Tooling & Productivity
Promptly
20 January 2026
Resolves simple user queries, so ops teams can focus on what matters.
.png)
https://promptly.hack2026.com.sg/
How it Began
It started with a problem one of our team members faced in her day-to-day work on a public-facing product. She was receiving a large volume of tickets from users asking questions already covered in publicly available guides.
She was not alone. Talking to other ops officers revealed the same problem across multiple products.
This raised the question:
❓If the answers are already available, can we deliver it to users before they submit a ticket to ops?
Opportunity - The Problem
Operations teams spend too much time repeatedly handling support tickets whose answers already exist and could be delivered instantly.
Who experiences this problem?
Ops Teams
Spend time retrieving answers that are already documented
Respond to inquiries that users could have resolved on their own
Lose bandwidth for higher value work due to repetitive lookups
End-Users
Submit tickets for information already available in guides or FAQs
Experience long wait times for straightforward questions (~3-7 days)
Don't know where or how to look, or don't know the resource exists
Background
We interviewed 20 ops officers across 9 products during problem discovery, and a few themes stood out to us.
Users prefer to ask rather than search.
“Just never read lor, didn't bother opening the guide. We ping pong 3 times to open the link. User was lazy to open the link”
"I put the IP addresses on the guide, sent the link, and the user still asked me what the IP addresses are."
Guides are scattered and hard to find.
"Information is just everywhere. You have AskGov, you have the guide, then the chatbot... people don't know where to go for information, that's why they contact us."
Some users can’t articulate their issue well.
“Most users who write in are uncles and aunties who are quite reliant on someone else explaining."
Why does it matter?
Each ticket takes 3 to 5 minutes to resolve. At scale, this adds up:
Smaller products: a few hours/week
High volume products: 800+ hours/month, over 2,000 during post-holiday spikes
"I was like, I cannot do this on my own… I need to fight for this because I really need help."— Officer on getting extra headcount to help with overwhelming ticket load
Teams can't scale because support doesn't scale.
"We no longer reach out to people, cannot manage at this point."— Ops Officer
What if we solve this problem?
Automating 20 to 40% of eligible tickets could save hundreds of hours monthly for ops teams, while giving users instant answers to simple questions.
1. For smaller teams, this means a few hours back each week.
2. For larger teams, it's the equivalent of multiple full-time officers.
Scale | Hours saved/month |
|---|---|
High volume | 330 to 670 hours |
Medium volume | 26 to 53 hours |
Low volume | 1.5 to 3 hours |
Velocity - Solution Overview
Promptly is an embeddable tool that allows end users to type questions or attach a screenshot. Promptly then retrieves responses that have been configured by ops officers themselves and passes this back to users automatically.
.png)
.png)
What sets Promptly apart?
Similar solutions exist, but most prioritise end user experience. Promptly prioritises offloading ops workload, with improved user experience as a natural outcome.
Screenshot interpretation: Accepts screenshots from end users who struggle to describe their issue.
Verified content only: Ops officers curate and consolidate different sources into Promptly, generating responses with no external dependencies or irrelevant answers from the wider internet.
Fully escalatable: Does not block users from raising tickets. If no match is found or answers are unsatisfactory, users can submit a ticket for human intervention.
How it works:
.png)
Admin users to configure two types of references (Public/Internal) within Promptly.
Public guides | Internal references |
Links to publicly available product guides are uploaded. Promptly scrapes the content and stores it for retrieval. | Internal content not found in public guides, such as CRM macros or internal playbooks, are uploaded to Promptly’s database. |
Promptly then uses Retrieval Augmented Generation (RAG) with vector search and Claude Haiku 4.5 to process user inputs and generate outputs.
Step | Method | What it does |
|---|---|---|
Query interpretation | Haiku 4.5 | Understands what the user is asking (text and/or screenshot) |
Retrieval | Vector search | Converts query to embedding and finds matching content |
Generation | Haiku 4.5 | Generates a response using retrieved content, not general knowledge |
🙁 What if Promptly cannot find a matching response?
If Promptly cannot find a matching response, it directs users to submit a ticket, ensuring that queries requiring human intervention still reach ops teams.
Publishing Promptly
Once configuration is completed, admin users can publish Promptly either by embedding the widget directly into their contact pages or redirecting users to their Promptly site when on-site embedding is restricted.
User Testing and Feedback
We conducted user tests with 5 ops officers across different products with real user queries to validate the setup flow and response accuracy.
Our User Testing Approach
Ops Officers were asked to configure Promptly for their product
Test Promptly with real queries from their users and compared this with the actual response from the officer
We observed their setup process, asked about points of confusion, and gathered feedback on response quality
.png)
We iterated and tested concurrently, making improvements based on real-time feedback from each user test and these were the outcomes.
Iterations from feedback
Time Period | Tester | Feedback | What we changed |
|---|---|---|---|
Week 3 | Tester 1 | Confused by various terminologies | Major copy adjustments |
Week 3 | Tester 2 | Unclear distinction between sources | Added descriptions and examples |
Week 3 | Tester 3 | Wanted to preview before publishing | Added preview functionality |
Week 4 | Tester 4 | Setup flow was clear | Minor copy adjustments |
Week 4 | Tester 5 | Found setup intuitive and clean | Minor UI tweaks |
Feedback: Response accuracy was high, significant improvements after copy changes.
Officers tested Promptly with real queries their users submit. Most found the generated responses accurate and aligned with what they would tell users.
"10 out of 10... this is basically what I would tell them."
Feedback: A tool to educate users
Officers also saw Promptly as a tool to actively educate users on self-help options before they escalate the issue to ops teams.
“The prompt should be able to like (prompt), have you tried?"
Interest by OGP officers in deploying Promptly:
When asked: "how likely will you use Promptly in the future?"
Officer | Response |
|---|---|
Officer A | "I think if everything is accurate, then yeah." |
Officer B | "As long as there’s the word beta, I will do it." |
Officer C | 7 out of 10 likelihood (3 points off for setup required, but once that’s done, it should be fine |
Traction - What has been done & what's next
Promptly is live and we are rolling out to selected OGP products for a beta launch.
When does Promptly work best?
Through our testing and analysis, we identified that Promptly's effectiveness varies significantly based on the nature of support queries and team characteristics.
For high volume products, automating 20-40% of queries can translate to hundreds of hours saved monthly. See Research and Approach for detailed breakdown.
Potential | Team | Queries Types |
|---|---|---|
High | Up-to-date guides, 20+ similar questions per week | Repetitive questions with answers already documented (e.g. how to reset password, how to use a feature) |
Medium | Guides exist but may be scattered, mix of simple and complex questions | Some repetitive, some require additional context from user |
Limited | New product with guides still being written, knowledge not consolidated yet | Each question is different, answers change frequently |
Not recommended | - No capacity to maintain knowledge sources | Sensitive handling, backend action, or case-by-case decisions needed |
What's Next?
Key features identified through user testing are essential for a robust beta launch:
Analytics: Visibility into what questions users ask, drop-off rates, and whether users escalate after using Promptly.
Email capture: Ability to identify users who interact with Promptly and later submit tickets for better user journey tracking.
Seamless escalation: Handoff to ticket submission should carry context from the Promptly interaction to provide continuity.
Measuring Success:
Success will be measured through reduced support ticket volumes and positive user feedback. If the beta demonstrates clear value, we will consider expanding Promptly to all OGP products and eventually to external government agencies.
Meet the team:
.png)
You may follow our guide to set up Promptly for your product.
For more on Promptly, reach out to us here: promptly@hack2026.com.sg