User Survey Synthesis

Key Findings

What the survey data shows across the full sample (N=101).

• 1
  The sample skews heavily toward early adopters / power users, not a representative cross-section of the general Vietnamese office workforce.

  82.2% of respondents use AI "Hàng ngày" (Daily), and only 1% "Rất hiếm khi / Chưa từng" (Rarely/Never) - the non-user branch (path B) has just n=1/101. This sits far above the broader market baseline ↗ Market Research - Gen Z: 40% daily; general office workers far lower, with the original PRD citing 19% daily AI use among Vietnamese office workers ↗ PRD. The gap likely comes from the distribution channel (personal networks and AI community groups). Every barrier and pricing figure below reflects people who already use AI regularly - the "rarely/never used" segment remains unvalidated, and interviews need to close that gap.
• 2
  Subscription dominates pay-per-task when asked directly: 73.3% chose Subscription, only 11.9% Pay-per-task, 14.9% Prepaid Credits.

  This is the first direct test of the pay-per-task hypothesis from market research ↗ Market Research ("willingness to pay per completed document" was an open, unvalidated question). Result: pay-per-task as the primary pricing model is not strongly supported in this sample.
• 3
  But most respondents are still willing to pay per task once given a concrete price point (Q7).

  Only 30.7% chose "not willing to pay per task, prefer subscription" and 8.9% "not willing to pay for an AI tool at all" (39.6% combined reject per-task entirely). The remaining 60.4% gave a specific price they'd pay per task (most common: 5,000-10,000 VND at 22.8%). Even within the group that chose Subscription in Q6, 46 of 74 people (62%) still gave a reasonable per-task price in Q7 instead of flatly refusing. Pay-per-task doesn't beat subscription as the primary model, but it has a place as an add-on / hybrid - not fully rejected.
• 4
  Concern about company data leakage (47.5%) and concern about AI eroding independent thinking (47.5%) are the two largest barriers, tied for first.

  Both outrank the price barrier ($20/month too high: 44.6%) and prompt-related barriers (not knowing how to ask: 22.8%; needing multiple back-and-forth turns: 37.6%; shallow AI answers: 26.7%). This is a sample that already uses AI daily - they keep using it while carrying both concerns in parallel, not concerns severe enough to make them stop.
• 5
  Claude has the highest average rating among tools evaluated (4.42/5) but far lower usage than ChatGPT and Gemini.

  66% have used Claude vs. 98% for ChatGPT and 94% for Gemini. This lines up directly with the "retention driven by lock-in, not quality" finding in market research ↗ Market Research: ChatGPT/Gemini win on reach (habit/integration), Claude wins on perceived quality but hasn't won on adoption.
• 6
  Task-by-role data shows a clear signal for Skill-first design.

  TA/HR (n=25, the largest group) concentrates heavily on matching CVs to JDs (76%), writing JDs (72%), and extracting CV data (68%) - exactly the priority Skills market research already proposed. Job Seekers (n=15) focus on optimizing a CV against a JD (73.3%), tied with automated multi-platform job search (66.7%) and mock interviews (66.7%) - automated job search shouldn't be dropped, since it's a fundamentally different Skill (sourcing/aggregation, not content generation). This is the first quantitative evidence backing the task inventory.
• 7
  "Productivity" framing beats "peace of mind" framing decisively: 85.1% vs. 14.9%.

  Consistent across nearly every role (TA/HR 22 vs. 3, Marketing 15 vs. 4, Sales 8 vs. 1, Job Seeker 13 vs. 2). A strong quantitative signal for positioning messaging.
• 8
  17 of 101 respondents (16.8%) self-entered "Other" for their role (Q1) - IT/Dev/QA, Purchasing, Logistics, Design, Legal, Project Manager, ERP consultant, Supply Chain.

  Most of this group sits outside Klerio's target audience (non-technical office workers: TA/HR, Marketing, Admin, Sales, Job Seeker, Accounting, Manager) - likely leakage from the personal-network distribution channel, not a signal to expand the role taxonomy.

Surprising / Unexpected

Findings that ran against expectations set by prior research.

• 1
  Pay-per-task loses decisively to subscription in the direct question (11.9% vs. 73.3%).

  Market research treated this as an unvalidated hypothesis that could still win ↗ Market Research; the survey data leans heavily toward subscription instead. This is the single most important finding from this run, and needs careful discussion before locking in the primary MVP pricing model.
• 2
  Concern about "reduced independent thinking" (cognitive dependency) is exactly as high as concern about data leakage (47.5% each).

  Market research frames Privacy Guard as the primary differentiator, but the data shows the fear of AI making users lazier thinkers is just as large. This is a second, independent axis of concern the product/positioning doesn't yet answer clearly - Privacy Guard addresses axis 1, not axis 2.
• 3
  Q10 (reasons for churn / not using AI) surfaces this exact "cognitive erosion" theme in respondents' own words.

  Multiple open-text answers mention "brain rot," reduced focus, AI giving sycophantic answers with no critical pushback, and noticing themselves getting lazier. This isn't noise - it's a clearly recurring theme in the open text that reinforces the 47.5% figure from Q4.
• 4
  Only 1 of 101 respondents fell into path B ("rarely / never used AI").

  The entire Label B content of Q4 and Framing B of Q5 (designed specifically for non-users) went essentially untested by this run. Understanding the non-user / light-user segment needs a different distribution channel next time, or interviews to cover the gap.

So What for Klerio

Product and positioning implications, and the open hypotheses behind them.

Open Questions

Unresolved gaps this run couldn't answer.

• 1The entire path B (rarely/never used AI) has almost no data (n=1). This is an important segment for understanding "why 81%+ know about AI but don't all use it regularly" (the paradox raised in the survey's own intro). Interviews should actively recruit this group.
• 2Accounting/Finance has only n=1 - not enough to say anything about this group, despite being one of the survey's 8 official roles. Needs more responses or interview coverage.
• 3Unclear how willingness to pay changes when tied to a specific Skill (Q7 asked about price in general, not tied to one Skill) - interviews can dig into this with real-Skill scenarios.
• 4Unclear whether "cognitive dependency" is a theoretical worry or a lived experience - Q10 gives quotes suggesting it's real ("noticing myself getting lazier"), but interviews are needed to understand the depth and specific context.
• 5Should responses from out-of-target roles (IT/Tech, Legal, Logistics, ERP, Supply Chain...) be excluded from the main analysis in the next run? Run #1 kept all N=101 to avoid losing data, but this needs a decision before the data is used to lock in product decisions.

Interview Follow-ups

Goal: distinguish the 3 competing hypotheses behind subscription's decisive win (unfamiliarity / habit / volume-confidence), and test reactions to the hybrid prepaid → auto-suggest-subscription idea.

• 1Test unfamiliarity bias: "Have you ever used a tool that charges 'pay for what you use' (top up, then top up again once it runs out)? In what area?" - a "never" answer supports unfamiliarity bias.
• 2Test the reason for choosing subscription (open-ended): "Why did you choose a fixed monthly fee over paying per use before? Did you actually compare the two options, or was it just the familiar choice?" - distinguishes habit bias from real deliberation.
• 3Test volume-confidence / loss-aversion (two-sided scenario): "Do you worry that with a subscription, if you use it lightly that month, you're wasting money? Conversely, if you paid per use and used it a lot, would you worry it costs more than a subscription?" - see which fear dominates.
• 4Direct scenario test: "If you had two options for the same AI tool: (A) 300,000 VND/month unlimited use, or (B) about 8,000 VND per completed task, pay for what you use - which would you choose, and why?" - forces a real comparison instead of an abstract answer.
• 5Test the light-user segment (especially valuable if recruiting from path B): "If you were just starting to use an AI tool and didn't know yet whether you'd use it a lot or a little, would you want to start with a fixed fee or top-up-as-you-go? Why?"
• 6Concept test for the hybrid model: "Imagine a tool where you top up in advance and pay for what you use; if the system sees your usage growing, it proactively suggests switching to a monthly plan because it would be cheaper for that level of use - what do you think of this idea? Is it appealing enough to make you try a new tool built this way?"

Details

Full survey data behind the findings above. Role names, answer options, and open-text quotes are kept in the original Vietnamese wording shown to respondents.

"Khác" (17) breakdown: IT/Dev/QA (5), Kỹ sư (2), Purchasing, Chuyên viên nghiệp vụ, Project Manager, Pháp chế, Logistics, Design, Tư vấn triển khai ERP, Quản lý chuỗi cung ứng & Vận hành.

Kế toán/Tài chính (n=1) and individual "Khác" groups (mostly n=1): too small to draw conclusions, noted qualitatively only.

Cross-tab (within the 74 who chose Subscription in Q6): 28 (37.8%) also chose "prefer subscription / not willing per-task" in Q7 - consistent; 46 (62.2%) still gave a specific per-task price (16 chose 5-10k, 9 chose <5k, 8 chose >20k, 6 chose 10-20k, 7 chose "not willing to pay for AI tool" generally).

By role (n≥5): TA/HR 22 productivity / 3 peace-of-mind; Marketing 15/4; Job Seeker 13/2; Sales 8/1; Manager 7/1; Admin 6/1. Pattern consistent across every role - none leans toward "peace of mind."

Sources

Where this data comes from.