Trust & privacy

What data it uses

Two kinds of information. HR records companies already keep (staff list with join and leave dates, attendance, leave and medical certificates, overtime), which employees are told about. And staff voice: voluntary check-ins for new hires, a short monthly team pulse and anonymous exit chats.

Built for Malaysian workflows

English, Malay and Simplified Chinese; Malaysian dates and time zone; RM action costs; state holiday calendars; multilingual import headers; KRA, MC and overtime terminology; manual WhatsApp sharing; configurable probation review reminders. Imports currently cover Malaysian citizens only. Legal benchmarks are prototype guidance, not legal certification: Peninsular Malaysia/Labuan and Sabah/Sarawak have different legal coverage. Check contractual terms and local requirements. Malay and Chinese still need native-speaker review.

What it never does

• No risk scores, rankings or "at-risk" lists for individual employees.
• No per-person workforce analytics or risk reports. Consented private support reviews are a separate, identifiable process.
• No secret monitoring, cameras or emotion detection.
• It never guesses how engaged a person is from their records.

How it protects people

• Workforce imports remove names in the browser. Roster identities, opt-in follow-up contacts and private review contacts and answers are stored separately and encrypted.
• Employee IDs are replaced with keyed codes that only work inside your company.
• Anonymous rates, scores and feedback are shown only for groups of 5 or more people. Separate private reviews require explicit consent.
• Feedback is shown as paraphrased themes; exact quotes only in groups of 10 or more.
• Anonymous feedback is not used to contact someone unless they ask. Private review invitations require prior agreement; participation is voluntary.

Staff check-ins

• Anonymous: no login and no name. An answer keeps only the team, the month and the week, never the exact time, the phone or the internet address.
• A random browser cookie remembers an answer for that month. Clearing cookies or using another browser can bypass this check; it does not prove one response per person.
• Results show only for groups of 5 or more. If a small team's answers could be worked out by subtracting from a total, that total is hidden too.
• AI groups comments into themes and writes neutral paraphrases; the boss never sees anyone's own words. Comments that try to give the AI instructions are treated as plain text.
• Comments about serious issues, such as harassment or a safety danger, are shown to HR only as a count.
• Someone who asks to be contacted gives a name and phone number in a separate form. It is encrypted, never linked to their answers, and deleted 30 days after HR closes it (90 days at most).
• Roster names and phone numbers support check-in invitations and probation review reminders. They are encrypted, never linked to anonymous answers, and deleted 30 days after the later of Day 90 or the configured probation review date. Remove an entry sooner when it is no longer needed. Comment text is deleted after 12 months.

Lead & new-hire reviews

A consenting lead and new hire each answer the same three private questions now and again one month later, so HR can see whether the new hire feels supported. Separate from the anonymous team results; never a performance grade. Private reviews and contacts expire 180 days after creation. Links expire after 30 days or earlier at review expiry. You may withdraw using your link, or ask HR to delete the review sooner. Each row is one person's self-reported rating, 0–10. Change = their after rating minus their before rating in points. Lead and new-hire views stay separate. This custom questionnaire is not a validated performance scale; changes do not prove what caused an improvement or predict probation confirmation. Skipped or pending answers are never converted to zero. Optional attachments support HR review and are not verified by AI.

Private to HR/admins and the task owner · Kept up to 180 days.

Causes, actions and checks

• SignalPath shows possible causes, never verdicts. A statistics engine chooses them from the records and staff feedback; AI only explains them in plain words, writes no numbers itself, and its text is checked before it is shown.
• The boss or HR decides what to do. Every decision is recorded with who made it, when and why, and owners and admins can read who did what in the audit log.
• After an action starts, SignalPath checks the numbers after 1 month and again at 3 and 6 months, against the team's own past and the other teams. With too few people it says "too early" rather than guess.
• The staff board shows what people raised (as themes) and what the company did, in Malay, English or Chinese. It never shows a name, an answer or a number about a small group.
• The sample company's results after an action are a simulated future, labelled on every screen. Real companies wait a month for the first check.

How AI is used

AI helps with language: reading messy file headers, grouping comments into themes in several languages, and explaining results in plain words. It runs on Google's Gemini API on a paid plan, under which Google does not use the data to improve its products, and names and IC numbers are removed before anything is sent. Every number comes from the statistics engine, never from the AI, and AI text is checked before it is shown.

Responsible AI check

Five questions we ask of every feature.

Could it be biased?

It compares situations such as teams and shifts. Before any comparison involving age, gender, race, religion or nationality, it checks whether another factor explains the difference, and it never recommends anything based on these traits.

Does it protect privacy?

Workforce names removed, keyed IDs, group results and a minimum reporting size of 5. Small-group context can still help someone guess who responded.

Could it create unfair labels?

It labels situations, never people.

Is there human oversight?

SignalPath suggests likely causes and options. The boss or HR decides what to do, and decisions are recorded.

Is AI genuinely useful here, and what new risk does it create?

AI can reduce manual reading and help across languages; real HR time savings have not been measured yet. Over-trust and misuse remain risks, so evidence, limits and human decisions stay visible.

About the sample company

Contoh Digital is a synthetic company. Every employee, number and comment is generated and is not based on any real company or person.

Malaysian context

Designed around the principles of Malaysia's Personal Data Protection Act 2010: tell people what is collected and why, collect only what is needed, and keep it secure.