There’s a classic Harvard Business School case about a rowing coach who forms his best crew with the eight strongest rowers and leaves the eight weakest in the junior boat. Against all odds, the "best" boat lost time and time again. The reason? Each star rowed with all their strength on their own, broke the shared rhythm and slowed the boat. The junior rowers, aware of their individual limits, rowed in sync. It’s the best image for understanding why you need to think about AI for the whole company and not just about making each person stronger: the collective result can be worse even if every individual performs at their peak.
Many companies live exactly this today with artificial intelligence. They have given powerful tools to every employee, and yet the business as a whole does not row any faster. The difference between optimizing the person and optimizing the system is what separates "we’ve tried AI" from "AI is making us money".
From individual performance to AI for the whole company
This phenomenon has a name: suboptimization. It occurs when one part of the system is maximized and, unintentionally, the whole is harmed. A recent SAP analysis explains it very well and provides a revealing figure from McKinsey’s State of AI: although AI adoption is almost universal (88% of organizations use it regularly in at least one function), only a minority report a real impact on their results at the company level (source: news.sap.com, "How to Optimize AI for the Entire Enterprise, Not Just the Individual", 2026).
We work with SMEs that are evaluating or already using Microsoft Dynamics 365 Business Central, and we see the same pattern on a smaller scale. Generative AI is exciting on each person’s desktop, but the value does not "rise" to the business. There are three reasons that explain it: intensity, context and prediction. Let’s go through all three, grounded in the reality of a private company.
Intensity: AI does not always remove work, sometimes it multiplies it
The promise is "you’ll do the same in less time". The practice is usually different: since generating things is now immediate, people generate more. More reports, more analyses, more email drafts. The problem is that someone has to read, review and absorb all that output. It’s like an eight-lane highway that ends at a single tollbooth: a lot is created, but people, and above all the organization, pay for the bottleneck.
The solution is not to use less AI, but to change where and how it appears. Useful AI understands what you need and returns the answer or executes the action, instead of producing one more document to add to the pile. In the Microsoft ecosystem, this translates into assistants integrated into the ERP itself and into Power Platform that act on the business’s real data, not into a separate app that forces you to jump from system to system.
Context: AI understands the world, but not your company
A generic model knows a lot about the world and almost nothing about how your company works. And a company runs on two types of knowledge: the one in the system of record (transactions, master data, process logic that lives in the ERP) and tacit knowledge (emails, chats, unwritten rules, criteria that only exist in people’s heads).
A seemingly trivial question like "which suppliers can I order this material from?" actually requires knowing purchasing, supply chain, compliance and finance all at once. If AI only sees the transactional part, its answer can be technically correct and at the same time wrong in practice, because it ignores that "we only buy from supplier A unless there’s an exception approved by management". That’s why the prior work matters so much: organized and governed data, well-modeled processes and explicit rules. Here Business Central works in your favor, because it concentrates finance, purchasing, inventory, sales or projects in a single system, and gives AI a reliable ground on which to respond with your business’s context, not just theory.
Prediction: a business decision is, almost always, a prediction
"Will the order arrive on time?", "do I reorder stock of this product?", "what default risk does this customer have?". These are prediction questions, and they rely on structured, tabular data. Here we should be honest: language models are excellent with text, but they are not the ideal tool for calculating reliable forecasts on numerical tables. And the traditional approach—extract data, pass it to a specialist and wait weeks—usually arrives late: by the time the analysis is ready, the question has already changed.
The practical conclusion is that forecasting and risk assessment should not be an individual trick, but a system capability, available to whoever decides. Combining the ERP with analytics tools like Power BI and models suitable for structured data allows an operations manager to ask the question and get an answer backed by the company’s real data, without depending on a specialist bottleneck.
How we ground this at Tisa
There is little point in chaining three concepts together if they don’t translate into concrete steps. Our approach with SMEs and private companies is this:
- Start with a measurable use case, not with "putting AI in place". A real problem, with a before and after that can be compared.
- Organize and govern the data before asking the model for miracles. Without clean data, AI hallucinates or confidently gets things wrong.
- Integrate AI where people already work: inside Business Central and Power Platform, to reduce intensity instead of multiplying it.
- Measure the impact on the whole, not just one person’s convenience.
We have been implementing management solutions since 1987 and have supported companies in retail, distribution, construction, the food industry, sports centers and hospitality. That sector experience is what allows us to distinguish the use case that moves the needle from the one that only generates noise.
Do you want to move from individual AI to AI for the whole company that shows in your results? Let’s talk with no obligation: call us at (+34) 971 305 885, write to us at info@grupotisa.com or visit grupotisa.com. We’ll study your case and tell you, frankly, where to start.
Source of inspiration: news.sap.com, "How to Optimize AI for the Entire Enterprise, Not Just the Individual" (Florian Kunzke, 2026). Interpretation and approach are Tisa’s own.