Why Your Early-Stage Startup Doesn’t Need an AI Engineer (Yet) but a Data MacGyver?

While everyone is talking about Agentic AI, RAG, LLMs and sharing prompts to make your work or life productive, I would take a moment to talk about Angus MacGyver. Well he is definitely not the father of any technology, but the protagonist of the TV series MacGyver. MacGyver is shown to possess a genius-level intellect, proficiency in multiple languages, superb engineering skills, excellent knowledge of applied physics and a preference for non-lethal resolutions to conflicts. In addition to his scientific knowledge and inventive use of common items, he always carries a Swiss Army knife and refuses to carry a gun.

Why am I talking about him, because he had a creative way of solving problems using any available random objects. Drop your keys down an elevator shaft? You may have to MacGyver them out using a shoelace, gum, and a magnet!

I want to take a moment of silence and remind ourselves not to confuse our creativity and imagination for the noise of tools available and in FOMO of being the first in the AI bandwagon. We need to MacGyver our way through problems and use what we often have readily available.

Just as MacGyver refused to carry a gun, a startup data person should often refuse the ‘heavy artillery’ of LLMs and complex neural networks when a simple Python script and a CSV will do the job better.

Startups need a Data MacGyver before an AI engineer

Here are some of the lessons I learnt from the Cardano startup I worked in..

Essentially Recalibrating my settings from being a Data consumer to a Data creator.

When I moved into the early-stage startup world, I realised my most valuable skill wasn’t my ability to build models — it was my ability to lower my expectations and roll up my sleeves.

Coming from a background of working with massive, pre-cleaned sensor data at large companies, I arrived with a “plug-and-play” mindset.

In big organisations, data is like electricity; you flip a switch, and it’s there — clean, formatted, and ready for analysis.

Well not always clean, but somewhat better than what it is at an early stage startup. At a startup, you aren’t flipping a switch. You’re digging the well.

My first “settings recalibration” was realising that if I didn’t find the data, it didn’t exist. I had to shift from being a “Data Consumer” to a “Data Scraper.” In the blockchain space, uncertainty is the only constant. I went from having millions of automated data points to manually copying blockchain entries into CSVs just to understand our first 10 users or stakeholders.

From messy CSV’s to clean data to infographics in spreadsheets for quick data analysis

It felt like a step backward, but it was actually the ultimate lesson in Data Awareness. When you manually copy a row, you notice the “ghost” in the machine — the weird formatting, the missing timestamps, and the subtle biases that automated systems often hide.

Expectations vs. Reality: The “Cleaning” Phase

I had to recalibrate my definition of “work.”

• In a Big Company: Work is 10% cleaning, 90% modeling.
• In a Startup: Work is 99% “carving” data into a shape that statistics can actually understand.

I spent weeks correcting date formats, deduplicating user IDs, and bridging the gap between our marketing funnel and our on-chain reality.

I learned that in a money-strapped ecosystem, clean data is more valuable than complex code.

Moving the Needle: The Achievement of the Pipe

The true “Aha!” moment wasn’t when I ran a regression; it was when I stopped the manual toil. When we moved the team from Python-generated JSON files to **Blockfrost APIs, funneling that data into [BigQuery ](https://cloud.google.com/bigquery)and finally into a self-service Looker** dashboard.

Seeing the Customer Support team — who used to ping a developer for a simple ticket number — suddenly able to download their own reports to fix the issue was a bigger win than any “AI project” could have been. Or being able to derive insights from the wallet contents of our users compared to having no knowledge about them, was a big win. We moved from “not knowing our users” to having a functional, automated pipeline.

The Advice: Don’t Jump the Bandwagon

To the founders and early-stage hires: Don’t jump on the AI bandwagon yet. If you have 30 to 40 users, you don’t need a Large Language Model to tell you what’s wrong. You need a clean table and a sharp eye.

1. Hire the “Bridge” Talent: People who can connect the marketing funnel to the technical pipeline.
2. Invest in the Link Early: Setting up Google Analytics and API streams early is a lifesaver. Missing your early customer journeys because “we’ll fix the data later” is a mistake you can never retroactively correct.
3. Respect the Scale: Use spreadsheets and foundational stats until you hit your “Magic 1,000” samples. Until then, focus on Automated Intelligence — making sure the data moves from A to B without you having to touch it.

This was about the mindset shift. In my next post, I’ll be diving into the technical specifics: What every startup founder needs to know about their data architecture from Day 1 (before you even think about AI).