Why Every ERP Vendor Suddenly Claims to Be AI-Powered (And How to Tell What's Real)

Open the homepage of any ERP vendor right now, old or new, enterprise or small business, and you will find the same word within the first screen: AI. Systems that have run on the same core logic for a decade or more now describe themselves as AI-powered. Vendors that added a single chat widget last quarter now lead with "AI-native." Even categories that had nothing to do with machine learning a few years ago, expense tracking, procurement, helpdesk routing, are suddenly "intelligent."

This is not a coincidence, and it is not really about the technology. It is about what buyers now expect to see on a pricing page, and what happens to a sales deck when a competitor gets there first.

Why this happened all at once

Two forces are doing most of the work here.

The first is competitive pressure. Once a handful of ERP vendors started shipping genuine AI features, usually large players with the engineering budget to build agents into their core workflows, every other vendor in the category faced a simple problem: a demo that does not mention AI now reads as behind the times, even if the underlying product is perfectly capable. Marketing teams responded the way marketing teams always respond to a category-wide shift, which is to update the language faster than the product changes.

The second force is that bolting a generic AI layer onto an existing system is genuinely easy, and building AI into the transactional core of an ERP is genuinely hard. Any vendor with a REST API and a weekend can wire up a chat interface that calls a large language model, wraps the response in company branding, and calls it an AI assistant. That chatbot can summarize a report, answer a question about a policy document, or draft an email. What it usually cannot do is act on live transactional data: adjust a reorder point because of a demand signal, flag a mismatched invoice before it posts, or reprioritize a production schedule when a supplier slips. That kind of AI requires rebuilding workflows around the model, not routing questions to it. It is the difference between adding a feature and redesigning a product, and most vendors under pressure to ship something this quarter chose the former.

G2's 2026 ERP buyer report gets at this directly, advising buyers to "look beyond feature claims and evaluate AI based on workflow impact, automation value, and decision support." That is good advice, but it is also where most guidance stops. Nobody hands buyers the actual questions to ask in the room. Even vendors selling AI evaluation frameworks tend to gate the useful part behind a form: Rillet's own AI ERP buyer's guide puts the actual content behind a download form, not in front of you. So here is the uncomfortable, specific version.

Three ways to spot AI-washing

The vague-specificity flag. Ask a vendor to name one AI feature and describe the exact workflow it changes, start to finish, and listen to what comes back. A vendor with real AI in the product will name it: "the reconciliation agent flags unmatched transactions before month end close, here is what it looks at and here is what it changes in the ledger." A vendor doing AI-washing answers with a category instead of a feature: "AI-powered forecasting," "smart insights," "intelligent automation." Categories are marketing. Named features tied to named workflows are product. Morph's guide to AI-washing calls this out as one of the clearest tells in B2B software claims broadly, ERP included.

The infrastructure-silence flag. Ask who actually runs the model behind the feature: their own infrastructure, a specific provider you could name, or something they would rather not specify. Vendors with a real architecture answer this without hesitation, because they made a deliberate choice and can defend it. Vendors that wrapped a single API call in a UI often go quiet or vague here, because admitting "we call a third-party API and mark up the response" undercuts the pitch. This tracks the same diagnostic that Iternal.ai uses to separate genuine AI partners from ones faking it: partners with a real practice answer with specific engagements and specific outcomes, while partners doing the wrapping answer in generalities and steer the conversation back to their vendor relationships.

The everything-costs-extra flag. AI gets marketed on every page of the site, in every email, in the first thirty seconds of every demo, and then turns out to be locked entirely behind the single most expensive tier, often with no visibility into what it actually costs at your headcount until a sales call. That alone is not damning. Building and maintaining real AI features costs money, and charging for that is fair. It is a signal, not proof, when the marketing intensity around AI is completely disconnected from how little of the product actually carries it. First Line Software's analysis of AI-washing describes exactly this pattern in general terms: public claims about AI capability running well ahead of the operational system built to support them, with the gap showing up in what the product actually does rather than in what the marketing says.

None of these three flags alone proves a vendor is faking it. A legitimate vendor might have a genuinely good reason for a vague answer on any single question. But when a vendor hits two or three of these in the same conversation, that is not a coincidence anymore.

The questions to actually ask

You do not need a scoring rubric or a maturity model. You need a short list of literal questions you can ask out loud in a demo or a sales call, and you need to notice when the answers dodge. Here are nine, pulled from the categories that expose the most.

On architecture:

1. "Show me an AI feature acting on live transactional data, not a reporting dashboard."
2. "If I removed this AI feature entirely, what would stop working?" A feature that is decorative rather than load-bearing usually reveals itself here.
3. "Is AI available across many modules, or only in one bolted-on add-on?"

On provider and infrastructure lock-in:

4. "Whose infrastructure does the AI run on: yours, or a model provider I can choose?"
5. "Can I switch AI providers, OpenAI, Anthropic, a local model, without switching ERPs?"
6. "If your AI vendor changes pricing or shuts down, what happens to this feature?"

On proof and specificity:

7. "Give me one specific, named AI feature and the exact workflow it changes, end to end."
8. "How many of your total modules have AI touching them today, by name?"
9. "What can this AI not do yet?" This last one is counterintuitive but reliable. A vendor with real AI in production has hit real limits and can describe them plainly. A vendor selling a marketing layer usually cannot answer this at all, because admitting a limitation means admitting the feature is not the everything-machine the homepage implies.

Write these down before your next demo. Ask them in order. Watch for the pivot back to a feature list when the answer to a specific question does not exist yet.

How Yukti answers a few of these

We would rather answer these questions plainly than pretend they do not apply to us, because they do.

On architecture and specificity: Yukti ships fourteen named AI agents, each embedded in a specific module and a specific workflow, not one chatbot layered over the whole system. The Reconciliation Agent works inside Accounting on unmatched transactions. The Stock Optimization Agent works inside Inventory on reorder decisions. The Ticket Routing Agent works inside Helpdesk on incoming requests. If you asked us to name one and trace the workflow end to end, we could, for any of the fourteen.

On provider lock-in: every one of the fourteen agents runs on infrastructure you choose. Point it at OpenAI, Anthropic, Google, or a local model through Ollama. Swap providers later and nothing about your workflows or your underlying ERP has to move. Pricing is published and flat: no negotiated quote, no per-module AI surcharge. See how that holds up next to NetSuite or Infor, where a public price list does not exist at all.

And here is the honest part, the one we would rather state upfront than have you discover in a sales call: the fourteen AI agents are an Enterprise-tier feature, not something included in the free Community edition. Community is free forever, fully open source under LGPL-3.0, self-hosted, with 50-plus core ERP modules and no user limits, but it does not include the AI agents. Those are part of Enterprise, priced at $9.99 per user per month billed monthly, or $6.66 per user per month effective on a two-year plan with a third year free. We are not going to claim AI is free everywhere when it is not. What we can tell you is that the price is one flat number, visible on our pricing page right now, not a quote that depends on how the sales call goes. When you do pay for it, you are paying for AI embedded in the modules you use every day, running on infrastructure you choose, not a chatbot with a new coat of paint.

Ask any vendor these nine questions before you sign anything. Ask us the same ones. The answer should never be a shrug.

Run them against your own shortlist, check the real numbers on our pricing page, or talk to our team if you would rather walk through the specifics live.