For reviewers
Programme showcase
A short walkthrough for programme reviewers: what problem this addresses, how the experience works, how it stays a complement to official services, and what a responsible pilot would require.
1. The problem
- Older adults in Singapore receive scam attempts through channels they cannot easily verify, often while alone and under time pressure.
- Official tools exist, but the moment of risk is emotional: the person needs a calm prompt to pause and a simple route to a trusted human.
- Digital confidence is uneven. Small barriers — tiny text, unfamiliar words, fear of doing it wrong — stop people from checking at all.
2. The experience
- One promise, three words: Pause. Check. Stay steady.
- Four huge actions on the home screen, 20px base text, 48px+ targets, captions, high contrast and reduced motion.
- A practice centre with fictional delivery, government impersonation, investment and job examples, so people rehearse before it matters.
- A voice-style demo and a scripted coach show how the same coaching would feel conversationally, without shipping a live model.
3. Positioning: a complement, not a replacement
- CheckFirst Voice is independent. It is not affiliated with, endorsed by or approved by ScamShield, any agency, bank or regulator.
- It never says a message is safe. Every path ends at an official channel, the 1799 helpline, or a trusted person.
- It adds the coaching and routing layer around official services: rehearse, slow down, then use the official route.
- No automated reporting. The person acts, in their own name, through the official service.
4. Safeguards built into the design
- No emotion, cognition, health or capacity inference. No profiling and no social scoring.
- Sensitive-input blocking: anything resembling an OTP, NRIC, password, card or bank login is refused, not processed.
- Confirmation before every help, reminder or trusted-contact action; all external actions are simulated in this prototype.
- Data minimisation by default: session-only demo storage, a visible privacy receipt, and one-click deletion.
- Permanent emergency notice, plus explicit medical, legal and financial boundaries.
5. Pilot metrics we would agree first
- Practice completion rate and repeat practice, by accessibility setting used.
- Help requests raised and time to human follow-up during service hours.
- False escalation review: every escalation reviewed by a person, with the cause recorded.
- Self-reported confidence before and after, collected offline by programme staff.
- All metrics anonymised and aggregated — never individual behavioural profiles.
6. Next-step co-design
- Co-design sessions with older adults and caregivers, including mother-tongue speakers, before any build.
- Content review with community partners and an independent safety reviewer for every scripted line.
- DPIA, data residency and vendor review before any real data is handled.
- A small, supervised pilot with paper fallbacks and a named human escalation route.
Status of this prototype
Everything here is a local demo for design and policy discussion. There is no database, account, analytics, external API or live AI, and no claim of PDPA compliance or agency approval — only PDPA-ready design considerations.
Continue to the Safety & Privacy centre or the pilot dashboard.