AI Is Running Out of Data. Australian Small Businesses Never Had Any.
The AI world is panicking about running out of training data. Researchers at Stanford called it “peak data” earlier this year, the moment when the useful, high-quality text available on the public internet starts to run dry. Big labs have consumed most of it. The model improvement that came from simply throwing more data at a problem is slowing down.
That’s the crisis getting all the coverage. Here’s the one that isn’t.
Australian small businesses hit their data ceiling before the AI boom started. And most don’t know it yet.
What the “Peak Data” Story Actually Says
The Stanford 2026 AI Index Report put it plainly: the available pool of high-quality human-written text for training large models has been exhausted. Synthetic data, AI generating training data from AI, is being explored as a workaround, but the research is not encouraging. Adding synthetic data doesn’t reliably improve performance, and there’s no evidence it can replace real-world text in pre-training.
Labs are in trouble because they built their business model on scale. More data, bigger model, better results. That flywheel is grinding to a halt.
Forbes reported this week that 63% of organisations still lack the data practices AI actually requires. That number should land differently when you run a small business in Australia and you’ve been wondering why your AI tools feel generic.
The SMB Version of This Problem Is Structural
The labs ran out of public data. Small businesses never had private data to begin with. Different problem, same outcome.
Think about what “data” actually looks like inside a typical small business. There’s a spreadsheet tracking customers, last updated eight months ago, missing half the fields. A CRM with gaps where the sales notes should be, because the rep who knew the context left eighteen months ago. Customer service conversations sitting in someone’s email inbox, in their head, or in a WhatsApp thread nobody exported. Pricing logic that exists as institutional knowledge: “we give this client a 15% discount because of what happened in 2023.”
None of that is structured. None of it is queryable. None of it is something an AI tool can learn from.
And that’s the problem. AI tools, the ones you’re actually using right now, Claude, ChatGPT, whatever’s in your CRM or your support platform, are only as useful as the context you can give them. They don’t know your customers. They don’t know your exceptions. They don’t know that you always waive the setup fee for referrals or that the onboarding process for retail clients takes three weeks, not one.
Generic context in, generic output out.
Why This Matters Right Now
The AI tools themselves are not the bottleneck anymore. They’re good enough. The bottleneck is business-specific context, and the businesses that are building it now, structuring their operational data, documenting their processes, cleaning their customer records, are laying down something that compounds.
In two or three years, the gap between businesses with clean internal data and those without it will be stark. The first group will have AI that actually knows how their business works. The second group will still be writing the same generic prompts and getting the same generic answers.
This is the moat that’s actually available to a small business right now. You can’t train a foundation model. You can build a knowledge base.
Three Things Worth Starting This Week
Document your top 20 customer FAQs. Not a marketing FAQ for the website, an internal one. The real questions your team answers repeatedly. How you handle returns. What your turnaround time actually is versus what it says on the site. What customers misunderstand about your pricing. Write it down in plain language and put it somewhere your team, and your AI tools, can access.
Export and clean your CRM, even if it hurts. A messy CRM export is still better than no data at all. Get it into a spreadsheet. Standardise the fields. Fill in what you know about your best 50 customers: industry, what they bought, why they came to you, what problem they were solving. That data will serve you for years.
Write down your pricing logic and exceptions. Every business has rules that aren’t in the pricing page. Write them down. When you give a discount, why? When do you say no? What does a good client look like versus one that costs you money to serve? This is the kind of context that makes AI tools genuinely useful for quoting, responding to leads, or training new staff.
The Window Is Open, Not For Long
Right now, the advantage of building data infrastructure is underappreciated because AI tools work well enough without it for basic tasks. That won’t be true in 2027. As the tools get more capable and more businesses use them, the differentiation will shift entirely to context quality.
The labs’ data problem is interesting. Your data problem is urgent.
Start building now.
Related: Parliament Is Investigating Why Small Businesses Aren’t Using AI. We’ve Been Tracking the Barriers for 18 Months — Here’s What They’ll Find. | How Australian Agribusinesses Use AI for Booking, Scheduling and Client Communication
📬 Subscribe to our weekly newsletter | 💼 Free resources & AI prompt packs
📖 More deep dives: News Deep Dives
📊 Compare AI tools side by side | 💼 Free resources & AI prompt packs
More hot takes: Hot Takes →
If this interests you
- Australia Leads the World in AI at Work — Here’s What the Numbers Actually Show
- Parliament Is Investigating Why Small Businesses Aren’t Using AI. We’ve Been Tracking the Barriers for 18 Months — Here’s What They’ll Find.
- Fair Work Claims Are Up 40%. AI Is Why. Here Is What Australian Small Businesses Need to Do.
More hot takes: Hot Takes →
