ServiceNow has one of the most active and generous communities in the enterprise ecosystem. Every day, Experts, MVPs, and Rising Stars invest their time answering questions, sharing solutions, and helping others move forward (often for the nth time 😄).
While working closely with ServiceNow forum, I kept running into a recurring problem:
Many questions asked today have already been answered months ago or even years ago — but are hard to discover because they are worded differently.
I know what you might be thinking:
“Just copy the question and Google it, right?”
And yes — that does help… sometimes.
But keyword-based search often fails when:
-
The question is phrased differently
-
Terminology has evolved across ServiceNow releases (hello Madrid → Quebec → Vancouver 👋)
-
The context is the same, but the words are not
In short: no keyword match = no results, even when the answer is already sitting there, quietly judging us.
This recurring gap led me to build a small experiment.
The Idea — ServiceNow Semantic Search
I’ve built and deployed an MVP (Minimum Viable Product) called:
ServiceNow Semantic Search
The goal is simple:
Identify previously answered ServiceNow forum questions based on meaning, not keywords, and surface the most relevant historical solutions, give them to you as an URL pointing to that solution.
Instead of matching text, the system matches intent.
Current State of the MVP
This is intentionally an early MVP, not a full-scale platform.
-
🔹 Indexed ~5,000 ServiceNow forum threads
(Yes, I know — that’s tiny compared to the total number of questions. The idea was to first build something that actually works and then ask: “Is this useful at all?” If the answer is no, that’s where it ends. No hard feelings (well I hope😄)
What the solution does today:
-
🔹 Focuses on question-level semantic similarity
-
🔹 Read-only and research-oriented (no production instances were harmed)
-
🔹 Built to validate feasibility and accuracy first
Even at this scale, the MVP already shows promising results in detecting:
-
Reworded questions
-
Similar issues across different ServiceNow releases
-
Repeated patterns in community problems (yes, some questions really do come back every few months 😅)
Deployments URL: Semantic Search
How This Works
It’s intentionally simple:
-
Copy the newly created question from the ServiceNow Community.
-
Paste it into the deployed web app.
-
The system does its thing ✨ — it checks whether the same or a semantically similar question has been asked before.
-
If a match is found, it points you to the most relevant existing thread in the community.
-
If no close match exists, then congratulations — this one’s truly new, and it’s time for you to step in as the expert and answer it.
The idea is not to replace expertise, but to make sure we don’t reinvent the wheel when the wheel already exists.
Vision (If Fully Implemented)
If expanded beyond the MVP scope, this idea could evolve into:
-
Indexing all historical ServiceNow forum questions (legacy → latest)
-
Scheduled, incremental ingestion runs (because manually re-indexing everything is not fun)
-
Reducing duplicate questions by surfacing prior solutions early
-
Supporting Experts and MVPs by:
-
Saving time on repeated answers
-
Highlighting canonical solutions
-
Improving discoverability of older but still valid content
-
This is not meant to replace the community —
it’s meant to support the experts who already power it (and maybe give them back a few minutes of their day).
How This Works (End-to-End)
This system works in two clear phases: data preparation and question discovery.
Phase 1: Discover → Collect → Index (Behind the Scenes)
Before any searching happens, there’s some groundwork involved:
-
ServiceNow Community threads are first discovered
Relevant forum threads are identified across different topics and releases. -
Questions and answers are then scraped
Each thread is parsed to extract:-
Thread title
-
Original question
-
Accepted solution (if available)
-
Source URL
-
-
Each thread is stored as a clean JSON object
One thread = one JSON file.
( just structured, readable data.) -
Only after this step does indexing happen
-
The title + question text is converted into a semantic vector
-
Vectors are added to the Semantic Similarity Index
-
Indexing is incremental, so only new threads are processed
-
This entire phase runs offline and ahead of time, so the search experience stays fast and lightweight.
Why I’m Sharing This Now
I’m sharing this early because feedback from the ServiceNow community experts matters more.
I’d genuinely love your thoughts on:
-
Is this useful for Experts / MVPs who answer questions daily?
-
Where do you see risks, limitations, or improvements?
I’m open to all feedback — positive or negative.
As I mentioned earlier, this feedback will genuinely decide whether I continue evolving this solution or stop here.
(So yes, you have more power here than you think 😄)
This is a community-driven experiment, and your perspective is invaluable.
Deployments URL here: Semantic Search
Important Note
This tool:
-
Is for educational and research purposes
-
Does not claim ownership over community content
-
Always respects original sources and links back to them
(credit where credit is due)
ServiceNow’s greatest strength has always been its people.
This MVP is a small attempt to honor and amplify the knowledge that the community has already built over the years — and maybe reduce a few déjà-vu moments along the way.
I’m grateful for any feedback, suggestions, or critical views — they’ll help shape what this could become (or whether it should). Hope this wont be that blog which is read by one. (or will it? haha)
Thank you for reading, and thank you for everything you contribute to the ServiceNow ecosystem. 💙
— Azar
P.S: Please tag in the fellow experts #community mvps, and #rising stars, appreciate it, thanks again.
Deployments URL here: Semantic Search
5 Comments
