Challenge
Cất Cánh is not a conventional talk show. Each episode features multiple speakers, each carrying a distinct story, and the audience spans grandparents tuning in to VTV1 on Saturday evenings all the way to Gen-Z viewers streaming on VTVGo and Facebook.
Before the VoA system, the production team had three feedback channels: Facebook page comments, livestream chat, and aggregate ratings. All three shared the same four limitations:
- Unstructured — comments mix opinions on MC, speakers, music, visuals, politics… extracting actionable signal was manual and slow.
- Not real-time — feedback arrived after the broadcast ended, eliminating any chance to adapt in-show.
- No coverage of the TV audience — older viewers watching traditional TV rarely comment online; this segment, while large, was effectively invisible to the team.
- Regulatory friction — national broadcast carries strict content-moderation requirements. Most open-feedback solutions (public forms, polling apps) cannot meet them.
In a national-television context, VoA is not "a survey tool." It is the mechanism for measuring real audience reaction per segment, per speaker — instantly, structured, and safe.
Solution
Filum.ai partnered with the Cất Cánh production team to build the first real-time VoA system deployed on a Vietnamese national-television program.
2.1 Dynamic QR on-air
Every speaker and every question in the show script has its own QR code. Viewers raise their phones, scan, and land on a survey.filum.ai page already contextualized for that specific moment in the show.
The QR is rendered both large-format for the studio audience (LED on stage) and as a persistent corner overlay for at-home viewers:
2.2 Live dashboard on the studio LED
Incoming responses flow into the Filum AI engine for real-time sentiment classification — topic, emotion (pride, admiration, satisfaction, empathy, hope…), word cloud — and the result is rendered directly onto the studio LED. The MC can read live audience feedback on-air.
Implementation note: Dynamic QR + context-aware survey routing via URL parameters (each question = one campaign ID). Multi-label Vietnamese sentiment analysis — not just Positive/Negative/Neutral, but resolving "Pride", "Admiration", "Inspiration", "Trust", "Respect"… calibrated for talk-show context. A time-window filter (e.g. 20:10 cohort) admits only responses submitted during on-air, eliminating post-broadcast noise.
2.3 Three tiers of risk control for national broadcast
The decisive difference between a standard survey and VoA for national TV is content risk. Any response that appears on the VTV1 broadcast feed must pass three checks:
- Tier 1 — Automated filter: strips abusive language before content enters the editorial queue.
- Tier 2 — Editorial review: every response is reviewed by the production-team editor before appearing on the broadcast feed.
- Tier 3 — Time gating: only responses submitted during the on-air window are accepted. Out-of-window submissions are discarded — anti-spam, anti-botting by design.
Results
| Metric | Before Filum | With Filum (Gala 2025 → Apr 2026) | Δ |
|---|---|---|---|
| Structured responses per episode | 0 (open comments only) | 120–200+ classified responses | +∞ |
| Feedback latency to production | D+1 (manual comment scan next day) | Real-time, in-show | −24h |
| Geographic coverage | Online commenters (mostly HCMC, Hanoi) | Nationwide — Hanoi, Nghệ An, Đà Nẵng, Huế, HCMC, Cần Thơ, Lào Cai… | +5 regions |
| Per-speaker emotion measurement | Infeasible | Resolved per question / per speaker | New capability |
| Response rate (scan → reply) | — | >30% | — |
For viewers: a simple gesture — scan, type, send — accessible to older and younger audiences alike. Responses appear on the broadcast screen instantly; viewers see themselves heard, on-air, in the moment.
For speakers: they receive live audience feedback while still presenting. Before, no such mechanism existed — speakers only learned what audiences thought after the broadcast ended.
For the production team: the dashboard is simultaneously an operating tool (real-time audience pulse) and a stage visual (a design element of the show). One product, two roles.
Testimonial — VTV elevated QR into the official 2026 season trailer
VTV positioned the QR interaction alongside other defining elements of Cất Cánh 2026 (celebrity guest commentators, surprise gifts for speakers) and lifted the description verbatim into the official trailer:
"Viewers simply raise their phone, scan the QR code, leave a comment — and their emotions flow straight to the program and appear on screen instantly." — Cất Cánh 2026 official trailer, VTV1
Discussion
Two lessons from operating VoA on national television — transferable to any live broadcast or large-scale event:
Real-time is not the same as unmoderated. Unlike consumer real-time products that default to "anything goes," national broadcast requires moderation. The answer is not to drop moderation but to make moderation fast — automated Tier-1 filter plus editorial Tier-2 review complete within seconds per entry. Speed preserved without compromising broadcast safety.
Counts are not the output — portraits are. Every response carries a name, a region, an age group, an emotion. What Cất Cánh now owns isn't "1,000 responses" — it's 1,000 specific audience portraits: who is watching, what they feel, in response to which speaker. This is the long-term data asset for VTV.
What's next
Every episode of Cất Cánh is a product-iteration loop. Filum and VTV are building this together, not simply running a vendor relationship. Forward directions:
- Accumulated response data per episode → long-term asset for VTV: understand audiences by topic, by guest, by moment.
- Pattern replicable to VTV3, HTV, regional broadcasters, conferences, award ceremonies — any live format where audience listening at scale matters.
References
- ▶ Video: Cất Cánh Khởi Tạo — Cất Cánh 2026 trailer
- 📘 Cất Cánh fanpage — VTV1: facebook.com/CatcanhVTV
- 📺 Cất Cánh April 2026 — Sức Bật: vtv.vn/video/cat-canh-suc-bat-108691947.htm
- 📺 Cất Cánh January 2026 — Thế Hệ Kiến Tạo "Hiện Tại": vtv.vn/video/cat-canh-17-01-2026-108686591.htm
How these numbers were measured
N ≈ 1,000+ free-text responses captured via QR codes broadcast on VTV1 and VTVGo, between Gala Cất Cánh 2025 (December 2025) and Cất Cánh April 2026. "Response rate" = (responses with content) ÷ (QR scans that opened the survey landing page). Sentiment analysis and topic classification run automatically on the Filum AI engine; a three-tier content filter (automated → editorial review → time-window gating) is applied before any response appears on the broadcast feed. This case study was reviewed and approved by the Cất Cánh production team prior to publication.
- Cất Cánh 2026 official trailer, VTV1 — YouTube: youtube.com/watch?v=ub5wZGiHzc8 (accessed May 2026).
- Per-episode response breakdown compiled by Cất Cánh production team: Gala 2025 (200+ responses / 5 prompts), January 2026 (120+ / 4), March 2026 (130+ / 4), April 2026 (150+ / 4). Cumulative N: 1,000+.
- "Response rate" defined here = (responses with written content) ÷ (QR scans opening the survey landing page). Scans that exit without engagement are excluded.
- Vietnamese multi-label sentiment analysis: Filum AI engine. Emotion labels used in-show ("Pride", "Admiration", "Satisfaction", "Empathy", "Hope", "Inspiration", "Trust", "Respect") were trained specifically for the talk-show television context.
- Three-tier content filter (automated → editorial → time-window) applied per VTV national-broadcast moderation requirements.
- Commercial figures undisclosed under customer agreement.
