Google Just Made AI Predictions on Your Spreadsheets Possible — Without the Headache

If you run a small business in Malaysia, you’ve probably got data scattered everywhere — customer lists in your CRM, sales numbers in Excel, inventory logs in your accounting software. And somewhere in the back of your mind, you know there are patterns in there that could help you make better decisions. Which customers are about to leave? What stock should you reorder next month? Which marketing channel actually drives sales?

The problem has always been the same: getting those answers usually takes weeks of data cleaning, coding, and tweaking models. You’d need a data scientist — or at least someone who really knows their way around Python and machine learning. That’s changing. Google Research just dropped something called TabFM, and for SME owners who’ve been sitting on their data without knowing what to do with it, this is worth paying attention to.

What Happened

Google’s research team introduced a new foundation model, TabFM, that can make predictions on a table of data it has never seen before — in a single pass. No training. No feature engineering. No months of pipeline building. You feed it historical examples and new rows you want to predict, and it gives you answers instantly (VentureBeat).

Normally, machine learning on tabular data — the kind in your spreadsheets and databases — requires a ton of prep work. You have to clean messy inputs, fill in missing values, turn categories into numbers, and then run endless loops to find the best settings. Once it’s live, you’re stuck monitoring for “data drift” and retraining every few months just to keep it accurate. As Weihao Kong, Research Scientist at Google, put it: these models “incur ongoing operational debt through data drift monitoring and retraining pipelines” (VentureBeat).

TabFM skips all of that by treating tabular prediction as an in-context learning problem. Instead of being trained on your specific dataset, it uses a hybrid architecture that preserves the two-dimensional structure of a table — rows and columns stay intact — so it doesn’t lose the relationships between your data points. It was pretrained on hundreds of millions of synthetic datasets using structural causal models, which means it learned the fundamental math of how table features interact without ever seeing real customer data (VentureBeat).

Why This Matters for Your Business

For a Malaysian SME with maybe 10 or 20 employees, the biggest barrier to using AI has always been complexity. You know the data is valuable, but you don’t have a team to build and maintain custom models. TabFM flips that. It gives you the ability to get a high-quality baseline prediction from your data in minutes — not weeks.

Imagine these scenarios:

• Customer churn prediction. You have a list of customers, their purchase history, and how long they’ve been inactive. Instead of building a bespoke pipeline, you feed the historical data and a few active sessions into TabFM, and it returns a churn probability for each customer. You can then act on it — send a promo or a personal follow-up — before you lose them.
• Inventory forecasting. Your sales data for the past year, plus some seasonal factors, can be used to predict which products will run out next month. No data science degree required.
• Lead scoring. You’ve got a list of leads with info like company size, industry, and website visits. TabFM can tell you which ones are most likely to convert, so your small sales team focuses on the right people.

The key here is speed. Google’s researchers tested TabFM on 51 different datasets — 38 classification and 13 regression tasks — and its zero-shot predictions matched or beat heavily tuned supervised models (VentureBeat). But the real value isn’t about beating every hyper-optimised production model. It’s about getting a working answer now, without hiring a dedicated data science team or waiting for a complex pipeline to be built.

“The true practical business value it unlocks for lean engineering teams is velocity,” Kong said. “It allows data analysts and backend engineers to instantly spin up high-quality baseline models without a dedicated data science team managing a complex lifecycle.” (VentureBeat)

For you as a business owner, that velocity means you can test ideas faster. Want to see if your customer data can predict next month’s revenue? You can try it this afternoon. That kind of quick feedback loop is something most SMEs have never had access to.

The Bigger Picture

TabFM is part of a larger shift that feels like it’s been a long time coming. The rest of the AI world — text, images, voice — has already moved to zero-shot inference, where you can just prompt a model and get an answer without training it first. Tabular data, despite being where most business data lives, has been stuck in the old way of doing things. This changes that.

What this trend means long-term is that predictive analytics is becoming a commodity. You won’t need to be a tech company or have a big IT budget to get useful forecasts from your data. The model does the heavy lifting of understanding how your columns and rows relate to each other, and you just bring the questions.

It also suggests that the role of “data science” for small businesses is shifting. Instead of spending months building and maintaining custom models, you (or a generalist on your team) can focus on interpreting results and making decisions. The grunt work gets handled by a foundation model that’s already been trained on the universal patterns of tabular data.

Now, it’s still early days. Google itself notes that TabFM isn’t going to replace every hyper-optimised production model on every enterprise workload (VentureBeat). But for the kind of data challenges most SMEs face — medium-sized tables, common prediction tasks, a need for speed over perfection — this is probably the most practical step forward in business AI this year.

If you’ve been collecting data but not doing anything with it because it felt too hard or too expensive, the window is opening. Tools like this mean you don’t need a PhD or a big budget. You just need to start.

Want to see how zero-shot predictions could work with your actual business data? Book a free 15-min call to see how this applies to your business →