AI in Business Central: risks and governance requirements

AI in Business Central brings real opportunities, but it also introduces new responsibilities. As artificial intelligence becomes embedded deeper into ERP workflows, businesses need to start considering the possible operational, financial and reputational risks.

In our Introduction to AI webinar, we took as long a look at the realities and risks as we did at capabilities. This isn’t because, here at TVT, we believe that AI is something to avoid. Rather we believe that successful AI adoption depends far more on governance and context than on features.

Data security and trust do not disappear with AI

One of the most common concerns around AI in Business Central is data exposure.

Generative AI relies on context. This context often includes your financial data, customer information or internal documentation. While Business Central Copilot operates within Microsoft’s trust boundary and respects existing permission models, problems can arise when data is passed to tools outside approved platforms.

The real risk is not AI itself. It is whether data is being sent through third-party tools that bypass enterprise security controls. From an ERP perspective, there is a simple sense check: would you be comfortable sending that information to another system manually? If not, that information should not be going through an AI interface either.

Shadow AI is already happening

When organisations delay decisions on AI, we find that AI usage rarely stops. Rather, it simply moves outside of official visibility.

Your staff will use their personal accounts, browser-based tools or unsanctioned extensions if there is no approved alternative. This results in a governance gap. Data handling audit trails and cost controls all become degraded when AI adoption happens informally.

It is therefore important to provide your staff with clear guidance. You need to ensure staff know which tools are permitted, what data can be shared and where AI should not be used as a practical first step.

Hallucinations and cascading errors

AI systems can generate outputs that are confidently wrong. While this has improved since early releases, it has not totally disappeared.

The risk increases when AI moves from assistance to action. When that happens one single incorrect assumption at the start of an automated process can be amplified as agents loop through multiple steps, creating compounding errors.

In an ERP environment, it is therefore important to ensure that there are human validation and approval points in the process as well as limits on what AI can do autonomously. Early implementations work best when agents are supporting processes rather than totally replacing human oversight.

Cost is becoming a real consideration

Another reality that organisations are beginning to face is cost. Although many AI tools were launched with low or no usage fees, as adoption increases and infrastructure costs become clearer, AI pricing models are changing. It is very common now for tech companies to have consumption-based licensing for agents. As a result unmonitored automation can lead to unplanned and unsustainable expense.

When AI becomes part of a core business process, you need to ensure that you give it the same cost scrutiny as you would any other operational system.

Skills, judgement and human roles still matter

AI can accelerate work, but it also changes how skills are developed. When systems are drafing emails, analysing data and suggesting actions, your staff risk losing familiarity with the underlying processes. This matters most in areas where judgement, context and customer relationships are critical.

AI is most effective when it supports your people rather than replacing their experience and critical thinking. For many organisations, this means prioritising internal processes first and being cautious about putting AI directly in front of customers.

Creating a sensible path forward

Your adoption of AI in Business Central does not require a leap of faith; it deserves the same preparation you would give any new process or extension.

We find that the businesses who see consistent value in AI adoption in Business Central are those that focus on:

• Strong data foundations
• Clear permission models
• Defined approval points for automation
• Practical policies around tool usage
• Measured rollout rather than blanket enablement

 

See how this evolution is playing out in real Business Central environments

In our AI in Practice: Business Central webinar series, we walk through Copilot and agent capabilities as they exist today, explain what’s practical, and discuss how to introduce AI without losing control over cost, data or decision‑making.

Watch the AI in Practice: Business Central webinar series

 

Prefer to talk it through in the context of your own Business Central setup?

We’re happy to have a straightforward conversation about where AI makes sense, where it doesn’t, and what a sensible next step could look like.

Talk to our team about AI in Business Central.