Delivering the same piece to several destinations used to take me most of an afternoon. Different shapes, different lengths, different loudness, different file requirements, each done by hand.
The workflow below produces all of them from one master and takes about twenty minutes, most of which is the computer working.
Master first, always
The principle everything else depends on. Export one high quality master and derive every delivery version from it or from the same timeline, rather than treating any delivery version as the source.
The master should be at your project's native resolution, in a lightly compressed intra-frame codec, with clean audio. It will be large and that is correct.
Deriving a small file from a large one loses a little. Deriving a small file from another small file loses a lot, because you are compressing already compressed material and the artefacts compound.
Aspect ratios: plan in the edit, not at export
The thing that costs most time if handled badly.
If you know a piece needs vertical and square versions, set up guides in the edit showing those crops and compose so the important content sits inside all of them.
Then the derived versions are a reframe rather than a rebuild. Most editing software supports auto-reframe, which tracks the subject and crops accordingly, and it works reasonably for single-subject content and poorly for anything complex.
Where it fails, manual keyframed position adjustment on a duplicated sequence takes ten minutes rather than the hour a full recut would take.
Text and graphics generally need redoing per ratio regardless, because a lower third designed for wide will not fit vertical. Keeping graphics on their own layers rather than baked into the picture makes this manageable.
Loudness is the requirement everyone ignores
Platforms normalise audio to a target loudness. Deliver something louder and it gets turned down. Deliver something quieter and it may be left alone or turned up.
The consequence is that mastering loud, which used to be the way to sound impressive, now just means your dynamics get squashed and then the whole thing gets reduced anyway.
Targets vary by platform and mostly sit in a similar range for online video. Mixing to a sensible integrated loudness with peaks kept below the ceiling produces something that survives normalisation everywhere.
Every serious editing application has a loudness meter. Use it rather than mixing by ear against whatever your speakers do.
Codec and bitrate for delivery
Every platform re-encodes what you upload, which changes what you should send.
Send a high quality file. Not the largest possible, but well above the bitrate the platform will end up using, because their encoder works from what you give it and giving it something already degraded produces worse results.
The common mistake is compressing hard to make upload faster. That saves minutes and costs quality permanently.
Match the frame rate to the source. Converting frame rates at export introduces judder that is entirely avoidable.
And keep the resolution native rather than upscaling. Upscaling to hit a higher tier does not improve anything and produces a softer image at a higher bitrate.
The naming convention that saves the confusion
With six versions of one piece, files get mixed up. A convention removes the problem.
Project name, then version, then aspect ratio, then destination, then date. Consistently, in that order.
The date in year-month-day form means files sort chronologically, and the version number means the client asking for "the one you sent Tuesday" is answerable.
Batch exporting
The step that turns an afternoon into twenty minutes.
Most editing applications have a queue or a separate encoding application. Set up every delivery version, add them all to the queue, and run it.
The presets can be saved, which means the second project takes no setup at all.
Run it while doing something else. This is the one part of the process where the computer works and you do not, and treating it as time to sit and watch a progress bar is wasted.
Checking before delivering
A short list that has caught real errors.
Open every exported file and play the first and last ten seconds. Truncated exports and files that failed silently are more common than they should be.
Check the resolution is what you intended, particularly if using proxies.
Check the audio is present on both channels. Mono material sitting on one side only is an easy mistake and completely invisible until someone listens on headphones.
And check the vertical and square versions for anything cropped out — a subtitle running off the edge, a graphic half visible, someone's head cut off.
That last one has embarrassed me twice, both times on versions I exported automatically and did not watch.
Thumbnails and stills from the same pass
A small addition that saves a separate job later.
While the export queue is running, pull the frames you will need for thumbnails, social posts and any promotional use, straight from the timeline at full resolution.
Exporting a still from the source is considerably better than screenshotting a compressed delivery file, which is what most people end up doing because they did not think about it at the time.
Three or four candidate frames, saved into the project folder, alongside everything else. It takes a minute and it means the promotional material comes from the same master as the video.