Video
in under
a minute.

Five numbers that summarise the whole story — the rest of the page explains how they were achieved.

The “20+” figure counts distinct client videos published within the first two days of the system operating — it excludes additional renders and re-rendered file corrections.

The client's marketing team regularly produces short-form video content for digital channels. The existing process required manual intervention at every stage — scene editing, voice-over recording, audio processing, asset preparation and rendering. None of these tasks was difficult on its own; their repetition created the bottleneck.

I analysed the workflow and designed an AI-enabled production system that automates the repeatable parts of the process while keeping creative decisions and quality control under human supervision. The result is a reusable production pipeline that takes a structured script through to a finished video — and it has been used to produce and publish 20+ real client videos.

The problem was not video generation. It was production efficiency. The team needed to produce short-form content regularly while keeping consistent visual standards, branding, predictable output formats, production quality and a reasonable turnaround time — but the workflow depended on repetitive manual operations.

The question: how much of this workflow could be automated without removing human creative control?

Rather than starting with a particular AI technology, I first broke the production workflow into individual activities — and mapped where automation was genuinely possible.

The objective was not to replace the marketing production process with an autonomous AI system. Instead, the workflow was redesigned around a human-in-the-loop model.

AI removes repetitive execution; humans keep control over creativity, messaging and quality.

A key design decision was to avoid creating another collection of disconnected AI tools — and instead connect them into a single repeatable workflow.

The solution is an AI-orchestrated production pipeline that uses MCP as the integration and orchestration layer between specialised capabilities.

The system receives a prepared script together with information about the type of video being produced. Rather than configuring every video from scratch, the appropriate production skill is selected — a bundle of predefined production rules covering colour palette, typography, animation style, composition, format and visual behaviour. The pipeline then uses those rules to execute the whole production workflow.

One key element of the solution was separating creative and production rules from execution. Instead of manually recreating the same design decisions for every video, production standards are encoded into reusable skills.

MCP provided a structured way to connect specialised capabilities and tools into a single repeatable production workflow. Instead of treating AI models as isolated tools, the pipeline uses them as components of a broader production system.

The solution was not stopped at the proof-of-concept stage. It was tested, refined and used in real production — resulting in 20+ videos produced and published.

The solution was evaluated not only technically, but also through actual operational use.

The pipeline was instrumented and tested using real production runs. For the measured short-scene runs, the full pipeline — from execution start to finished video file — completed in approximately:

This measurement includes the complete pipeline, including TTS and final output generation. Processing time varies with the complexity of the requested production workflow — scenes requiring additional visual generation or more complex processing can take significantly longer. The benchmark represents observed production performance, not a fixed processing guarantee.

The voice-generation component was measured separately. Across two production tests, 23 dialogue segments were generated:

In practical terms: 4 min 08 sec of finished voice-over in 1 min 25 sec. A human recording would require at least the real-time duration of the final audio — 4:08 — before accounting for retakes, editing, audio processing and synchronisation. The AI generation removes the real-time human recording requirement from this part of the workflow.

Important distinction: these are not identical measurement scopes. The manual benchmark covers the editing component only; the automated benchmark covers the complete pipeline. So the numbers should not be presented as a simple “2–3 minutes vs 56 seconds.” The meaningful conclusion is: the automated pipeline performs multiple production stages within approximately one minute for many short-scene runs, rather than automating only the manual editing step.

An important finding: processing time depends on the complexity of the requested output. Workflows requiring additional asset generation or more complex visual processing take considerably longer than simple scenes.

This led to a design principle — AI automation should be evaluated against the complete process, not a single benchmark number. The system therefore measures actual production runs rather than relying solely on theoretical model performance.

Shorter is better — human execution time drops while creative control stays.

The goal was not to eliminate the human from the process. It was to eliminate unnecessary human execution from the process.

AI Transformation · Business Process Analysis · Solution Design · Implementation

• Analysed the existing marketing production workflow
• Identified repetitive production activities
• Assessed automation opportunities
• Designed the target AI-enabled workflow
• Defined the human-in-the-loop model
• Designed reusable production skills
• Implemented the MCP orchestration layer
• Integrated AI-based generation components
• Tested the workflow using real production scenarios
• Measured production performance
• Refined the system based on observed results
• Supported the transition into real production use

• AI transformation starts with the process. The technology follows the process analysis, not the other way around.
• The value is in orchestration. A single AI model can solve an individual task; greater productivity gains come from connecting capabilities into a complete workflow.
• Not everything should be automated. Creative decisions, quality control and final approval remain valuable human activities.
• Reusability is essential. Encoding production knowledge into reusable skills makes automation scalable.
• Measure real workflows. Actual production measurements beat theoretical model benchmarks.

How I would approach a similar AI transformation:

1. 01 — UnderstandMap the current process.
2. 02 — AnalyseIdentify bottlenecks, repetitive activities and human time costs.
3. 03 — Identify opportunitiesDetermine where AI can create meaningful value.
4. 04 — PrioritiseEvaluate options by business impact, feasibility and complexity.
5. 05 — PrototypeBuild a focused proof of concept.
6. 06 — MeasureCompare the new workflow against the existing process.
7. 07 — ImplementTurn successful experiments into reusable operational systems.
8. 08 — ScaleExtend the solution to additional processes and use cases.

And this is where I work best — at the intersection of business needs and technical possibilities.

The “20+” figure counts distinct client videos published within the first two days of the system operating — it does not include additional renders or re-rendered file corrections.

A real marketing workflow transformed from repetitive manual production into a measurable, reusable AI-enabled system.

Have a process, a problem or an idea worth analysing? Write to me and we will turn it into a concrete next step.

Client confidentiality: client identity and commercially sensitive information have been anonymised. This case study presents the methodology, implementation approach and aggregated production metrics without exposing confidential client information.