Philippines staffing guide
Run a bookkeeping document-intake pilot with a virtual assistant
Evaluate document organization while payments and financial judgment remain protected.
Decide what the exercise must reveal
Run a bookkeeping document-intake pilot with a virtual assistant should help owners with recurring receipts, statements, invoices, and missing-document follow-up make a hiring decision from observable work. Define the position as bookkeeping support assistant, then select a task resembling the first safe assignment. The exercise is not a contest to see who tolerates unpaid project work or guesses hidden preferences. Keep it short, use fictional or properly redacted information, state the time limit, and explain how work will be assessed. A useful exercise reveals how someone reads instructions, handles uncertainty, communicates a stop condition, and returns work for review. It must not require access to a live customer, employee, tenant, patient, candidate, financial record, or business account.
Use a realistic but controlled scenario
Build the exercise around this situation: two possible duplicates, one missing receipt, and a vendor-name mismatch. Give every candidate the same facts and explain which source controls if notes conflict. Include one ordinary item, one incomplete item, and one case that belongs with an owner. That combination is more informative than a large volume of easy work. It shows whether the candidate can preserve the record while continuing safe parts. Make the scenario specific enough to review consistently, but never disguise real private information as a test. Replace names, identifiers, account details, and confidential facts with invented values, and label them clearly so nobody mistakes the exercise for live work.
Prepare one fair input packet
The packet should contain redacted documents, a naming convention, a read-only register, and an exception template. Add a one-page instruction sheet naming the output, due time and time zone, permitted references, communication channel, and clarification contact. If writing quality matters, say whether tone, structure, accuracy, and brevity are scored. If tool skill matters, offer a test environment or portable format rather than production. Candidates should not purchase software, disclose credentials, or use personal accounts. Record which materials were supplied so reviewers do not judge people against different information. Remove accidental clues that reveal a preferred answer while retaining the rules needed to complete the assignment fairly.
Score evidence rather than confidence
Use four or five behaviors tied to evidence in the return: instruction accuracy, factual traceability, exception recognition, clarity of handoff, and respect for authority. Describe acceptable work before reviewing submissions. “Flags the conflicting record and names its source” is reviewable; “shows good judgment” is not. Score the artifact once for completion and once for risk. A polished response that crosses a boundary should not outrank a plain response that identifies the limit and asks a useful question. Keep interview impressions separate from work-sample scores. Compare candidates on the same dimensions, and preserve brief reviewer notes that point to the work rather than personality labels.
Make the stop rules explicit
The candidate should stop or escalate before initiating payments, deciding tax treatment, changing final books, or guessing at matches. Put those limits in both the exercise and eventual SOP. A stop is not failure when the next step exceeds the assignment. The useful response identifies the item, states what was checked, explains what is missing or protected, and routes it to a named owner. Distinguish a cautious question from avoidance: the candidate can often complete safe fields, gather permitted facts, or prepare a draft while a decision waits. Reward that separation because it prevents silent guessing without turning every routine detail into an escalation. Never imply that owner silence expands the assistant’s authority.
Protect candidate and business information
Collect only information needed for evaluation, limit who can view it, and define when test files will be removed. Do not put credentials or sensitive records in the packet. The Federal Trade Commission advises businesses to understand what personal information they hold, keep only what they need, protect it, and plan for incidents. NIST’s Cybersecurity Framework 2.0 treats governance, access control, awareness, protection, and response as connected risk-management activities. These are general references, not legal advice or a complete control set. The business owner remains responsible for employment, privacy, security, professional, and jurisdiction-specific requirements.
Run a structured review conversation
Ask the candidate to explain one ordinary choice and one exception. Useful questions are: Which source controlled? What did you leave unchanged? What would you ask before continuing? How would the answer change if the deadline moved? Avoid revealing the preferred answer before the explanation. The goal is to see whether the return reflects a repeatable method. Let the candidate identify unclear instructions; that can expose a defect in the exercise. Record evidence from the artifact and explanation, but do not invite confidential information from a previous employer or client. Tell the candidate what happens next and when the test material will be deleted.
Translate the result into a first-week lane
A work sample is only one hiring input. If the person moves forward, process one historical week in read-only systems and review every exception. Keep live scope narrower than the exercise when systems contain sensitive data or irreversible actions. Supply accepted examples, a named reviewer, fixed check-in times, and a fast exception route. Access follows the approved task rather than the job title. Prefer individual identities, multifactor authentication, role-based permissions, and audit history where tools support them. Increase scope only after the return format, correction loop, and boundaries work reliably. One successful simulation does not justify broad inbox, financial, customer, candidate, or administrative access.
Measure the pilot with a denominator
Track documents indexed, evidence-supported matches, missing items, and corrections. Every count needs the reviewed volume, task class, and correction reason. One defect in five high-risk records differs from one formatting fix in five hundred low-risk fields. Review a representative sample plus every protected or unusual item. Separate source defects from assistant errors; missing information correctly flagged is not inaccurate work. Look for repeated causes that can be fixed in the instruction, example, permission design, or training. Use the scorecard for coaching and scope decisions, not to manufacture a performance claim from a tiny sample. Record changes to the rules before measuring the next batch.
Choose the next step deliberately
At the pilot’s end, decide whether to keep the lane, narrow it, coach one behavior, improve the input, or stop. Write down the evidence. Do not expand access merely because easy cases were completed quickly. Expansion follows stable quality on the exact task, correct treatment of exceptions, and a review routine the owner can maintain. Add one adjacent task at a time and update the boundary first. If the workflow caused confusion, repair it before blaming the candidate or assistant. This produces a fairer hiring decision, a safer start, and a routine that can survive handoffs without depending on private memory.
Authoritative background and next step
Review the FTC guide to protecting personal information and the NIST Cybersecurity Framework 2.0. Apply employment, privacy, security, tax, professional, and sector rules with qualified advisers.
See the related staffing service or request a role and first-task plan.