If you stack together and read the ERP announcements from the first week of November, you will find a rather interesting phenomenon: almost every one uses "launched," "released," or "deployed," but if you carefully turn to the second page, their status differs by three levels.
Sage Intacct's finance agent is now generally available in six markets, but it is disabled by default and must be turned on by an administrator; half of NetSuite's batch of AI tools are labeled "planned for a future release"; Organizely says factories in six states have already deployed it, and the report then adds—a demo is a demo, not actual test results. CMiC released NexusAI, but did not provide pricing, availability timing, customer names, or performance figures—not one.
This information isn't some scandal dug up by someone—it's all written by the manufacturers themselves in their announcements or documents. It's just that it doesn't easily appear on the first page of sales materials.
If you come back from the selection meeting with only a feature checklist, no matter how many boxes are ticked, it's useless.You first have to break down the word "release".
1. Three tiers of status—don't mix them in one table
I rearranged this week's information by "can we open today or not," and it roughly looks like this:
| Manufacturer / Product | Announcement statement | Actual status | The sentence that should be questioned further |
|---|---|---|---|
| Sage Intacct R4 / Finance Intelligence Agent | 11-07 report, six markets officially available | Already GA, but disabled by default and must be explicitly enabled by an administrator; connected Excel and the new approval workflow are still in Early Adopter | In my region, for this license tier, who needs to approve it if it's opened? |
| NetSuite Next / Ask Oracle | SuiteSkills 45+ features, AI Workspaces, and industry SuiteAgents launched | Some are already available; Studio, SuiteFlow Agent, Management Space, Connectors, Data Intelligence Agent, etc. are marked future release; phased rollout by US/CA → UK/AU/NZ | Which items in the list can I see in my account today? |
| CMiC NexusAI (CONNECT 2026, 11-02) | Plain-language AI entry released, driven by MCP Server, following request—review—confirmation | Pricing, availability date, launch timing, customer names, and performance figures were not disclosed; the three benefit claims have no data. | When can it be signed into the contract, and how will it be priced? |
| Organizely (YC Fall 2026 Batch) | Factories in six states have deployed AI-native manufacturing ERP | The state was not named, and neither installation numbers nor customers were provided; the phrases "relocate production lines" and "approve POs by SMS" on the official website are explicitly marked as company demonstrations; the only named customer is a founder-affiliated entity | Has any non-affiliated customer ever produced verifiable figures? |
| Infor 2026.10 Distribution Edition (11-04 Detailed Explanation) | a series of agents such as Quoting / Purchasing / Product / Batch Wave | The announcement does not provide measured customer outcomes; cost reduction and speed improvements are all vendor-stated benefits; there is no per-product entitlement / availability matrix. | On which page are the enablement steps and license checklist? |
| ENTERSOFTONE (Romania, 11-03) | AI version of EBS / SOFT1 ERP, "data processing time reduced by up to 70%" | Product has been released; the 70% figure is the manufacturer's claim, with no third-party testing. | What is the denominator of this 70%, and who measured it? |
Not one company in this table is lying. The problem is that "launch" is too broad a word: it can mean code merged into the main branch, a canary rollout in one region, a demo shown at a customer conference, or something that can actually be signed into a contract and go live next month. If procurement teams use the same press release to set their schedules, it is only a matter of time before they are caught short.
My own habit is this: whenever an announcement contains "launched," "released," or "already deployed," I always downgrade it to "pending verification" until the other party provides all four items: version, region, license tier, and enablement steps. If they can't provide them, I treat it as not shipped and don't write it into the rollout plan.
2. Even if it is marked "GA", there are still two gates ahead
The Sage example is worth discussing separately, because it is the most "solid" one in this batch, and also the most easily misread.
Finance Intelligence Agent is powered by Sage Copilot. Finance staff can directly ask in natural language, "Which suppliers' bills are overdue?" The system provides a summarized answer, along with an analysis panel showing how it reached its conclusion. It is officially available in six markets: the US, UK, Canada, Australia, New Zealand, and South Africa. Sounds like it could go live tomorrow.
But the document states three things: disabled by default and requires an administrator to explicitly enable it; inherits the underlying user permissions—in other words, the accuracy and security of AI output depend entirely on how strictly this company's data access controls are configured; in addition, the connected Excel reporting and new approval workflows promoted at the same time are still in Early Adopter. The implementation checklist provided by the assessor is also quite plain: first audit role permissions, confirm that the operating regions are on the supported list, and then check whether the company's own IT or finance system administrator has the capacity to enable and maintain these Copilot features.
Look, two of the three have nothing to do with "whether AI is good or not"; they're about permissions and people.
The most accident-prone combination is "permissions were already configured loosely" + "the administrator turns all the switches on to rush a demo." When user permissions are loose, the agent can summarize financial data it should not have been able to see into a fluent paragraph and present it, and it even looks quite convincing. Defaulting to off is actually a well-intentioned design: it forces you to take a look at the permission table before turning it on.
Odoo has a similar threshold here, just in a different form: the Community edition has no built-in AI agent, so you have to bring models in through custom modules or external automation platforms, and permissions, data scope, and invocation costs all have to be fully configured by yourself. The upside is that no vendor decides for you "which ones are enabled by default"; the downside is that no vendor writes the checklist for you either. Open capabilities do not equal completed configuration—this statement holds true for both closed source and open source.
3. What really stalled the project was that 86% figure
In the November issue of much. Consulting, there was a set of data I read over several times: among 150 surveyed manufacturing companies, 86% listed “lack of internal resources” as the biggest obstacle to introducing or modernizing ERP; at the same time, 64% still plan to budget for ERP, digitalization, or process automation in the next two years.
86%
Stuck on a lack of internal resources, not a lack of money (150 manufacturing companies)
64%
Still plan to allocate budget for ERP / digitalization / process automation in the next two years
That judgment in the original text was quite blunt: whoever drags the business department and daily operations in only at the launch stage will pay the price later. Applied to today’s batch of AI features, it is even more apt—for each row in the table above, the “one question that should be asked” requires someone who understands both business definitions and system configuration to ask and verify. If this person is not present, no matter how detailed the checklist is, no one will execute it.
The same issue also noted two small items that are somewhat interesting when viewed together. Xentral founder Benedikt Sauter has stepped back from day-to-day operations, with Domenico Cipolla taking sole charge; the recently disclosed 2024 figures are revenue of €14.5M (+22%), a net loss of €9.4M, cash down from €39.6M to €32.1M, cumulative funding of about US$95M, and the last round in August 2021. The report's interpretation is that five years without new funding shows it is now hard to get venture capital in the ERP space, and competitors already include Odoo and self-built AI solutions. On the other side, Odoo is setting up an office and hiring in Munich, and Business Shows has also changed from that universal “template IKEA store” to walking through real processes by industry.
On one side, capital is receding; on the other, demos are being turned into real workflows. The direction is actually the same: the market no longer buys “having an AI”; it buys “this thing works in my industry and in my workflow,” and validating that depends on people, not budget.
The proportion planning to invest money, and the proportion blocked by "staff shortage". The gap between the two is the one slot in the project schedule that should be filled first
IV. Before Entering the Selection Meeting, First Do Four Effortless Things
1. Turn the announcement into a three-column status table. For each feature, state clearly "available today / must wait for a version or region / demo only", and attach the vendor-provided entitlement matrix and enablement steps. If the matrix cannot be obtained, treat it as not shipped. This step can prevent most rework during the launch phase.
2. Ask about the denominator, not the adjectives.After "70% reduction," "efficiency improvement," and "real-time visibility," follow up with three questions: over what time period, how many documents, and who measured it; and whether any non-affiliated customer has independently rerun it. When vendors themselves write "company demonstration" or "expected results," record them verbatim in the minutes—do not let them quietly turn into "measured results" when paraphrasing.
3. Before enabling features, first perform a permissions and data checkup. Go through role permissions, master data consistency rate, and approval chain completeness, then select one high-frequency process for a small-scale trial. For features that are disabled by default, don't turn them all on at once just to rush a demo.
4. Reserve business-side personnel in the schedule first. It's not "come participate in UAT during the launch phase"; it's being present throughout, from requirements grooming to acceptance, and making it clear that they have the authority to say no to the agent's output. For that 86% question, the answer can only be a human.
The feature list is written by the vendor; the conditions for starting work are your own. The truly valuable information this week is not in those headlines about "launch" and "release," but in the wording on the second page of the announcement: disabled by default, planned for a future release, company demonstrations, undisclosed. Picking out these words brings you closer to the truth of the project than comparing ten more features.
Next time someone comes with a checklist full of checkmarks and says, "This AI is the most complete," you might just ask one question: which of these items can I actually use today, in my version and my region?
This article is compiled and analyzed based on public information: much. Consulting's "Odoo News November 2026" (a survey of 150 manufacturing enterprises, Xentral operating data, Odoo Munich and Business Shows), ERP News' November reports on Sage Intacct R4, Organizely and Infor 2026.10 distribution edition, Decision Intelligent's assessment of Sage and manufacturing AI scheduling (2026-11-07), IT Brief Asia / ADTmag / Snapshot Design's reports on NetSuite Next, AI Frontier Post's report on CMiC NexusAI (2026-11-02), and Business Catalog's report on ENTERSOFTONE (2026-11-03). The data is current as of 2026-11-08 and is for reference by enterprise digitalization leaders; it does not constitute procurement advice.
Zhongdi Information · Focused on Odoo and digital solutions for enterprise management
