How Zipang screens Indonesian candidates (5-gate funnel + 30/90-day guarantee)

The 5-gate structure is designed to fail honestly at each gate, not to inflate placement numbers. A 2-gate funnel (CV + interview) misses too many of the failure modes that show up after deployment: ramp-and-reality mismatch, English proficiency drift on real tasks, and competence gaps that only show up in actual production work. An 8-gate funnel creates too much candidate drop-off from fatigue and discourages the strongest candidates from completing the process. The 5-gate structure balances signal quality (each gate adds real signal) against candidate experience (candidates don't drop out of fatigue).

The funnel is also calibrated to a specific operational outcome: 88%+ 12-month retention and 90%+ sustained accuracy at month 3. A funnel that lets through candidates with weaker ramp or weaker English would push those numbers down. The gates are designed to surface failure modes early — at gates 1–3, the cost of cutting is low; at gate 5, the cost is highest (the candidate has done real work).

Gate 1 is a CV review against role-specific criteria. The criteria vary by track: for customer support, the bar is a completed degree, demonstrable English (curriculum, work experience, or test), and prior customer-facing or admin work. For AI data annotation, the bar is more about cognitive tests, attention to detail, and language coverage than work history. For virtual assistance, prior admin / coordination experience matters more. The CV is scored against a 12-criterion rubric (education, English signal, role-relevant experience, tools, communication quality in CV itself, etc.).

Pass-through at Gate 1 is roughly 10% of applicants — the funnel is intentionally tight here to filter out applicants without a credible base for the role. The cost of cutting at Gate 1 is low (no interview time invested yet), so the bar is high. Strong applicants are progressed to Gate 2 within 24–48 hours; weak applicants receive a polite rejection email with general feedback. The honest read: if your CV does not pass Gate 1, the issue is usually a missing signal (no English evidence, no role-relevant experience, generic CV) rather than a hard cap on candidate quality.

Gate 2 is an English proficiency assessment aligned to the Common European Framework of Reference (CEFR). The assessment has two parts: a 30-minute writing task (typically a customer email response, a back-office summary, or a process description) and a 15-minute speaking task with a Bahasa Indonesia-fluent assessor. The writing task is scored on grammar, vocabulary, task completion, and tone. The speaking task is scored on fluency, comprehensibility, accent clarity, and listening comprehension.

The CEFR target varies by role: customer support typically requires B2 (upper-intermediate), back-office and data entry typically require B1+ (intermediate), and AI data annotation for English-language content typically requires B2+ with strong reading comprehension. Pass-through at Gate 2 is roughly 50–60% of Gate 1 passers — meaning the effective funnel rate after Gate 2 is ~5–6% of applicants. Candidates who fail narrowly are sometimes invited to re-test after a 2-week self-study period; candidates who fail clearly are cut. The honest read: most candidates who fail Gate 2 are B1 or below, and the production-quality work requires B2+.

Gate 3 is a role-specific quiz that simulates the actual work. The quiz is different for each track: customer support candidates respond to 8–10 simulated tickets (refund request, billing dispute, account verification, escalation scenario); data entry candidates complete a 60-minute structured data entry task with accuracy scoring; AI annotation candidates complete a 90-minute annotation task with multi-domain content (image, text, video); virtual assistant candidates complete an admin coordination simulation (calendar management, email triage, document prep).

The quiz is scored against production-quality benchmarks: for CS, the score is response quality (issue resolution, tone, compliance) plus time-to-respond. For data entry, the score is accuracy and throughput. For AI annotation, the score is inter-annotator agreement and edge-case handling. For VA, the score is task completion, calendar hygiene, and communication clarity. Pass-through at Gate 3 is roughly 40–50% of Gate 2 passers — meaning the effective funnel rate after Gate 3 is ~2–3% of applicants. The quiz is the strongest single signal of production readiness.

Gate 4 is a 45-minute structured video interview using the STAR method (Situation, Task, Action, Result). The interview is conducted by a Bahasa Indonesia-fluent Zipang assessor and covers four areas: motivation (why this role, why this company, why remote work), reliability (attendance, equipment, environment), competence (past experience using STAR for the role's core scenarios), and culture fit (collaboration, accountability, growth orientation). Each area is scored 1–5, and a candidate needs ≥ 3.5 average to pass.

Pass-through at Gate 4 is roughly 60–70% of Gate 3 passers — meaning the effective funnel rate after Gate 4 is ~1.5–2% of applicants. The interview is the only gate where human judgment dominates; the others are mostly scored against rubrics. The interview is also where the candidate's first impression of Zipang's employer brand matters most. Strong candidates who pass Gate 4 are progressed to Gate 5 (paid trial) within 2–3 days; the trial is scheduled, not optional.

Gate 5 is a paid trial task that simulates a real client engagement. The trial is 3–5 days of paid work, scored against a client-specific KPI. For CS programs, the KPI is a quality score (60% weight) and response time (40% weight). For data entry, it's accuracy (70% weight) and throughput (30%). For AI annotation, it's inter-annotator agreement (50%), accuracy against gold standard (30%), and edge-case handling (20%). For VA, it's task completion (60%), calendar hygiene (20%), and communication clarity (20%).

Pass-through at Gate 5 is roughly 50% of trial completers — meaning the effective end-to-end pass rate from Gate 1 to active deployment is 6–12% of applicants, depending on role and the specific KPI bar. The trial is paid at the full role rate (e.g., a customer support trial is paid the same Rp 8–12M / month pro-rated for the trial week) — Zipang does not run unpaid trials. The trial is the highest-fidelity signal of production readiness, and it is also where the candidate gets a real preview of the work.

The aggregate pass-rate from Gate 1 to active deployment is 6–12%, depending on role. Customer support is on the lower end (6–8%) because of the higher English bar; data entry is on the higher end (10–12%) because the bar is technical rather than linguistic. The most common reasons candidates are cut: at Gate 1, generic CV, no English signal, no role-relevant experience. At Gate 2, English below the role's CEFR bar (most common overall reason for cut). At Gate 3, inability to hit the production-quality benchmark on the role-specific quiz. At Gate 4, weak STAR responses, low motivation signal, or poor reliability signal. At Gate 5, failure to hit the client KPI bar in the trial.

Time-to-decision from Gate 1 to Gate 5 is 7–14 days, depending on candidate responsiveness and trial scheduling. Zipang's target is 10 days median, with the upper bound driven by Gate 5 trial scheduling. The funnel is fast on the front end (Gate 1 to Gate 4 in 3–5 days) and slower at the back end (Gate 5 trial 3–5 days plus scoring). For employers, this means a candidate submitted on Monday can be in active deployment by the following Monday–Friday, depending on the role and trial complexity.

What the client sees: a PII-redacted talent card for each shortlisted candidate, including role-relevant scores (English CEFR, quiz score, interview score, trial KPI), anonymized work samples, and the role for which the candidate is being proposed. The card is designed to let the client evaluate fit without seeing PII that would bias the review (name, photo, age, gender, university, exact years of experience). Zipang's PII redaction is consistent with UU PDP (Indonesia's GDPR-equivalent) and with most international buyer compliance frameworks.

What the client does NOT see: the candidate's name, photo, age, gender, exact address, university, current employer, family status, or any other PII. The client also does not see the candidate's full CV, full interview transcript, or the full trial task output. The client sees the talent card, the work sample, and the role fit. If the client wants to interview the candidate directly, Zipang facilitates a structured interview with the candidate's consent — but the client still does not see PII beyond what the candidate chooses to share.

The 30/90-day guarantee: if a deployed candidate fails the client's KPI bar or leaves voluntarily within 30 days, Zipang replaces the candidate free of charge. If the candidate leaves or is cut between 31 and 90 days, Zipang replaces at 50% of the placement cost. The guarantee sits on top of the 88%+ 12-month retention track record and is the operational commitment behind the 432 deployed professionals and 3.4M monthly production tasks. For employers, the guarantee is the practical answer to 'what if the candidate doesn't work out' — and the funnel is the operational answer to 'how do I know the next candidate will.'