AI in Marketing: How we built an Effective Agency System around LLMs
Authored by: Mayur Nathani
“Only the use of AI tools didn’t move the needle.”
When I started my agency, I fell into the same trap as everyone else: I thought AI was about the tools. I assumed that having access to the latest LLM would automatically translate into better results for my clients. I was wrong. I spent months on the surface layer, wasting time and getting inaccurate results because I didn’t understand the operational reality of generative AI.
The breakthrough didn’t come from a new tool. It came when I realized that AI in marketing isn’t a magic button; it’s a system that requires a disciplined architecture.
The Complexity Trap:
Most agencies underperform on AI because they treat it as an execution shortcut. Most of them use generic prompts for complex tasks and wonder why the output feels “robotic” or flat.
Through hundreds of A/B tests, I realized that different sub-areas of marketing require different “LLMs.” There is no single best LLM; there is only the best LLM for a specific use case:
- ChatGPT for content.
- Claude for copy.
- Claude + MCP + Skills for lead generation.
- Gemini for content ideas.
The Node Structure:
The question I had to answer was:
How do we make this a repeatable system in an industry that evolves every week?
The solution was creating a Node Structure. Instead of one long, fragile prompt, we broke our workflows into nodes. Nodes are simply steps that connect one dot to another, creating a broader concept as you progress.
For example, we helped a client by scraping comments to understand target audience sentiment and then planned relevant posts for the following two months, resulting in 4,000 followers in 90 days.
This system allowed me, as a LinkedIn Top Voice, to generate 20 high-value posts in minutes. But the system isn’t just the AI; it is the human-in-the-loop. I still perform the final edits to ensure the Mega Prompt output delivers genuine value. This combination attracts the inbound leads that generic AI content misses.
The LinkedIn transformation for Personal Branding:
Before I built my system for LinkedIn Personal Branding, most of my time used to go towards coming up with ideas which can be focused on adding value to my community. Again brainstorming the right ideas plus content creation used to take so much time that I was unable to provide the time for commenting and DM. So basically I was spending more time on LinkedIn even when I hit my lowest just to build my authority and thought leadership.
Now because the process is entirely proactive and system-driven. With the system in place, it started to help me with the content ideas, content creation and how to comment effectively without using chrome extension. The final content used to be published only once the human intervention is done. Apart from this, the LinkedIn Lead Generation system is in place with the reply rates boosted from 31% to 47%. Both of the systems are created for LinkedIn and can be repeatable which saves hours of manual labor every week.
5 Steps to implement an AI workflow in your Agency:
If you want to move beyond the surface layer and build a production-ready AI stack, follow this framework:
1. Expertise in your domain.
2. Understanding where the gap is.
3. Leverage the power of Advanced Prompting.
4. A/B test till you get better outputs.
5. You must be clear with your outputs.
The Shift from Tool to Asset
By understanding the gaps, we ensured that AI helped us bridge them while saving us both time and money. When your system is built on A/B testing rather than trends, your agency doesn’t just keep up with AI; it evolves alongside it.
Author Bio:
Mayur Nathani is the founder of Growfluence, a marketing agency specializing in Personal Branding, SEO, and Lead Generation. Recognized as a LinkedIn Top Voice for 2024 and 2025, he is the co-author of Crossroads and The Unsaid Emotions and has hosted seven offline networking events in Mumbai.
