From prototype to production line: how we helped Road Skills Online standardise their training videos

Road Skills Online (RSO) is an e-learning platform for the transport industry, providing online training to drivers, operators and the businesses that employ them, mostly in the UK. Their courses are split into toolbox talks, with each one consisting of several short videos, known as learning points, and concludes with a multiple-choice assessment. Learners have to watch the videos and pass the assessment before they can progress.

That makes the videos the heart of Road Skills Online’s product. They’re the main way learners take in the material, and the assessment only works if the video before it has explained things clearly. For a learner, the videos are the course.

Each video is built like a slideshow: a series of scenes, each with on-screen text, images and a spoken voiceover. Traditionally, each video was made by hand. Someone put each video together slide by slide, then paid a voice actor to record a narration from a script, which was laid over the top. It worked, but it was costly and very time consuming. With a need to publish at least 6 videos every month, those costs and hours were adding up quickly.

That’s where we came in. What started as a two-day prototyping session to test whether we could automate making these videos has grown into the tool Road Skills Online now relies on to produce them quickly, affordably and consistently.

The real challenge: consistency

Making a video was never the hard part. The hard part was making videos that Road Skills Online’s existing learners wouldn’t notice that the production process had changed. Moving from a human-made video to an AI-generated one had to be close to invisible, and not just once but on every video the tool produced.

Making something predictable out of something as unpredictable as AI was incredibly difficult. A generated voiceover never sounds quite the same twice, and every part of the video has to fit around it. There are many ways for it to go wrong or drift out of sync:

• Fonts that render differently from one slide to the next, or overflow the space they’re given
• Speech speed that changes between voiceovers, so a script reads quickly in one video and slowly in another
• Pacing, where text appears before the narrator reaches it or lingers long after they’ve moved on
• Styling that slowly drifts, so videos in the same course stop looking like they belong together
• Transitions that cut in too early or too late and break the flow between slides

Any one of these on its own is a small annoyance. Across a whole library of courses, they’re the difference between training that feels professional and training that feels thrown together.

Two days to prove the concept

We began earlier this year with a focused two-day prototyping session, the kind we can run with clients to test an idea quickly before committing to a full build. The goal was simple: find out whether the idea was possible at all. By the end of the second day we had a working prototype that covered the whole journey. You typed in the content for each slide, the tool generated a voiceover from the script, added images (either uploaded or generated by AI), and sent everything off to be rendered into a finished video.

It was rough, but it answered the most important question. The concept worked, and it was worth building properly.

Matching the hand-made originals

The gap between that prototype and a video Road Skills Online could put in front of learners was still large, so we closed it step by step. The first target was matching the original hand-made videos as closely as possible: the same line spacing and font sizes, blocks of text appearing one at a time, and text boxes sliding smoothly into frame instead of simply appearing.

The rendering tool we used proved to be a challenge – there were many undocumented features (some of which proved to be vital for us) whilst some of the documented features no longer worked at all or were due to be retired – a common pitfall of rapid development. A great deal of our progress came from trial and error: trying what the documentation said, finding it didn’t work, and experimenting until we found something that did – the exact reason an early investigation phase can be critical to software projects.

The clearest example was the variables feature of the service – which allows us to set a value once and reuse it throughout a video. It turned out to be the only way to tie each slide’s timing to the length of its voiceover, so the words appear on screen as the narrator says them. Yet it wasn’t documented at all, and we only found them through persistence and experimentation. Without it, the timing would have had to be set by hand on every slide, which is exactly the manual work the tool was meant to remove.

We soon had our first version that could reliably produce a video to match the hand-made originals.

Building a tool people want to use

With the videos looking right, our focus moved to the people making them. A tool that produces great results but is frustrating to use doesn’t get used, so we then spent time refining a working engine into a dependable everyday tool.

• A live preview. Before a video is sent for rendering, users can see exactly how each slide will look. We take the same sizes and positions we send to the rendering tool and rebuild the slide on screen as the user edits it, so there’s no guesswork and no waiting for a render to then spot a typo.
• A shared media library. Images are uploaded once and reused across videos. That saves time and keeps the same visuals appearing consistently throughout a course.
• House style built in. Colours, fonts and slide layouts are set up in the tool, so every video ensures brand consistency, without anyone having to remember the rules.
• Easy editing. Slides, and the blocks of content within them, can be reordered by dragging them into place. If something is missing or wrong, the tool says so clearly before the video is rendered.
• Clear statuses. Rendering a video takes time, so every video shows where it is: draft, queued, processing, ready or failed. The platform we developed checks on progress automatically, so users always know whether their video is complete, and if something goes wrong it says so clearly, instead of failing silently or with a confusing error.

Looking back, every feature had the same aim: remove a manual step, or remove a decision someone would otherwise have had to make every time. That’s what standardising the process really meant.

The results

The way Road Skills Online produces videos has changed completely. What used to mean building every slide by hand and booking a voice actor for each script now means entering the content, checking the preview and pressing a button.

• Faster. A video that used to take days to produce can now be completed in well under an hour, and minor changes no longer mean re-recording an entire narration.
• Cheaper. There’s no longer a voice actor to pay for every script read, dramatically reducing the costs involved.
• Consistent. Every video follows the same pacing, styling and structure, so a course feels like one product rather than a collection of separate videos.

What we learned

A prototype proves the idea, not the product. Two days was enough to show the concept worked. To turn it into an effortless tool that Road Skills Online could rely on took a considerable amount of effort. That isn’t a shortcoming of the prototyping session. It’s the nature of building something real, and it’s worth planning for from the start.

The last 20% is what people notice. The prototype could produce a video almost straight away. Getting the timing, styling and transitions right took far longer than building the whole pipeline. Learners would never notice that the pipeline works, but they’d spot a word appearing a second too late straight away.

Consider the tooling you rely on. The documentation for the rendering tool was far from perfect – we managed to work around it, but decisions on dependencies and suppliers should be carefully considered.

Build the rules into the tool. A style guide only works if everyone remembers to follow it. When the colours, fonts and pacing are built into the tool, consistency happens automatically, and that’s what turned a clever experiment into a standardised process.

Road Skills Online didn’t just need to produce a video. They needed a way to do so regularly, in the same way, quickly and affordably, without learners noticing the difference. We got that right – turning a daunting task into an effortless process.