A hybrid townhall that nobody watches has usually failed in four specific ways, none of which looks like a technical fault on the day. The remote office drifts out of the habit of attending. The Q&A produces a handful of polite questions, because everyone's name is attached to theirs. The chief executive's message registers in the room and arrives thinner everywhere else. The recap turns up three weeks later, long after the announcement was chewed over in every group chat in the company.

All four are decided during planning, by people who do not think of themselves as making a production decision. So this is ordered the way the problem reaches whoever owns internal comms: failure mode first, then the mechanism that closes it.


1. The remote office quietly stops attending

How it shows up. Not as a complaint. Jakarta's concurrent viewer count is a little lower this quarter than last, and lower again the quarter after. Nobody escalates, because raising it costs more effort than not joining. By the time it reaches you it is a culture conversation, and it started as an audio problem eleven months earlier.

Why it happens. The room is checked and the feed is not. Somebody claps into the mic at 8.40am, the engineer nods, and everyone assumes that covered the people watching from another country. The failures that drive remote offices away are dull and cumulative: a question asked from the floor without a microphone, so the answer arrives as a reply to something nobody heard; a slide built for a six-metre LED wall being read on a 14-inch laptop; a locked wide shot that turns your leadership team into grey shapes behind a table.

What closes it. Run the checks on the outbound stream, with a named person in each office joining the technical rehearsal as an ordinary viewer and describing what they get. Not the engineer watching their own monitor. Make repeating in-room questions into a handheld mic a rule the show caller enforces rather than a courtesy the presenter remembers. Proof the deck at laptop size, and put a camera on whoever is speaking rather than one locked off at the back.

None of that is expensive. It is what makes the thing a broadcast, not a video call pointed at a stage.

2. The Q&A gets three softballs

How it shows up. The Q&A block runs for twelve minutes, produces four questions about parking and the new expense system, and ends early. Leadership concludes the organisation is content. It isn't. The questions people actually have were composed, read back, and deleted.

Why it happens. Attribution. When submitting means putting your name on a screen in front of the people who decide your next promotion, the only staff who submit are the ones with nothing at stake, and they do not ask the question the room is thinking about.

What closes it. Anonymous submission by default, moderated. Attendees scan a QR code, submit from their own phones with nothing to install, and the in-room and remote audiences drop into one queue rather than two nobody reconciles. We run this on Questro, Interframe's own Q&A and captioning product: a producer merges duplicates so the same concern asked eleven ways arrives once, upvotes sort the queue by what the company wants answered, and anything sensitive is flagged to the comms lead before it reaches the confidence monitor. Anonymous never means unvetted.

Then the part that isn't technical, and matters more. Agree beforehand what gets answered live, what gets a written follow-up and what gets declined openly, then publish the unanswered questions with a line each. Staff cannot see how many questions were submitted. Their only evidence that submitting works is whether last quarter's difficult question was ever acknowledged.

The honest limit: we can make submission anonymous and the queue clean, and we cannot make anyone answer. If the same subject is filtered out two quarters running, your submission volume will tell you before your engagement survey does.

3. The message lands differently in each office

How it shows up. The room reads the announcement as measured and confident. Manila reads the same eleven minutes as evasive. Nobody is wrong, because they did not receive the same thing.

Why it happens. The room gets tone, pauses and the reaction of the people around it. The stream carries a compressed version, and for anyone working in a second language it compresses further. A dry aside that reassures in person reads as dismissive a week later.

What closes it. Live captions on the stream and in the room, with speaker names, product names and any restructuring vocabulary loaded in beforehand, so the terms that carry the announcement are the terms that appear on screen. A host in each office with a live mic, which gives every location a visible person rather than a portal. And a scripted read for the segment that matters, timed at rehearsal, because an executive who improvises for four minutes produces four minutes that travel badly.

One caveat the platform vendors have no reason to state: AI captions are strong for clear, prepared speech and weaker exactly where the stakes are highest. For a scheduled quarterly update they are enough. For a restructuring announcement, or anything where a mistranslated clause becomes an industrial-relations problem, budget for a human interpreter on the languages that carry contractual weight.

4. The recap arrives after the news cycle died

How it shows up. A 90-minute unnavigable file, posted to the intranet a fortnight later, opened by roughly nobody. Meanwhile the version of events circulating internally was set within two hours of the broadcast ending, by people who were half-listening.

Why it happens. Nobody decided during the show what the recap would be, so someone has to watch the whole thing back to find out.

What closes it. Decide the cut before the show and capture it during. The show caller marks segment in and out points on the run sheet live, against timecode, so the edit is a trim rather than a search. The captions already exist as a transcript, so the written summary starts as a subedit instead of a transcription job. A captioned, chaptered recording then goes to your intranet the same working day, with a couple of short clips for the offices that missed the live slot. Publish the question list alongside it, and the recap does a second job: it proves the Q&A was real.


What you can actually count

You can count four things, and you should be careful about the rest.

  • Concurrent viewers by office — the only number that answers whether Jakarta watched. Read it against that office's headcount, not against the total.
  • Drop-off timing — the minute people leave, which points at a segment rather than at general disinterest.
  • Questions submitted, per office — a better attention signal than viewing figures, because it costs effort.
  • Replay views in the first 48 hours against the following fortnight, which separates catching up from never intending to watch live.

Say your Kuala Lumpur office has 180 staff and peaks at 40 concurrent viewers, with a drop at minute 34. That is not a result, it is a question. Minute 34 is a segment with a name on it, and next quarter's run-of-show can move it or cut it to a written note.

What you cannot do is compare any of this to an industry average. Published attendance figures describe other companies, with other headcounts and other reasons to care. Your only honest comparison is your own last three townhalls, measured the same way.

The seam nobody owns

Internal comms owns the message. HR owns the agenda and the policy content inside it. IT owns the network, the platform and who gets through the firewall. Facilities or the venue owns the room and the meeting-room screens. Every slice has a name against it. The join between them has none.

That join is where the failure lives. The VPN blocking the player is IT's. The meeting-room TV running a two-year-old app is facilities'. But the fact that one office could not watch the chief executive is yours. Name a single technical owner for the whole path before the calendar invite goes out, and give them the standing to test the office end rather than only the stage end.

When you don't need any of this

A 40-person all-hands in one office needs a microphone people can actually hear, a room where nobody sits behind a pillar, and a recording for whoever is on leave. It does not need a production company, and a vendor who tells you otherwise is quoting on the wrong problem.

Production earns its cost when the audience is split across offices and you are accountable for whether the far ones showed up, when leadership is announcing something that cannot be walked back, or when the recording has an audience beyond the people who missed it. Below that, spend the money on a better mic and go back to the script.


Frequently asked questions

How do you stop a remote office from dropping out of a hybrid townhall? Check the feed, not the room. A named person in each office joins the rehearsal as a viewer, in-room questions get repeated into a mic, and slides are proofed at laptop size.

Should townhall questions be anonymous? Yes, by default — attribution is the reason you get polite questions. Moderation still applies, so nothing unvetted reaches the stage.

What can you actually measure after a hybrid townhall? Concurrent viewers by office, the minute drop-off happens, questions submitted per office, and replay views in the first 48 hours. Compare against your own previous townhalls, not a published average.

Do we need a production company for an all-hands meeting? Not for 40 people in one office — buy a good mic instead. Production earns its cost across multiple offices, on irreversible announcements, or when attendance is a number you report upwards.

How quickly can the recording and recap go out? Same working day, if the cutting points were marked on the run sheet during the show instead of hunted for afterwards.


Interframe produces corporate townhalls and internal communications for Singapore and regional teams — stage AV, multi-office streaming, moderated Q&A on Questro and same-day recap assets, on one technical plan. Tell us your employee count, office locations and date, and we'll come back within one business day with a plan and a quote.