This one’s about AI, but it was written by a human. (Probably.)

How can we produce high-quality video for ads quickly and cheaply with AI?

Systematiq.ai founder Rory Flynn has cracked the code, so to speak, on producing video assets at scale with AI workflows. In less than a year, he went from barely able to keep up with the needs of 90 clients to effectively serving more than 500.

Rory shared the shift that made this possible, along with the processes he put in place, at a recent Affiliate World event, and guess what? We were there, and we saw the whole thing.

Turning Data Into Creative 🖼️

The robust and systematized AI tools that Rory has developed with SuperSide, an AI-forward creative agency, make it possible to go from a creative brief to a first-draft video in seconds, not hours or days. They have created and integrated multiple AI agents in ways that make it easier to keep up with both the quality and volume that top brands expect.

Rory’s team and SuperSide are turning out video assets for ads at scale with a four-step process.
Ad Design System

1. Knowledge Graphs 📈

Knowledge graphs, which provide context and a structure for AI tools to understand real-world facts and their relationships, are used to train the AI agents that will execute the video tasks and provide quality control.
Knowledge Graphs

For this system to work, each client will need a branded Knowledge Graph to maintain consistency and brand standards.

2.️ Agents to Execute 🚀

AI agents can execute many of the video production tasks, but the key to getting usable results is to break down every element that you want from your video in granular detail.

This includes lens type, lighting, composition, and other modifiers that professional videographers would incorporate in real-life production.
AI Building Blocks

The best way to do this is by reverse-engineering a successful output. If you’re creating a static image ad, for example, you can break that down into component parts like what the image should look like, what the copy should be based on the language and market you’re targeting, and so on.

3.️Integrations 🔀

System integrations make it possible for each trained AI agent to work together. Again, by breaking down tasks into mundane or repeatable events that AI can handle based on your prompts, you can implement a level of quality control that gets first drafts to a usable status that can then be improved by humans.

Integrating AI agents to perform each of these tasks, in sequence, based on Knowledge Graphs can drastically reduce creative output time without sacrificing quality. One agent can create images, one agent can create copy, and yet another can assemble the elements into a template. Rory recommends a tool like LangChain to assist with agent guidance and integration.

4. Coding 🧑‍💻

Coding is the final piece of the process that ensures all of the AI agents can actually communicate.

The connector language that Rory uses to get these agents to talk to each other is called JavaScript Object Notation (JSON), which is a standard text-based format for representing structured data. And if you’re not familiar with writing JavaScript code, don’t worry — ChatGPT can generate the code you need to connect the agents with a few simple prompts.

Outputs at Scale 🔥

We mentioned earlier that Rory and his team were able to use these processes to scale their client list from 90 to over 500, but other success metrics behind this process are equally impressive.

For example, after more than 1,000 completed projects over eight months, they discovered that Brief-to-Delivery time for clients was 200% faster. Project hours were also reduced by more than half, and the revenue saved totaled more than $1.4 million.

As AI tools shift from marketplace novelties to workflow essentials, mastering these tools and using them in conjunction to compound outputs will become the norm. And the faster you can adapt to this new normal, the more successful your creatives will be.