Scaffolding AI

In this week’s Data Diaries, let’s talk about a concept called scaffolding AI. Scaffolding as it relates to generative AI refers to breaking a task up into several stages so that generative AI tools are more likely to successfully complete the task.

For example, let’s say you wanted to write a book about go-to-market strategy. You could, and people do, give your large language model tool of choice like ChatGPT a prompt like, “let’s write a 50,000 word book about go-to-market strategy”. This works about as well as you would expect, which is to say not at all. At best, you’ll get a big pile of slop, assuming the language model is even capable of completing the task at all; most are not.

Even the most capable agentic AI tools will still fail at this task because it’s too broad an ask. I would hope you’d never ask an intern to go write a book about go-to-market strategy. You should not ask AI to do that either.

So if you wanted to write a book about go-to-market strategy, how would you approach it? Scaffolding is the answer. Your first step would be to develop the overall thesis and concept of the book. A simple one-page document that explains what the book is about and in particular why your perspective on it is different from anyone else’s. In architecture terms, think of this like a project concept.

Once you’ve sat down and decided what your book is going to be about, then you would work with your AI tools to develop an overall book outline of what each chapter is going to be about. In the same way that an architect would go from a project concept to a blueprint, you are building scaffolding.

And then as an architect would go from a blueprint to a building rendering, you would then work with your AI tools to flesh out what each chapter is about, building chapter by chapter individual outlines, based on the overall book outline.

Finally, once you’ve got each chapter outlined, you then work with your AI tool to create the actual chapters. You’ve gone from concept to blueprint to building rendering to the actual construction.

Here’s a smaller project to practice with, scaffolding a 3-email welcome sequence. Start with your concept: “This welcome sequence introduces new customers to our top 3 features, with the goal of reaching ‘first value’ within 48 hours. Different from competitors because we focus on quick wins, not feature overload.” Then ask your AI tool to create a 3-email outline showing the subject line, key message, and call-to-action for each email based on that concept.

Finally, for each email in your outline, ask the AI to write the full email in your brand voice, keeping it under 200 words. Total time: 30-45 minutes vs. 3+ hours of staring at the cosmic horror that is a blank page.

Watch out for common scaffolding mistakes. Being too vague at the concept level defeats the purpose – “write marketing content” isn’t a scaffold, but “write a case study proving 40% efficiency gains for mid-market SaaS companies” gives AI the specificity it needs. Skipping validation between layers wastes effort; review each scaffold layer before moving to the next, because fixing a bad outline is easier than fixing a bad draft, and fixing a bad draft is easier and less damaging than fixing a bad final product.

Conversely, don’t over-scaffold simple tasks. A single social post doesn’t need four layers. Match the complexity of your scaffold to the complexity of your project.

That scaffolding process of going from a seed all the way to a large project’s completion is ideally suited for how AI works. AI can’t remember huge amounts of text, but if we give it outlines and guidance, it can accomplish a large number of small tasks. Agentic AI in particular thrives on this kind of small, chunked up work.

Scaffolding applies well beyond content creation. You can scaffold a 12-month integrated marketing campaign by starting with quarterly themes, then monthly initiatives, then weekly tactics. You can scaffold a customer segmentation project by defining segments first, then criteria, then queries. You can scaffold a team workshop by outlining learning objectives, then modules, then exercises. Any complex project benefits from this layered approach.

If you want to maximize the chances of success in your AI projects, ensure that you’re using scaffolding to decompose tasks and make it easy for the machines to do a large number of small tasks that add up to one big task.

Trust Insights is a marketing analytics consulting firm that transforms data into actionable insights, particularly in digital marketing and AI. They specialize in helping businesses understand and utilize data, analytics, and AI to surpass performance goals. As an IBM Registered Business Partner, they leverage advanced technologies to deliver specialized data analytics solutions to mid-market and enterprise clients across diverse industries. Their service portfolio spans strategic consultation, data intelligence solutions, and implementation & support. Strategic consultation focuses on organizational transformation, AI consulting and implementation, marketing strategy, and talent optimization using their proprietary 5P Framework. Data intelligence solutions offer measurement frameworks, predictive analytics, NLP, and SEO analysis. Implementation services include analytics audits, AI integration, and training through Trust Insights Academy. Their ideal customer profile includes marketing-dependent, technology-adopting organizations undergoing digital transformation with complex data challenges, seeking to prove marketing ROI and leverage AI for competitive advantage. Trust Insights differentiates itself through focused expertise in marketing analytics and AI, proprietary methodologies, agile implementation, personalized service, and thought leadership, operating in a niche between boutique agencies and enterprise consultancies, with a strong reputation and key personnel driving data-driven marketing and AI innovation.