When AI gets it wrong, who takes responsibility?

AI is rapidly entering property businesses, but accountability cannot be outsourced to an algorithm. Governance is becoming as important as adoption.

• AI can touch tenant data, leases, payments and investment decisions, making governance a business risk issue rather than simply an IT function.
• POPIA obligations remain with property businesses when AI tools process personal information; adopting technology does not transfer organisational accountability.
• Effective AI governance needs clear ownership, human oversight and fast approval processes that enable innovation without exposing sensitive property and client data.

AI is moving deeper into property

Artificial intelligence is moving rapidly from experimentation into the day-to-day operations of South Africa’s property industry.

Agents are using it to create listings and marketing campaigns. Property managers can use AI to interrogate portfolios, analyse arrears and communicate with tenants.

Developers and investors are exploring it for feasibility studies, research and data analysis, while businesses are increasingly embedding AI into customer service, administration and document-heavy workflows.

The opportunity is significant. But so is the question that follows: When AI makes a mistake, exposes confidential information or generates an inaccurate instruction, who is actually accountable?

For property businesses, that question matters because the sector sits at the intersection of high-value transactions and highly sensitive information.

Property companies routinely handle identity documents, bank details, credit information, tenant applications, leases, payment instructions, property valuations, contracts and other personal or commercially sensitive information.

South Africa’s Protection of Personal Information Act (POPIA) places obligations on organisations processing personal information, including requirements around security safeguards and the responsible handling of data. Introducing AI into a workflow does not make those obligations disappear.

If an employee uploads confidential tenant information into an unauthorised AI platform, or an AI-enabled system is connected to sensitive company data without adequate controls, the technology may introduce a new point of vulnerability.

The real issue is therefore no longer simply: “Should our property business use AI?” It is becoming: “Who decides how we use it, what information it can access, and who is accountable when something goes wrong?”

That is why formal AI governance is becoming increasingly important.

Cross-functional representation is essential

AI governance cannot be left solely to the IT department. An effective governance structure should bring together the people who understand technology, legal obligations, customers, employees, finances and operational risk.

That can include representatives from IT and information security, legal and compliance, product development, client support, sales and marketing, finance and human resources.

Finance is particularly relevant as AI vendors increasingly move away from simple per-user subscriptions towards consumption-based pricing. A seemingly inexpensive experiment can become considerably more expensive as usage scales across a business.

HR also has an important role. AI may automate elements of people’s jobs, but human oversight remains essential where decisions involve accuracy, compliance, ethics or significant financial consequences.

The technology is therefore likely to change not only how people work, but what businesses expect from them. Omobolanle Adekola, Product Owner at MRI Software, says the businesses benefiting most from AI are approaching governance as an enabler rather than an obstacle.

“AI isn’t going to replace the people who run this industry; however, it is already changing what we expect of them.”

The challenge, Adekola says, is to create sufficient controls without suffocating experimentation.

“The organisations getting the most from AI right now are the ones treating governance as a growth enabler with guardrails, not brakes.”

This becomes particularly important when AI applications interact with platforms already containing large volumes of property and customer information.

“In a market like ours, where a huge share of listed REITs and residential portfolios already run on shared platforms, the risk of one ungoverned AI tool touching the wrong dataset is not a hypothetical scenario.”

Governance therefore creates a repeatable process for evaluating that risk.

“A properly constituted board turns that risk into a manageable, repeatable process instead of a potential crisis every time someone wants to trial a new tool.”

Representation should not be confined to executives either. Junior employees and middle managers often have the clearest view of how AI is actually being used in daily workflows, while senior leadership needs to ensure adoption remains aligned with the organisation’s strategy and risk appetite.

Businesses operating across several African or international markets face another challenge: one set of rules may not fit every jurisdiction. Data-protection and regulatory requirements can differ between countries, making local legal and compliance input important even where businesses adopt group-wide AI principles.

What a governance board actually does

An AI governance board should not exist merely to discuss technology. Its central function is to establish who can use what, for which purpose, with which data and subject to which controls.

One of its most important responsibilities is managing the intake and approval of new AI tools. Rather than allowing individual employees or departments to independently subscribe to platforms, a business can establish a structured process that asks several fundamental questions:

• What problem does this AI tool solve?
• What information will it access?
• Where will that information go?
• Does it integrate with other company systems?
• What will it cost?
• Who will monitor its outputs?
• What happens if it fails?

Not every AI application requires the same level of scrutiny. A tiered approval structure can therefore make sense. A low-risk productivity tool that does not touch confidential information may require relatively straightforward approval.

A more significant application involving cost, integration or operational processes should bring technology and finance into the decision.

At the highest level, an AI system interacting with personal information, third-party platforms, financial information or sensitive business data should receive scrutiny from information security, legal and governance, risk and compliance functions.

The objective isn’t to make AI difficult to use. It is to ensure that the level of governance is proportionate to the risk.

Where governance boards go wrong

Creating an AI governance structure does not automatically create good governance. Two opposite failures can undermine the entire process.

The first failure is becoming a rubber stamp

If every proposed AI application is approved without meaningful questioning, the organisation has created another administrative layer rather than an effective control.

A governance committee should occasionally send applications back for clarification, modification or additional safeguards.

The second failure is potentially dangerous: moving too slowly

If an employee has to wait six weeks for approval to use a relatively low-risk AI application, governance can inadvertently encourage exactly the behaviour it was created to prevent.

Employees may simply bypass the process and start using freely available consumer AI applications themselves. That creates the phenomenon increasingly described as shadow AI: employees using AI tools without formal organisational approval, visibility or controls.

In a property environment, that can become particularly problematic if staff begin placing tenant information, contracts, financial records or other confidential material into unauthorised platforms.

Speed can therefore be a governance control in itself. Organisations can establish different approval tiers and target turnaround times so that low-risk requests move quickly while genuinely sensitive applications receive deeper scrutiny.

Good AI governance needs to be rigorous enough to protect the business and fast enough that employees actually use it.

A formal board isn’t essential, but accountability is

Not every property business needs an AI governance board. A large listed property company, national agency group or major property manager may require a formal cross-functional structure.

A smaller property business may not. The governance function could instead sit within an existing risk, compliance, technology or management committee. But the size of the organisation does not eliminate the need for accountability.

Someone still needs to own the fundamental responsibilities: AI policy, tool approval, data protection, staff training, human oversight, security and monitoring emerging threats.

And those responsibilities should not simply be handed to one person in IT. One of the simplest starting points for a smaller property company is to create an approved AI-tool register.

Employees should know which platforms they are permitted to use, what information can and cannot be uploaded, which activities require human verification and who they should approach when they want to introduce a new tool.

Businesses should also establish clear rules around high-risk information. Tenant identity documents, banking information, contracts, confidential transaction documents and personal information should not simply be uploaded into public AI applications because doing so is convenient.

AI governance should enable innovation, not stop it

The property sector has good reason to embrace artificial intelligence.

Used effectively, AI can improve productivity, analyse information faster, reduce repetitive administration, strengthen customer service and allow professionals to spend more time on higher-value work.

But adoption without accountability creates a different category of risk. The critical question is no longer whether a property company uses AI. Increasingly, employees will use it regardless.

The more important questions are whether the organisation knows which tools are being used, what data they can access, who checks their outputs and who owns the consequences when something goes wrong.

For property businesses that have yet to establish any AI governance framework, the starting point does not need to be complicated.

Identify the people responsible for technology, compliance, finance, operations and employees. Establish what AI tools are already being used. Determine what data those applications can touch. Create an approval process. Train staff. And assign clear accountability.

Because ultimately, AI can generate the answer, recommendation or document, but it cannot carry the organisation’s responsibility for using it.