The master and its cuts: why a video is the wrong unit of production#
Ask a small software team how their video marketing is going and you will hear a count of videos: we shipped a landing clip, a couple of social posts, a YouTube walkthrough, an onboarding email. Count them as separate projects and the math is brutal, because each one is a script, a recording, a voice track, and an edit. That is the trap. The unit of production is not the video, it is the recording you cut many ways. One capture of the product doing its job becomes a landing hero, a square feed clip, a vertical Short, an email GIF, an app-store preview, and a YouTube upload, and the gap between the cheap version of that sentence and the expensive one is entirely whether you planned for it before you hit record.
The channels are not optional, because your buyers are not on one surface. Wistia's 2026 State of Video, drawn from its hosting corpus, reports that 81% of teams share video on LinkedIn and 76% on YouTube, with about half on Instagram, just under half on Facebook, and under a quarter on TikTok1. HubSpot's 2024 Video Marketing Report, surveying more than 500 marketers, found 83% naming social media as their highest-ROI video channel and 81% using it at all2. Landing pages, email, and product docs sit on top of that. A demo that exists on exactly one of these surfaces has left most of its production cost unrecovered.
Teams already feel the pull. Over half the companies in Wistia's data repurpose video into social clips, and nearly 90% reuse webinar footage as replays, cuts, and emails. What most lack is a spec that says which cut each channel actually wants, so the repurposing stays ad hoc and the awkward formats, portrait and square and the app-store preview, quietly get skipped. This is that spec. The production playbook covers making the master; this pillar is what to do with it afterward.
The channel matrix: format, length, aspect, and the number each grades on#
Here is one recording, delivered to nine surfaces. Length is the retention math worked out placement by placement elsewhere; the columns that genuinely differ per channel, and that most guides leave out, are the container, the aspect ratio, and the single number that channel is judged on.
| Channel | Format | Length | Aspect | Grades on |
|---|---|---|---|---|
| Landing-page hero | MP4/H.264, muted autoplay loop | 30-60 s | 16:9 | Play rate, then time on page |
| Product docs / README | MP4 or looping GIF, no audio | 10-30 s | 16:9 | Task completion |
| YouTube | MP4, H.264 High Profile, AAC, 1080p | 1-5 min | 16:9 | Watch time, then click-through |
| LinkedIn feed | Native MP4, not a pasted link | 15-30 s | 1:1 or 4:5 | Three-second views |
| TikTok / Reels / Shorts | MP4, captions burned in | 15-30 s | 9:16 | Completion, replays |
| Animated GIF or poster linking out | first frame + 5-10 s | ~16:9 | Click-through to a hosted page | |
| Sales follow-up | MP4 link behind a poster | 45-90 s | 16:9 | Reply rate |
| Product Hunt | YouTube link in the gallery | up to ~2 min | 16:9 | Upvotes, Product-of-the-Day |
| App Store / Play preview | .mov or .mp4, H.264 | 15-30 s | 886x1920 portrait | Install conversion |
A few of those cells are load-bearing and earn a source. YouTube treats 16:9 as its standard aspect and ingests an MP4 in H.264 with AAC without complaint, which is unremarkable precisely because that container is what MDN calls the industry standard for video, widely supported across devices and browsers3; a 16:9 H.264 file is the master you keep, and everything else on the table is a re-fit of it. LinkedIn accepts 4:5, 9:16, 16:9, and 1:1, and recommends 15-30 seconds so a clip qualifies for every feed placement4, which is why a square or vertical native upload out-reaches a pasted YouTube link. Product Hunt is blunter than most: its gallery accepts only YouTube video links, and about 53% of Product-of-the-Day winners since 2021 included a video5. And Apple's App Store Connect nails its corner of the table down hard: an app preview runs 15-30 seconds, ships as .mov/.m4v/.mp4 in H.264, sits at a portrait 886x1920 for current iPhones, allows up to three per language, and caps at 500 MB6.
Read straight down the aspect column and the whole case for a master falls out. Five of these surfaces want 16:9, two want 9:16, one wants 1:1, and one wants a fixed 886x1920 portrait. That is not nine shoots. It is one recording and a handful of reframes, plus a landing-hero placement that does not wreck load time, a README loop that stays current, and a Product Hunt cut built the night before launch.
What changes from the master, and what never does#
The invariant across every row above is the recording itself: the pixels of your product performing one real task, cursor and all. What varies is a short, bounded list of transforms applied after the fact.
| Transform | What it does | Driven by |
|---|---|---|
| Trim | Cut to the channel's length budget | Retention curve for the placement |
| Reframe | Pad or crop to the target aspect | Feed vs hero vs store |
| Caption burn | Bake word-timed captions into pixels | Muted autoplay |
| Re-encode | Container and codec for the platform | Upload requirements |
| CTA swap | The single ask, per funnel stage | Where the viewer sits |
None of those is a reshoot. Trimming to length is a compose-time cut, not a re-record. Burning captions is mandatory the moment a clip autoplays muted, which on a feed is always, and it is a rasterizing step, not a rewrite of the take. Reframing is the one with teeth: taking a 16:9 desktop capture to a 9:16 feed clip by cropping throws away two-thirds of the frame, and the crop-versus-pad math is genuinely unkind, which is why the honest vertical answer is to record a narrow viewport rather than crop a wide one. That single exception aside, every channel deliverable is the master plus a transform.
This is where treating the demo as a spec rather than a captured performance pays for itself. If the master is a storyboard, retargeting it to a new aspect is a re-render at a different output size instead of a new take, and the reframe problem dissolves because you render the flow into a portrait viewport rather than crop a landscape one. aidemo, our open-source engine, does this from an agent-authored storyboard; the honest limits are that it is browser-only and has no GUI timeline editor, so it is a poor fit for a one-off screencast and a good one when the same flow has to reach nine places and stay honest as the product ships. Automated product demo videos is the general form of that argument.
The arithmetic of one recording across nine placements#
Put numbers on it. A hand-made 60-to-90-second demo is a two-to-three-hour job once you have scripted it, recorded a clean take, voiced it, and keyframed the polish. Treat each channel as its own project and nine placements is nine of those, call it twenty to thirty hours, and every one of them is atomic, so a single UI change reruns most of the bill.
Now amortize. Record the master once, then produce each channel cut as trim plus reframe plus caption plus re-encode. Even done by hand in an editor that is ten to twenty minutes a cut, so nine cuts land in two or three hours on top of the one master, roughly a quarter of the bespoke cost. Generated from a spec, it is closer to a batch render than an editing session, and the marginal cost of the tenth placement rounds to the compute it takes to encode. The recurring bill is where the gap becomes decisive: when the UI moves, the bespoke shop re-performs nine takes and re-syncs nine sets of captions, while the master-and-cuts shop re-records one flow and re-renders nine outputs from the same storyboard. The distribution version of regenerate-rather-than-re-record is the same lever the automation pillar pulls, applied across channels instead of across releases.
Read the metric the channel grades on, not "views"#
The matrix's last column is the one teams flatten and should not. "Views" is not a metric; it is nine different metrics wearing one label. A landing hero is graded on play rate and whether the clip lifts time-on-page and conversion, not on raw plays. A YouTube upload is graded on watch time first and click-through second, because the ranking runs on the former. A feed clip lives or dies on the three-second view, since a muted autoplay that fails to arrest the scroll in three seconds was never really watched. A sales follow-up is graded on a single reply. A Product Hunt video is graded in upvotes. Average all of those into one "views" number and you have hidden every signal that would tell you which cut to fix. Which numbers to instrument, by goal, is a spoke of its own, as is the honest read on whether video actually lifts conversion: the short version is that it helps when the clip is relevant and the page still loads fast, and backfires when a heavy hero tanks the load time it was meant to improve.
Where one recording genuinely isn't enough#
"One recording, every channel" is the default, and it holds for most of the matrix. Four places it breaks, named plainly so you can budget for them.
First, the App Store. Apple's preview must be actual in-app footage at a fixed portrait resolution, so a landscape web-app demo cannot simply be letterboxed to fit; a mobile app earns its own capture. Second, hard vertical. A 9:16 cut of a wide desktop UI is the reframe with real loss, and when the crop would gut the content, you record a narrow viewport instead, which is a second take by design. Third, the interactive tour. A clickable product walkthrough is a different artifact than a video, not a cut of one; it captures HTML and replays clicks, and it cannot ride inside an email or a README the way an MP4 can. Fourth, depth. A 30-second hero loop and a ten-minute YouTube tutorial serve viewers with opposite amounts of attention to spend, and past a point that is a different script, sometimes a different recording, not a trim.
These four are the exceptions that prove the rule: they are the only rows where the aspect, the source footage, or the medium is not something you can reach from the master with a mechanical pass. Everything else on the matrix is one master and a transform. The teams that win the distribution game are not the ones making the most videos; Wistia's data shows most are already shipping at least one a month. They are the ones making a single recording earn its cost nine times over, by deciding the cuts before they record and treating each channel deliverable as a render, not a reshoot. Where to put the demo once you have it turns this matrix into a placement checklist.
Sources#
- Wistia — State of Video Report: Video Marketing Statistics for 2026
- HubSpot — 2024 Video Marketing Report
- MDN — Media container formats (MP4 as the industry standard)
- LinkedIn Help — Video ads advertising specifications
- Product Hunt — Preparing for your launch
- Apple — App preview specifications (App Store Connect)
FAQ#
Do I need a different video for each marketing channel?#
No, you need one master recording and a different cut per channel. The flow you capture stays the same; what changes is the trim to length, the aspect (16:9 for a hero and YouTube, 1:1 or 4:5 on LinkedIn, 9:16 for Shorts), burned captions for muted feeds, and the container. Two genuine exceptions need their own capture: an App Store preview must be real in-app footage at 886x1920 portrait, and a deep YouTube tutorial is a different script than a 30-second hero loop.
What kinds of marketing videos does a software company actually need?#
Think in placements, not genres. The working set is a landing-page hero, a product or docs clip, a social cut, an email GIF, a sales follow-up, a Product Hunt video, an app-store preview if you have a mobile app, and a YouTube upload. Wistia's 2026 data shows where the audience is: 81% of teams post video on LinkedIn and 76% on YouTube. Each of those placements can be a cut of the same recorded flow rather than a separate shoot.
Is video marketing worth it for a small software startup?#
It is worth it when one recording serves many channels, and a money pit when every channel gets its own shoot. HubSpot found 83% of marketers name social media their highest-ROI video channel, and short-form the highest-ROI format, so the return is real. The cost that sinks small teams is per-channel re-recording, especially when the product keeps shipping. Solve distribution as a render matrix off one master, and a two-person team can cover nine surfaces.
