From Low-Code to Production-Grade AI
How I built a content generation system that created 1:1 messaging at scale
In my last post, I made a case for why Operations skills are a superpower for the AI era. I argued that our ability to think in systems, map processes, and understand data makes us the ideal “last-mile” builders, especially with our experience using low-code and marketing automation tools.
That post was the “what”. Today’s post is the “how”: How I, a MOPs leader with limited coding experience, used Python with the help of AI, to build a 1:1, research-powered ad content engine for personalized messaging at scale.
The result? Significantly higher engagement and click-through rates across our target accounts, while transforming our campaign creation process from weeks of manual work to systematic, scalable execution within days.
But it’s not just about the results. It’s about how this project helped me understand that AI democratizes development, proving that learning to code may not be the barrier it used to be.
Personalization at Scale is an Operational Nightmare
Most of us have experienced the personalization at scale challenge. We have a target account list, key personas, and the goal of launching a personalized campaign. But when you’re facing hundreds of accounts, multiple personas, and various ad formats, the math becomes exponential. True 1:1 personalization can be an operational nightmare to build and implement.
The Math: (Accounts) x (Personas) x (Ad Formats) = Thousands of unique ad variations
So we tend to fall back to templates, and create generic “one-size-fits-all” content. But with AI and all the new skills I’ve learned, I wanted to see if I could analyze account data at scale and create messaging that spoke to each account and persona’s specific challenges.
Why Low-Code Wasn’t Enough
My first instinct was to use the tools I knew: a low-code automation platform. I figured I could build a workflow that would:
Read a spreadsheet of accounts and personas
Call an LLM API with a prompt to perform research and create messaging
Write the results back to the spreadsheet
But as I started to map this out, it was obvious that this wasn’t going to work.
First, the level of scale and limits. To run this for a couple hundred accounts and 4 personas, each requiring multiple steps, I’d be looking at thousands of workflow tasks and would immediately hit my platform limits—likely out of the question.
Second, the quality control problem. Even if it could run, I’d have validation issues. Low-code tools are great at calling an API and chaining steps together for an end-to-end workflow. But they are not so great at quality validation at scale: checking character limits for different ad sizes, validating CTAs, or enforcing brand messaging.
Asking AI for a Better Way
Facing these limitations, I turned to my AI assistant. I described my problem: I need to process hundreds of rows from an excel file, run a complex set of generation and validation steps for each, and do it all in one batch without hitting limits. What’s the right way to do this?
The answer it gave me pointed to a more powerful, professional-grade stack: Python, Jupyter Notebooks, and Azure Machine Learning Studio.
For a non-coder, these words might as well be in a different language, and sound more suited to an Engineer or Data Scientist, not exactly Marketing Operations. But AI helped me understand them in simple, practical terms:
Azure Machine Learning Studio: The secure “sandbox” or “workbench.” It’s the place that I can build solutions using approved tools in Azure. This is a “platform” that IT can provide access to.
Jupyter Notebooks: This is the tool inside that workbench. It’s like an interactive lab notebook. You can write a small piece of code in a “cell,” run it immediately, and see the results right on your screen. It’s perfect for testing and building step-by-step.
Python: This is the “language” you use to write instructions in the notebook. It’s the instructions for data handling, logic, and stitching all the pieces together.
None of this felt like coding in the traditional sense, but more like building workflows with much more powerful tools. I would recommend partnering with your IT team and ask if they have an approved sandbox or area where you can experiment with AI and tools.
For this use case, I used AI as my coding tutor to help me build the data analysis and content framework using Python and ran them in my notebook.
First was the setup: how it could read my Excel file and connect to my Azure OpenAI model.
Next, it gave me the blocks of code for the content creation “engine”. This kicked off days of iterative, back and forth testing.
My first attempts were a reality check. I had multiple failures with incomplete headlines, missing CTAs, copy that cut off mid-sentence. It was frustrating at first, but taught me exactly what validations to build in. I wasn’t just “prompting”, I was building a system to:
Add validation to my data
Review and recommend CTAs
Abide by branding guidelines
Check if headlines and copy were within character limits
Loop the engine for hundreds of accounts
This back and forth is how I learned more about Python, Azure, and a more robust way to handle AI production at scale.
An Intelligence-Driven Content Engine
My final workflow wasn’t a simple “generate content and done” process. It was an intelligence-driven system that incorporates reliability, quality, and scale (and only took about 1.5 weeks to build):
Research: Pull account intelligence and pain points.
Analyze: Use AI to extract key insights from account data, including strategic priorities and persona-specific challenges.
Create: Generate tailored content based on the analysis.
Validate: Run the content through quality control: character limits, persona keywords, tone, and brand.
QA: Anything that needed QA was flagged for further review.
Review: Once the scripts ran (in minutes), it exported a complete Excel file.
Human-in-the-Loop: The marketing team reviewed the file and made final refinements, ensuring a final quality-control step before launch.
The combined power of AI, the logical rigor of Python, and the human-in-the-loop quality control enabled us to personalize at scale, all while building a reusable “engine” for future campaigns.
The results validated our approach: significantly improved engagement and response rates compared to previous campaigns.
We were able to deliver personalized ads at scale and speak to our buyers about their specific problems, and they were responding accordingly.
The “Beyond Low-Code” Framework
This project taught me a repeatable framework for building quality AI solutions at scale:
Start with AI for intelligence & creative solutions
Analyze accounts and challenges
Map to personas
Identify and generate relevant messaging
Use Python for quality and validation
Enforce character limits
Check copy alignment
Detect tone or brand issues
Use cloud infrastructure for scale
Jupyter notebooks for development
Azure for AI generation
Export to excel for deployment
Test with real campaigns
Test against other campaigns
Measure engagement and performance
Iterate based on results
Get to measurable improvement before scaling
This process is also a great answer to the “context dilution” problem I wrote about in an earlier post. Instead of a “telephone game” of “agents” where rules and data were lost with each handoff, the AI has all the rules, all the data, and all at once. It’s a true, high-context Synthesis Task.
Your First “Beyond Low-Code” Project
This content system is more than just a one-off tool. It’s a repeatable engine for taking structured data, applying AI-driven insights, and enforcing Ops-level quality control.
And AI didn’t just teach me to use Python, it also taught me that the barrier between “Ops” and “Developer” is a willingness to ask the right questions and dig a bit deeper to learn and build.
This is the new “how” for Operations. We start with the business problem, identify where current tools hit their limits, and then use AI as our tutor to build a scalable, enterprise-grade solution.
And this is just beginning. Think about what else you could do with this framework:
For Sales: What if you used this system to analyze prospect companies and create research-backed, personalized outreach that references their specific business challenges?
For Content: What if you connected it to a Google News search? You could create messaging that references a prospect’s latest strategic initiatives or market challenges.
For Your Team: What’s the time-consuming manual process that’s preventing your team from scaling? Is it campaign setup? Naming conventions? Perhaps that’s your first use case.
This is the democratization of development I’m talking about. It’s not a passive “it’s happening.” It’s an active invitation for every Ops pro to move from “workflow builder” to “solution builder.”
What manual processes could you start with? I’d love to hear about your operational challenges and brainstorm solutions!