AI in ERP Systems: Hype or Real Value?

A View from Odoo Experience

AI in business systems is not a new story. SAP’s Joule has been on the market for more than three years, and Odoo built AI agents and automated actions into version 19. What is new is something else: both vendors have moved from AI that advises to AI agents that carry out repetitive tasks inside the ERP on their own.

At the end of September, our team attended Odoo Experience in Brussels, an event with around 45,000 expected visitors, where this ambition was the central theme. Odoo 20 was unveiled with a strong emphasis on integrated AI (1). In his opening keynote, Fabien Pinckaers, Odoo’s founder and CEO, said that AI no longer just analyzes data but also does the work through agents. It’s an appealing picture: a business system that needs no instructions because it understands and acts on its own. That is why we watched the conference with one more question in mind: how much of this holds up in practice?

Promise vs. Experience

Among companies that buy “AI-ready” or “AI-powered” ERP modules, we increasingly see the same pattern: the functionality exists and the demo looks impressive, but real benefits only arrive after months of preparation. The reason is simple. AI inside an ERP is not a magic layer added on top of an existing system; it is a tool that depends on the quality of what lies beneath it. If data is inconsistent, if items are duplicated under different names, if financial records have gaps, AI doesn’t fix the mess. It inherits and amplifies it.

Duplicates are the simplest example. AI can suggest that two records are the same supplier or item only if the records contain enough reliable data to support that conclusion, such as a name, an address, and a tax ID. The more incomplete the records, the less reliable the suggestions.

In other words, AI in an ERP doesn’t create value out of nothing. It amplifies what the system already does, for better or worse.

Where AI Delivers Real Value

Once the marketing layer is stripped away, a few areas with clearly measurable impact remain:

– Procurement and inventory forecasting: According to McKinsey, better demand forecasting can reduce inventory levels by 20–30% in distribution companies (2). This is an estimate of potential; actual results depend on the industry and data quality.

– Document entry automation: OCR and language models can now read invoices, delivery notes, and purchase orders reliably. This was the most concrete thing shown in Brussels, with one important detail: the system proposes, and a human confirms.

– Anomaly detection: An AI layer monitors financial transactions in real time and flags suspicious patterns before a human would notice them.

– Conversing with the system: Users ask the system a question as they would a colleague, in writing or by voice, instead of building a report across several modules. This especially helps smaller teams without dedicated BI support.

Odoo itself offers an interesting example. According to a post by Pinckaers from April 2025, two months after introducing Odoo AI and live chat, the same team of 650 support engineers handles twice as many cases, and 70% of issues are resolved immediately, on the spot (3). These figures were published by Odoo itself and do not separate the effect of AI from the effect of human chat, but they show a direction: here AI serves as additional capacity alongside people, not as a replacement for them.

The common denominator of all these examples is that they work best when solving a well-defined problem, not when trying to “run the whole business.”

“When we talk about AI in ERP, and in other IT systems too, the key is to start from a concrete business problem, not from the technology itself. When processes are well organized and data is clean, AI can take over part of the routine work, speed up activities, and spot deviations much faster. That way we don’t just save users time; we enable them to get from existing data to better decisions more quickly. For me, AI is above all an upgrade that makes ERP faster and more useful in everyday work.”— Ivanka Vranješ, Director of Engineering & Delivery at Notus

Where the Hype Is Greatest

Most of the disappointment comes from promises of fully autonomous AI agents that supposedly make business decisions on their own. In Brussels, this idea was the loudest: a user can create an agent simply by describing it in writing or by voice. The demonstration is convincing, but in practice such autonomy requires quality data and process discipline that few companies truly have. The question therefore shifts from “can it be done?” to “who is allowed to run it, what is it allowed to change, and who is accountable when it makes a mistake?”

Another common scenario: a company buys a premium AI package but uses only a fraction of its capabilities in the first year, while the license bill grows with the number of users or queries.

“In my experience so far, AI in ERP systems today is not here to replace us in making key business decisions. Nor is it here to replace the human operator in their regular tasks, but to free up everyone’s time from administrative drudgery. Companies looking for a quick fix will be disappointed fast, and may even lose people in the process, but those looking for a tool that will offer the right information at the right moment and point to solutions will gain a huge advantage in the market. Today we talk about AI the way we once talked about AutoCAD or even ERP systems, which back then stirred fear of job losses and are now an integral part of the process. Where do I see things most often getting stuck? The biggest mistake is trying to introduce AI into every part of the system at once. It’s best to treat AI as a new, highly efficient assistant to whom we assign one very specific task. It is most effective when given a task that is clear and has measurable results, such as automating the entry of incoming invoices. Then we review and measure the results, and we can extend the lessons learned further.”— Adis Petrović, Head of the Odoo Department at Notus

Conclusion: It’s Not About Hype or Value, but Maturity

It is a mistake to frame the question as a choice between two extremes. AI in ERP systems is neither pure marketing fiction nor a universal solution. Its value depends directly on the maturity of the data, the clarity of the problem being addressed, and the realistic expectations of the company introducing it.

Companies that treat AI as a magic wand that will “fix the ERP” will be disappointed within the first few months. Those that set it up as a tool for a specific task, such as inventory forecasting, document entry automation, and anomaly detection, will see a real, measurable return on investment. Technology rewards discipline and preparation, not just enthusiasm.

Do You Recognize Yourself in These Questions?

Before you begin selecting or upgrading an ERP system with AI components, it’s worth answering four questions honestly:

1. How clean and structured is our data today, realistically, not optimistically?

2. What specific problem should AI solve in the first six months?

3. What is the real cost per transaction or query, not just the license price?

4. Who within the company or partner organization will maintain and oversee the AI layer after implementation?

If you don’t have a clear answer to any of these questions, that is not a reason to give up on AI in ERP, but a sign that you need a partner who will lay the foundations first and only then build on them.

Sources

1. The Brussels Times / Belga (September 24, 2026). Odoo announces €10-billion valuation and a partial price increase. https://www.brusselstimes.com/2332306/odoo-announces-e10-billion-valuation-and-a-partial-price-increase

2. Oca, A., Panikkar, R., Sampat, C., Brown, T. (November 15, 2024). Harnessing the power of AI in distribution operations. McKinsey & Company. https://www.mckinsey.com/industries/industrials/our-insights/distribution-blog/harnessing-the-power-of-ai-in-distribution-operations

3. Pinckaers, F. (April 2025). LinkedIn post on Odoo’s support. https://www.linkedin.com/posts/fpodoo_odoos-support-used-to-be-by-email-through-share-7314576215279919104-cp-i/