In early June, IDC published a rather inconspicuous blog post titled "The Next Stop for ERP is 'AI That Can Execute'." If you don't look closely, it's easy to mistake it for yet another vendor-sponsored industry press release — after all, the term "AI+ERP" has been overused by 2026.
But I took a closer look at the numbers inside: in 2025, the size of China's AI-enabled ERP market was $315.7 million, a year-on-year increase of 96.1%. Doubling in one year. And this doesn't even include traditional ERP license revenue—it only counts revenue generated by AI products, AI capabilities, and AI-related modules in ERP scenarios. In other words, companies are genuinely increasing their investment, not just rebranding old systems and selling them again.
96.1% year-on-year growth rate of China's AI-enabled ERP market in 2025, reaching a scale of $315.7 million
But what's more worth watching is not the growth rate—there has never been a shortage of growth rates in AI news over the past two years—but the core judgment put forward by IDC: ERP is moving from a "system of record" to a "system of execution".
What does this mean? Let me translate: In the past, the core value of ERP was to "record" — record business, record processes, record data. When an order came in, people entered documents into the system, and the system managed approvals, workflows, and reports. But now that AI has arrived, companies are no longer satisfied with just "recording"; what they want is for the system to "get things done for me." This is not just about changing the menu from a functional list to a dialog box; it's about the entire product logic of ERP undergoing a transformation.
IDC breaks it down into three lines: the entry point shifts from functional menus to business intent, agents evolve from auxiliary applications to process execution carriers, and the competitive focus moves from feature completeness to AI implementation. These three lines point in the same direction: ERP vendors will no longer compete on who has more features, but on whose AI can actually enter business processes and get the work done.
It's not very interesting to just talk about trends; I happened to come across a practical Odoo case that clearly lays out the results of implementing this IDC theory.
A distributor with annual revenue of $9.3 million burns $43,000 per month in "invisible waste"
Meridian Supply Co. — a Dallas-based industrial parts distributor with 4 warehouses, $9.3 million in annual revenue, and an inventory system that "updates Excel every 24 hours."
$43,200Meridian's monthly hidden operational waste caused by process fragmentation
Its core problem is not "having no system," but that the system is like a handful of loose sand. QuickBooks handles finances, an Access database manages inventory, plus a time-tracking tool that no one uses seriously. Data between these three systems is manually transferred by people. And what's the result?
- Inventory accuracy rate 67% — 33 out of every 100 SKUs cannot be found. Out of the $2.1 million inventory value, $690,000 is "phantom inventory".
- Order fulfillment rate 79.3% — industry average 94.7%. This gap burns approximately $18,400 per month in B2B renewal revenue.
- Operations Director spends 41 hours per week on reconciliation—almost a full-time manual effort to reconcile POs.
In a word: if the system lacks the ability to "execute," people have to do the work for the system. And the more they do, the more mistakes they make.
Six-week deployment, three-tier architecture: How Odoo becomes an "execution machine"
Meridian did not take the NetSuite route — the quote from their partner was $187,500 in implementation fees and $42,000 in annual subscription fees, which is not realistic for a mid-sized company with less than ten million in revenue.
They chose Odoo, giving their implementer Braincuber only 6 weeks—because the Q4 peak season was approaching.
Braincuber built a three-tier architecture:
Layer 1: Core ERP Foundation—Odoo 17's inventory, procurement, sales, and accounting modules cover all 4 warehouses. Replaces QuickBooks, Access databases, and two redundant SaaS services. Inventory data updates in real time, eliminating 24-hour batch processing delays.
Layer 2: AI Automation Engine – A demand forecasting model built on the Odoo AI platform plus custom Python services. It pulls 36 months of historical data, supplier lead times, and seasonal shipping velocity to predict stockouts 14 days in advance. Document AI reads supplier PDF invoices, automatically matches them to POs, and flags discrepancies over $50.
Layer 3: Warehouse Intelligence Layer — AI-driven putaway rules that automatically assign fast-moving SKUs to picking locations within 12 feet of the packing station.
6 weeks from data migration to AI layer activation, full-process implementation cycle
90 days later, the numbers speak for themselves
| Indicator | Before deployment | 90 days later | Improvement magnitude |
|---|---|---|---|
| Inventory accuracy rate | 67% | 96.3% | +$690,000 capital visibility |
| Order fulfillment rate | 79.3% | 93.8% | Monthly recovery of $18,400 in revenue |
| Single order picking time | 4.7 minutes | 1.6 minutes | Save 190 working hours per month |
| Invoice processing time | 18 minutes/piece | 1.9 minutes/sheet | Release Operations Director 41 hours/week |
| Out of stock event | 9.4 times/month | 1.1 times/month | Reduce by 88% |
That operations director Marcus, who used to spend 41 hours a week manually matching POs, now only needs 6.5 hours a week for exception review. The rest of his time is spent on supplier negotiations and capacity planning—that's what he was hired to do.
Moreover, a simple calculation shows: the monthly recovered $43,200 in operational waste covers Odoo's entire annual SaaS subscription fee in just 23 days.
Mastering AI is not the key; the key is embedding AI into the right data foundation
Finally, I want to say one thing. What makes the Meridian case interesting is not AI itself—demand forecasting models are everywhere on the market, and Document AI is nothing new. What is interesting is Braincuber's summary: AI alone, mounted on fragmented operational systems, is like using a smart brain to command a paralyzed body.
This is something that IDC's blog post didn't explicitly say, but the entire report implies. The "AI execution" of ERP is not something that can be achieved by simply adding an AI layer; it requires enterprises to first clean up their data foundation—dismantling the fragmented structure of QuickBooks, Access, and Excel, and replacing it with a unified data model. Meridian was doing exactly this two weeks ago: clearing out 2,700 duplicate SKUs, 430 duplicate supplier records, and correcting conflicting pricing data.
This process is not glamorous, and it doesn't involve the term "AI," but it determines whether the subsequent AI layer can deliver results. IDC statistics show that the AI-enabled ERP market has doubled in a year, which is from the supply-side perspective. From the demand side, the companies that are likely to see returns are probably those that first organize their data and then layer AI capabilities on top—rather than companies that buy a bunch of AI tools but are still working with messy data.
For CIOs, the choice in the second half of 2026 may not be "whether to adopt AI," but "whether to first lay a solid foundation before adopting it."
