothing to match against. Also less useful in highly bespoke or one-off case environments where each case is genuinely unique. Most useful for tier-1 support where the same types of issues recur.

Configuration decisions

Enable similar case suggestions in Copilot Service Admin Center under the AI settings section: it's not on by default
Configure the field weights for similarity matching: which fields does the AI use to find similar cases? Title and description are the most valuable; subject tree classification adds precision
Decide on the minimum similarity threshold: lower thresholds show more suggestions but with lower confidence; higher thresholds show fewer but more accurate matches
Populate historical case resolutions before relying on the feature: cases with empty or thin resolution fields produce poor suggestions; run a data quality check on the resolution field in historical cases before enabling

Gotchas

The AI model uses case data in the current environment. If you're in a new environment migrated from an old system, migrated cases need to include resolution text for the model to learn from. Case records migrated without resolution detail produce no useful training signal.
Similar case suggestions are not Copilot: they're a separate AI feature. Agents who understand Copilot may assume this feature works the same way; it doesn't. It's a text similarity match against cases, not a generative AI response.
New case data takes time to incorporate into the model. Recently resolved cases may not immediately appear as suggestions. There's a model refresh cycle: new resolutions don't instantly become available as suggestions.
The feature surfaces similar cases but does not apply the resolution: agents must review the suggested resolution and manually apply it. This is intentional (agents need to validate the match), but train agents not to blindly copy resolution text from a similar case without checking it's actually applicable.
GDPR consideration: similar cases can include personally identifiable information from the resolution text (customer names, account numbers mentioned in resolutions). Make sure resolution text follows a PII-scrubbed or anonymised format, especially if multiple agents with different customer access levels can see the suggestions.

Consultant notes

Data quality is the entire story with similar case suggestions. The feature is only as good as the resolution text in historical cases. Before enabling, pull a sample of 50 resolved cases and look at the resolution field: if the most common entry is "Resolved" or "Called customer" with no further detail, the AI has nothing to learn from. Include a case resolution data quality standard in the operational runbook before go-live if this feature is in scope.
Set realistic expectations with clients about the new environment timeline. A deployment that goes live with zero historical case data will produce no useful similar case suggestions for the first several months. Frame it as a feature that gets better over time, not one that works on day one.
The PII concern in resolution text is worth raising with the client's data protection officer if the feature is used. If agents write case resolutions that include customer-specific details ("Account 1234567 was credited £50 per their complaint"), those details will appear in similar case suggestions shown to agents handling different customers. Resolution text standards should either be anonymised or agents should be trained to keep resolution text generic.

Source last updated: 2026-04-30 | Revisit if Microsoft updates the AI model training frequency or adds generative AI-powered resolution suggestions