Demand intake gets messy over time. Different teams submit requests that overlap, and nobody notices until two projects are funded for the same need. Now Assist for Strategic Portfolio Management (SPM) has a Similar Demands skill that helps here. From a demand record, it finds existing demands that look like it, so a demand manager can spot duplicates and related work during review.
This skill doesn't write anything. It uses AI Search to find existing records, so there's no generated text and no rating to check.
What you get
When a demand manager opens a demand and selects Identify similar demands:
- A banner lists up to three matching demands, each with its number, a link and its title. A count shows how many more matched.
- The banner asks you to review the Similar Demands related list and confirm the matches.
- The matches are saved on the demand in the
similar_demandsfield, so they're still there when you reopen it. - If nothing matches, the form says "No similar demands identified."
How it works
- Your current demand becomes the query. The skill joins the values of the indexed fields into one block of text.
- AI Search looks through the indexed demands. It uses the
SimilarDemandsemantic index and a match threshold. Your current demand is left out of the results. - The matches come back to the form. The ids are saved to the demand, and the banner shows the first three.
- Later form loads read the saved list. Each demand is shown only if you have access to it.
If the demand has no content in the indexed fields, the skill returns "Search context cannot be empty" instead of guessing.
What is compared
Three fields on the demand are compared, in this order:
|
Ordr |
Field |
Why it matters |
|---|---|---|
|
100 |
Short description |
The title of the request. It's the quickest signal of intent. |
|
200 |
Description |
The detail of what is being asked for. |
|
300 |
Business case |
The reason for the request. Two demands with different wording can still share a business case. |
The skill joins the non-empty values of these fields and uses that text as the search query. The same fields are what the index holds for every other demand. So a demand is matched on its title, description and business case, and not on category, cost, dates, people or any other field.
A few things follow from this:
- Empty fields count for nothing. A demand with only a short description is matched on that alone, so results are weaker.
- Attachments aren't indexed. The data source has attachment indexing switched off.
- How close is close enough. A candidate must score at least 0.85 against the query to count as a match.
- Chunking. The search breaks text into chunks of 750 words, so a long business case is still searched in full.
Where to configure it
Everything below is part of the AI Search setup, not the skill record.
|
What |
Where |
Default |
|---|---|---|
|
Fields compared |
Semantic index configuration |
|
|
Which table is searched |
AI Search data source |
|
|
Embedding model |
The index configuration and data source |
|
|
Match threshold |
|
0.85 |
|
Search profile |
|
Shipped |
To compare another field, such as a custom one, add a component field row to the SimilarDemand index configuration, with the source set to dmn_demand, the field name and an order. The skill reads the list at run time, so the query text picks up the new field. Re-index the demand table afterwards so the stored demands include it too.
The threshold is a constant in code. Changing it means changing the script include, which makes it a customization to maintain across upgrades. Lowering it returns more matches with weaker similarity, and raising it returns fewer, closer matches.
What an admin sets up
- Turn the skill on in Now Assist admin. This activates the AI Search data source and semantic index. Turning it off deactivates them.
- Make sure AI Search is enabled on the instance.
- Index the demand table. An install and upgrade fix script does this once, only if AI Search is on, the skill is on and the table hasn't been indexed already.
The button only appears on demand records, and only when the skill is on.
FAQ
Does it use a large language model? No. It's a search over indexed demand records. No text is generated.
Which fields decide whether two demands are similar? Short description, description and business case. Nothing else is compared.
Can I add or remove fields? Yes, in the SimilarDemand semantic index configuration. Add or remove its component field rows, then re-index the demand table.
Can I change how strict the match is? The threshold is 0.85 and set in code (SPMNowAssistConstants), not in a settings page. Changing it needs a script change.
Why does the button show nothing, or not appear at all? Check that the skill is active, you have access to it and AI Search is enabled. Also check that the demand table has been indexed.
Why did I get "Search context cannot be empty"? The demand has no values in the compared fields. Fill in at least the short description or description and try again.
Why did it find nothing when I know a similar demand exists? The other demand may not be indexed yet, or it may score below the threshold. It may also be worded very differently in its short description, description and business case. If the table was only recently indexed, the index may still be processing.
Why do I only see three matches? The banner shows three and counts the rest. The related list holds the full set.
Are the matches always right? No. They're suggestions. The banner asks you to review the Similar Demands list and confirm them.
Can I see demands I don't have access to? No. The saved ids are loaded through a secured query, so only demands you can read are shown.
Does running it again overwrite the earlier list? Yes. Each run writes a new set of ids to similar_demands.
Does it change anything else on the demand? It doesn't change the demand's other fields. It writes the ids to similar_demands and shows a message.
What happens if I turn the skill off? The button disappears, and the AI Search data source and semantic index are deactivated. Turning it on again reactivates them.
