---
title: GenAI owns the conversation. Tabular AI owns the P&L.
description: Yann discusses the multi-billion enterprise opportunity of Tabular AI.
image: https://blog.probabl.ai/hubfs/TL-YL-GitHub.jpg
---

[Skip to content](https://blog.probabl.ai/tabular-ai-enterprise-opportunity#main-content)

[![](https://blog.probabl.ai/hubfs/logo-probable-black.svg)](https://probabl.ai/)

- [Company](https://probabl.ai/)
- [Join Us](https://join.probabl.ai/)
- [Blog](https://blog.probabl.ai/)

Open main navigation

Close main navigation

- [Company](https://probabl.ai/)
- [Join Us](https://join.probabl.ai/)
- [Blog](https://blog.probabl.ai/)
- [Contact us](https://www.probabl.ai/contact/)

[Contact us](https://www.probabl.ai/contact/)

 Wednesday, September 30 2026

# GenAI owns the conversation. Tabular AI owns the P&L.

![Picture of Yann Lechelle](https://blog.probabl.ai/hs-fs/hubfs/2017-06%20-%20Dark%20BG%20-%20Yann%20Lechelle%20-%20Square.jpg?width=50&name=2017-06%20-%20Dark%20BG%20-%20Yann%20Lechelle%20-%20Square.jpg) [Yann Lechelle](https://blog.probabl.ai/author/yann-lechelle)

Share: [linkedin-in icon](http://www.linkedin.com/shareArticle?mini=true&url=https://blog.probabl.ai/tabular-ai-enterprise-opportunity) [Twitter icon](https://twitter.com/intent/tweet?url=https://blog.probabl.ai/tabular-ai-enterprise-opportunity) [envelope icon](mailto:?body=https://blog.probabl.ai/tabular-ai-enterprise-opportunity)

GenAI owns the conversation. Tabular AI owns the P&L.

9:57

 

A quick test. Think of the last few decisions an algorithm made about you: the post you're reading right now, ranked into your feed. Your credit limit. Your insurance premium. The price of your last Uber. Whether your card payment was flagged. These decisions weren’t made by LLMs. They were made by AI systems that run on data that lives in tables.

Since the launch of ChatGPT in late 2022, we have been mesmerized by LLMs. Suddenly we had chatbots at our fingertips that could converse with us and generate content on demand. Prose. Code. Images. Poetry. Even life advice. LLMs quickly became a household name and redefined what AI means for most people. Boardrooms were no exception. Dazzled by the demos, investors and business leaders extrapolated to massive productivity gains.

However, the leap into LLMs owes more to wishful thinking than to evidence. It’s also an appreciation error. The AI that generates is the AI we can watch while the AI that decides is the AI that pays. The AI that decides has a name: Tabular AI. It’s the largest and least discussed category of enterprise AI, and the one best placed to move the P&L.

# **A definition first. What is Tabular AI?**

When you hear "Tabular AI", you might think of tabular foundation models (TFMs), the fast-growing category of foundation models that can make highly accurate predictions on tables straight out of the box. You may already be familiar with TabICL by Inria-SODA, TabPFN by Prior Labs, NEXUS by Fundamental, or others that can be found on the fast-growing TabArena Leaderboard \[1\].

TFMs are certainly an exciting branch of Tabular AI (which, for transparency, we're involved in too, as the company supporting TabICL, the SOTA fully open source TFM \[2\]), but in fact Tabular AI is much bigger than TFMs. Consider, for instance, the program of the "AI for Tabular Data" workshop at EurIPS 2025, the European satellite of the prestigious NeurIPS conference. It mapped out numerous areas of research and innovation in Tabular AI, including methods (e.g. TFMs), benchmarks, applications, systems, and interdisciplinary perspectives \[3\].

> **Tabular AI is the subset of AI concerned with grounding decisions on tabular data; that is, any table-like structured data organized in rows and columns, including spreadsheets, relational databases, and time-series data. Tabular AI encompasses diverse goals, including statistical and predictive modeling (regression, classification, forecasting), generative modeling, understanding tables, and causal reasoning. Tabular AI tools include ML frameworks, TFMs, language models, and agentic frameworks.**

Evidently, Tabular AI is a category that goes far beyond just TFMs. Reducing Tabular AI to the latest foundation model would be like reducing software to one programming language.

# **Tabular data is the data of enterprises**

In plain English, tabular data is structured data organized in rows and columns. Each row represents an observation, such as a customer or a transaction, and each column represents an attribute of that observation, such as an age or an amount, as defined by a schema.

Tabular data is the lingua franca of enterprise operations and decision-making \[4\]. It is the information enterprises create and accumulate as they operate: transactions, customers, products, assets, risks, you name it. It has been at the heart of enterprise information systems since the first ledgers. It lives in the CRMs and ERPs that businesses run on. And for as long as businesses need to count, compare, and decide, it will remain the ground truth on which every model, agent, and decision is built.

Tabular data is also one of the most valuable and most proprietary assets an enterprise owns: data that no public foundation model has been trained on and no competitor can copy. The impact of tabular data is visible in every industry: it powers credit decisions, fraud detection, predictive maintenance, pricing changes, churn prediction, and operations models \[5-6\].

> **Tabular data is where enterprises record what they do, and Tabular AI is how they decide what to do next. AI that can reason over it natively is not a niche. Unlike many GenAI use cases, it goes straight to the core of the business. For instance, a better forecast translates directly into P&L gains \[7\].**

The clearest sign of this came from SAP's CTO Philipp Herzig at this year's WELT AI Summit in Berlin. Philipp said the greatest untapped opportunity in enterprise AI isn’t LLMs but systems that can make sense of the tables enterprises run on. "We're sitting on at least 400,000 tables, right, in the so-called ERP--finance, supply chain, logistics, procurement," he said. "When it comes to decision intelligence in companies, what are business leaders asking? They're saying, 'Okay, what's the impact of this initiative that I'm doing on my cash flow, on my working capital? How can I get my inventory costs down?' That's where the money is" \[8\].

# **"But 90% of our enterprise data is unstructured"**

It's true that IDC estimates 80 to 90% of enterprise data is unstructured: documents, emails, images, recordings. But that measures how much data companies store, not which data drives their decisions. About 60% of technology spend already goes to the structured minority, because that's where the operational decisions are made \[9\].

Consider how often each kind of data gets used. A contract is read a handful of times. A transaction table is scored millions of times a day. Unstructured data dominates storage, structured data dominates decisions.

# **The size of the Tabular AI opportunity**

No one has published a clean census of enterprise AI by data type, so let's triangulate:

**Value at stake.** McKinsey's much-quoted 2023 estimate gave GenAI 2.6 to 4.4 trillion dollars a year. The part that rarely makes the slide is that this sits on top of 11 to 17.7 trillion for non-generative AI and analytics. On a use-case basis, GenAI accounts for between a fifth and a third, depending on how generously you count. Counting every diffuse productivity gain, it reaches about a third \[10\].

**Spend.** IDC puts GenAI at 17% of global AI spending, heading to 32% by 2028. It is the fastest-growing slice, but still the minority of spend \[11\].

**Occurrences.** A Gartner analyst summed up GenAI as 90% of the airwaves and 5% of the use cases \[12\]. Practitioners confirm it: two thirds of data professionals work primarily on relational data, and above 80% in retail and insurance \[13-14\].

**Where value lands.** BCG locates 62 to 70% of AI value in core functions such as operations, pricing, supply chain, and R&D. That is the bread and butter of Tabular AI systems used for forecasting, scoring, and risk modeling \[15\].

**Realized returns.** 88% of companies use AI, but only 39% see any EBIT impact \[16\]. Meanwhile, fraud scoring, credit risk, and demand forecasting have been compounding for two decades.

Now apply a conservative haircut for vision, speech, and classic language processing. Tabular AI still comes out at no less than half, and probably two thirds, of enterprise AI value, and it is the vast majority of what actually runs in production. The reality is that the AI that holds most of the value receives a sliver of the attention.

# **Why the biggest category of enterprise AI became invisible**

The data underlines that Tabular AI is the biggest category of enterprise AI. But the headlines and analyst reports haven't focused on it. Why? Three biases are at work.

**Demonstrability.** A chatbot demos in thirty seconds to anyone who can see or hear. A fraud model that saves you tens of millions a year requires an advanced degree and does not demo at all: the loss simply never happens. We bias toward what we see working.

**Novelty.** Tabular machine learning has worked for twenty years, mostly on gradient-boosted trees, so it looks like the old world. Deep learning transformed text and images but stalled on tables for a decade. Research moved on and tables became the understudied modality of AI, but this is changing \[3, 17\].

**Narrative economics.** GenAI created a consumer market, venture capital needs stories, and advisory firms follow their clients. None of these signals measures value.

# **Three recommendations for enterprise leaders**

First, re-read your AI portfolio by data type, not by hype. Ask what share of your budget goes to what you can demo, and what share goes to what decides. The split should mirror value, not attention.

Second, reopen the shelf. Use cases killed on build cost two years ago deserve a new business case, because tabular foundation models and agents have changed the unit economics.

Third, put rigor in the loop. As agents generate more pipelines, invest in evaluation and methodology. A wrong forecast does not hallucinate visibly. It misprices inventory, risk or credit, at scale.

# **The next scoreboard**

The first chapter of enterprise GenAI was judged on how well AI talks. The next will be judged on how well it predicts, measured in EBIT. The largest AI market in the enterprise is not coming. It is already in production: underfunded, under-discussed, and about to be re-tooled. The companies that see it first will compound.

# **References**

1. TabArena, TabArena Leaderboard, Hugging Face. huggingface.co/spaces/TabArena/leaderboard
2. soda-inria, TabICL, GitHub. github.com/soda-inria/tabicl
3. EurIPS'25 Workshop on AI for Tabular Data, 6 December 2025, University of Copenhagen. sites.google.com/view/eurips25-ai-td
4. Neuralk, "The Tabular AI Shift: The Use Case Gap", 2026. neuralk.ai/post/the-tabular-ai-shift-the-use-case-gap
5. Prior Labs, TabPFNv3 technical report, 2026. arxiv.org/pdf/2605.13986
6. Google Research, "Introducing TabFM: A zero-shot foundation model for tabular data", June 2026. research.google/blog/introducing-tabfm-a-zero-shot-foundation-model-for-tabular-data
7. Brayer, M., "Should you care about Tabular AI?", 2025. sotabymariebrayer.substack.com/p/should-you-care-about-tabular-ai
8. etn interview with Philipp Herzig, WELT AI Summit, Berlin, 2026. linkedin.com/posts/every-board-asked-should-we-build-our-own-ugcPost-7509282448355274753-ERQB
9. IDC, Untapped Value: What Every Executive Needs to Know About Unstructured Data, white paper sponsored by Box, 2023. blog.box.com/90-your-data-unstructured-and-its-full-untapped-value
10. McKinsey & Company, The economic potential of generative AI: The next productivity frontier, June 2023. mckinsey.com/capabilities/tech-and-ai/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier
11. IDC, Worldwide AI and Generative AI Spending Guide, 2024. idc.com/resource-center/blog/a-deep-dive-into-idcs-global-ai-and-generative-ai-spending
12. Gartner, Erick Brethenoux quoted in The Register, 10 September 2024. theregister.com/2024/09/10/brute\_force\_ai\_era\_gartner
13. Kaggle, The State of Data Science & Machine Learning, 2017.
14. Olteanu, "Learning Models over Relational Data: A Brief Tutorial", 2019. arxiv.org/abs/1911.06577
15. BCG, Where's the Value in AI?, October 2024, and The Widening AI Value Gap, September 2025. bcg.com/publications/2024/wheres-value-in-ai
16. McKinsey & Company, The state of AI in 2025: Agents, innovation, and transformation, November 2025. mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
17. Hollmann et al., "Accurate predictions on small data with a tabular foundation model", Nature, January 2025. nature.com/articles/s41586-024-08328-6

# For more from Probabl

- Follow our latest updates on [LinkedIn](https://www.linkedin.com/company/probabl)
- Subscribe to our [monthly newsletter](https://hello.probabl.ai/subscribe-probabl-newsletter)
- Watch over 100 technical videos on our [YouTube channel](https://www.youtube.com/@probabl_ai)

[artificial-intelligence](https://blog.probabl.ai/tag/artificial-intelligence), [enterprise](https://blog.probabl.ai/tag/enterprise)

## Related posts

[![](https://blog.probabl.ai/hs-fs/hubfs/TL-GV-Nico.jpg?noresize&width=600&height=600&name=TL-GV-Nico.jpg)](https://blog.probabl.ai/creating-the-conditions-for-enterprises-to-truly-own-and-trust-the-models-they-put-into-production)

[open source](https://blog.probabl.ai/tag/open-source), [machine-learning](https://blog.probabl.ai/tag/machine-learning)

## [Creating the conditions for enterprises to own and trust their models: A conversation with Nicolas Flores-Herr](https://blog.probabl.ai/creating-the-conditions-for-enterprises-to-truly-own-and-trust-the-models-they-put-into-production)

![Picture of Gaël Varoquaux](https://blog.probabl.ai/hs-fs/hubfs/GV.jpg?width=50&name=GV.jpg) [Gaël Varoquaux](https://blog.probabl.ai/author/gael-varoquaux) 

 Thursday, April 30 2026

[Read more](https://blog.probabl.ai/creating-the-conditions-for-enterprises-to-truly-own-and-trust-the-models-they-put-into-production)

[![](https://blog.probabl.ai/hs-fs/hubfs/TL-Yann-AIBOM.jpg?noresize&width=600&height=600&name=TL-Yann-AIBOM.jpg)](https://blog.probabl.ai/the-enterprise-case-for-ai-bill-of-materials-understanding-ai-ownership-risk-and-control)

[machine-learning](https://blog.probabl.ai/tag/machine-learning), [data science](https://blog.probabl.ai/tag/data-science), [artificial-intelligence](https://blog.probabl.ai/tag/artificial-intelligence)

## [The Enterprise Case for AI Bills of Materials: Understanding AI Ownership, Risk, and Control](https://blog.probabl.ai/the-enterprise-case-for-ai-bill-of-materials-understanding-ai-ownership-risk-and-control)

![Picture of Yann Lechelle](https://blog.probabl.ai/hs-fs/hubfs/2017-06%20-%20Dark%20BG%20-%20Yann%20Lechelle%20-%20Square.jpg?width=50&name=2017-06%20-%20Dark%20BG%20-%20Yann%20Lechelle%20-%20Square.jpg) [Yann Lechelle](https://blog.probabl.ai/author/yann-lechelle) 

 Tuesday, April 21 2026

By Yann Lechelle, Executive President & Chairman of Probabl & Daniel Whitenack, CEO of Prediction...

[Read more](https://blog.probabl.ai/the-enterprise-case-for-ai-bill-of-materials-understanding-ai-ownership-risk-and-control)

[![](https://blog.probabl.ai/hs-fs/hubfs/undefined-May-07-2026-01-43-12-9639-PM.jpeg?noresize&width=600&height=600&name=undefined-May-07-2026-01-43-12-9639-PM.jpeg)](https://blog.probabl.ai/highlights-from-the-own-your-data-science-and-ai-workshop)

[open source](https://blog.probabl.ai/tag/open-source), [machine-learning](https://blog.probabl.ai/tag/machine-learning), [data science](https://blog.probabl.ai/tag/data-science)

## [Highlights from our Own Your Data Science and AI Workshop](https://blog.probabl.ai/highlights-from-the-own-your-data-science-and-ai-workshop)

![Picture of Cailean Osborne](https://blog.probabl.ai/hs-fs/hubfs/680c86be-b873-4f13-a8c5-7805b22a4e6d.jpg?width=50&name=680c86be-b873-4f13-a8c5-7805b22a4e6d.jpg) [Cailean Osborne](https://blog.probabl.ai/author/cailean-osborne) 

 Thursday, May 7 2026

On Tuesday, May 5th, we packed the Open Stage at Station F for a full day dedicated to “Own Your...

[Read more](https://blog.probabl.ai/highlights-from-the-own-your-data-science-and-ai-workshop)

![illustration-probable](https://blog.probabl.ai/hubfs/illustration-probable.svg "illustration-probable")

- [Company](https://probabl.ai)
- [Join Us](https://join.probabl.ai/)
- [Blog](https://blog.probabl.ai)

[![69ee669dda802928688b7b3e\_logo-probabl (2)](https://blog.probabl.ai/hubfs/69ee669dda802928688b7b3e_logo-probabl%20(2).svg "69ee669dda802928688b7b3e_logo-probabl (2)")](http://probabl.ai)

[Privacy Policy](https://probabl.ai/privacy)               © 2026 Probabl, all rights reserved.

```json
{
  "@context" : "https://schema.org",
  "@type" : "BlogPosting",
  "author" : {
    "@type" : "Person",
    "name" : "Yann Lechelle",
    "url" : "https://blog.probabl.ai/author/yann-lechelle"
  },
  "dateModified" : "2026-09-30T13:23:39.392Z",
  "datePublished" : "2026-09-30T13:15:15.000Z",
  "headline" : "GenAI owns the conversation. Tabular AI owns the P&L.",
  "image" : [ "https://blog.probabl.ai/hubfs/TL-YL-GitHub.jpg" ],
  "mainEntityOfPage" : {
    "@id" : "https://blog.probabl.ai/tabular-ai-enterprise-opportunity",
    "@type" : "WebPage"
  },
  "publisher" : {
    "@type" : "Organization",
    "logo" : {
      "@type" : "ImageObject",
      "url" : "https://blog.probabl.ai/hubfs/logo-probable-black.svg"
    },
    "name" : "Probabl"
  }
}
```