The conversation around artificial intelligence has taken an interesting turn: now it’s the manufacturers themselves who are putting realistic figures on the table. When a hospitality software provider like Infor acknowledges that its AI agent "only" resolves around 5% of the work
At Tisa we have been implementing management systems since 1987, and we have seen enough technological waves come and go to know one thing: the technology that pays off is the one that solves a specific problem, not the one that promises to solve everything. Let’s break down what that "5%" means, why it is good news and how to translate it into useful decisions for your company.
When the provider admits the agent "only" does 5%
For a manufacturer to acknowledge the limits of its own AI is unusual and very healthy. What matters is not the exact percentage —which is worth verifying before quoting it— but the underlying message: today’s AI agents automate a narrow band of tasks, usually the most repetitive and rule-based ones. The rest still needs human judgment, business context and supervision.
That nuance completely changes the conversation. Instead of asking "will AI replace my team?", the useful question becomes "which specific tasks can my team offload so they can devote themselves to what adds value?". The answer is rarely "everything", but that 5% —or whatever the percentage is in each case— can represent hours of administrative work freed up every week.
What agentic AI in the ERP is (and what it is not)
It is worth clarifying the terms, because marketing mixes them up. An AI assistant answers questions and helps to draft or summarize. An agent goes a step further: it executes multi-step tasks using tools, it does not just respond. In the context of an ERP, that means actions such as preparing a draft order, reconciling transactions, classifying invoices or proposing a response to a customer based on the system’s data.
Agentic AI in the ERP is promising precisely because the ERP concentrates the company’s operational data: finance, purchasing, sales, inventory, projects. But that is also where its limits appear. An agent that acts on messy or incomplete data does not save work: it multiplies it, because it generates errors that someone will have to correct. That is why hallucination —when the model produces incorrect information that looks true— is a real risk that must be governed, not ignored.
Assistant, agent and automation: they are not the same
A Power BI dashboard informs you. An assistant like Copilot helps you draft or query. A Power Automate flow automates a process with fixed rules. And an agent combines several of these elements to execute a complete task with a degree of autonomy. Knowing which rung of that ladder you are on prevents you from paying for an "agent" when what you need is, quite simply, a good report or a basic automation.
Why that 5% is good news for your SME
A vendor who acknowledges the limits of their product is giving you realistic expectations, and realistic expectations are what make a project succeed. These are the practical takeaways from that 5% band:
- The use case is what matters. It is not about "adding AI", but about identifying a measurable task that can be automated reliably. A well-chosen 5% —the one that removes your most tedious work— is worth more than a theoretical 50% that never works in production.
- The human stays at the center. Agentic AI removes friction, not responsibility. The decisions and the supervision remain yours.
- The investment can be staged. Starting with a specific case lets you measure the return before scaling, without committing large budgets up front.
This approach —AI applied by measurable use case, as opposed to the noise— is exactly the one we advocate with our clients. We prefer a small project that works to a big promise that disappoints.
Without orderly data, no agent is worth it
Here is the key point that many providers do not stress enough: AI performs to the extent that the data is orderly and governed. An agent on Business Central can help you reconcile or draft, but only if the accounting, customer or inventory information is reliable.
That is why, before talking about agents, we talk about data. Data governance —policies on quality, access, privacy and use— is not bureaucracy: it is what separates an AI that helps from one that makes things up. And techniques such as RAG (giving the model access to the company’s own knowledge base) allow the AI to respond with verifiable information instead of improvising.
In practical terms, a well-implemented ERP with clean data is the best "training" you can give to any AI you plan to incorporate afterwards. Getting things in order beforehand is half the project.
How to leverage agentic AI in the ERP step by step
If you run a private company and want to move from the headline to action without burning through your budget, this is the path we recommend:
- Get the foundation in order. Make sure you have a solid ERP —Business Central, in the Microsoft ecosystem— with reliable data in its key areas: finance, purchasing, sales, inventory.
- Choose a measurable use case. A specific, repetitive task with a quantifiable result. For example, classifying and pre-coding supplier invoices, or preparing draft responses to frequent queries.
- Start with what you already have. Copilot integrated into Business Central and Power Platform
We apply this same criterion in sectors where we have a track record —retail, distribution, construction, food industry, hospitality (with our Tisa PMS vertical on Business Central), sports centers or fleets—: first the process and the data, then the AI layer that delivers a specific benefit.
The "5%" is not little: it is realism
The 5% band that an honest provider acknowledges is not a disappointing ceiling, it is a sensible starting point. Agentic AI in the ERP is advancing quickly, and it will do so even more in 2025–2026 with AI embedded in management systems. But the competitive advantage will not go to whoever boasts most about AI, but to whoever puts their data in order, chooses their use cases well and measures the results.
At Tisa we help you do exactly that: start from a solid ERP, identify where AI adds real value and land it without hype. If you want a no-obligation assessment of where to start in your company, write to us at info@grupotisa.com, call us at (+34) 971 305 885 or visit us at grupotisa.com. We’ll talk about your specific case, not about headlines.