Everyone's watching the AI. The data is doing the work.
The world made roughly 180 zettabytes of data last year. Almost none of it is in a state your AI could actually use.
The part nobody puts on the slide
Every AI announcement of the last two years has been about the model. Bigger models, faster models, cheaper models, models that reason, models that see. It's the part that demos well.
The part that doesn't demo well — and therefore doesn't get talked about — is the data underneath it. And that's the part that decides whether any of this works for your business.
Here's the uncomfortable version: the model is mostly a commodity now. Anyone can rent a capable one by the hour. What nobody can rent is your data — the record of how your specific operation actually runs. That's the fuel. The model is just the engine, and engines are for sale everywhere.
The numbers that matter
The world generated something in the order of 180 zettabytes of data in 2025, according to IDC's long-running projections. The figure roughly doubles every few years. It sounds like an abundance problem — surely with that much data, feeding an AI is the easy part.
It isn't. The volume is the trap.
Informatica's 2025 survey of data leaders found that only about 12% of organisations report having data of sufficient quality and accessibility for AI applications. Deloitte and others consistently put data quality at the top of the list of AI adoption barriers — around 60% of enterprises name it as the number one obstacle, ahead of talent, cost, and compute.
Read that again: the thing stopping most AI projects is not the model, the budget, or the skills. It's that the data going in is messy, scattered, or locked in formats a system can't read. The industry has a blunt phrase for what happens next — garbage in, garbage out. An expensive model fed bad data just produces confident, expensive nonsense.
Why this is the whole game
There's a reason the biggest AI companies guard their data more closely than their code. The model architecture is largely public knowledge. The training data is the moat.
The same logic scales all the way down to a four-guide tour operator. Two businesses can buy access to the exact same AI model. The one that wins is the one whose data is clean, connected, and structured — because that's the business the model can actually be useful to.
Your booking history, your manifests, your guest communications, your maintenance logs, your rostering patterns. Individually they look like admin. Together they're a proprietary record of how your operation works that no competitor has and no vendor can sell you. That's the asset. Most operators are sitting on it without realising it's worth anything.
What it means if you run an operation
The takeaway isn't "go adopt AI." It's the opposite of the hype, actually.
Before AI is worth anything to your business, your operational data has to be in a state a system can use — documented, consistent, and connected rather than trapped in one person's head and six unlinked spreadsheets. The businesses that get real value from AI over the next few years won't be the ones who move fastest on the model. They'll be the ones who did the unglamorous work on the data first.
That's not a reason to wait. It's a reason to start at the right end. The highest-value first move for most operators isn't a chatbot — it's getting the weekly workflows out of spreadsheets and into something structured. Do that, and you've built the foundation AI actually runs on. Skip it, and you've bought an engine with no fuel.
The AI era is real. But it was never really about the AI. It's about whether the data underneath it is worth feeding in.
blnk.nz · July 2026