I Built an ML Model with AI — No Human Code Required
Last week, I set myself a challenge: build a machine learning (ML) model entirely with AI, no human coding allowed.
Just me, some creative prompts, and seeing how far AI could take the wheel.
The result? A big win. For the first time at our company, we’ve got a retention ML model built entirely by AI. It wrote all the SQL and R code, transformed the data, and had a working model up and running in under an hour.
Not a single line of code came from me — which felt both weird and thrilling. Though more on the thrilling side of the house.
Where AI Excelled. And Where it Didn’t.
The AI handled the SQL and R coding impressively well, delivering functional and accurate scripts in record time. It was able to construct complex queries, transform tables, and generate the final model outputs.
The catch: I still had to play guide.
Knowing both SQL and R meant I could steer the AI — spotting mistakes, refining prompts, and reprompting where needed.
Without that knowledge, it would’ve been like trying to assemble flat-pack furniture without the instructions. You’ll eventually get there, but it’ll take longer (and involve some swearing or broken drawers you’ll relabel as ‘art’).
Then there was the early data prep. While AI could’ve handled it, our data has its quirks. Jumping in to clean and structure the tables myself was faster than training the AI to understand all those nuances.
Once the prep work was done, though, AI hit the ground running — transforming tables, building the model, and making me feel a little redundant (in a good way).
The Overall Process.
I start out with some initial data cleaning.
The AI took the semi-cleaned data and transformed it into useable base tables with some SQL magic.
AI then built, trained and validated a working ML Retention model using R.
Once I was happy with it, I prompted the AI to also set up an automated model re-run monthly and add new model insights to an outputs data table.
This outputs tables now powers a new Retention dashboard that I created in Looker Studio for business folk.
The Takeaway.
AI is turbocharging the way we build models.
For experienced analysts and data scientists, it’s a game-changer.
But let’s be clear: AI isn’t a magic button. You still need to know how machine learning and coding languages (like SQL, R, or Python) work to get the best results.
Without that expertise, you’re just throwing darts at a tiny board in the dark, hoping for a bullseye.
But if you know how to guide it, AI becomes less of a tool and more of a collaborator — one that works at lightning speed.
Looking ahead, this blend of human expertise and AI power feels like a massive leap forward for analytics.
AI might be taking over the heavy lifting, but human judgment and context still steer the ship. And honestly, I’m excited to see where we can go from here — maybe with a little less data cleaning next time.
