What a property business should do when AI gets the answer wrong

A property description includes a feature that was never confirmed. A meeting summary misses an important condition. A suggested shortlist contains premises that fail a stated requirement. These are useful scenarios for any property business planning to put AI into everyday work.

Joe Sokolic, founder and CEO of Sokolic Property Group, believes the response to an error should be designed before a tool becomes part of a client-facing process.

“We need a clear answer to a simple question,” says Sokolic. “When someone finds a mistake, how do we correct it and stop the same mistake travelling further?”

The first step is to understand the source. The underlying property record may be wrong. The system may have misunderstood a correct record. A user may have supplied incomplete instructions. Each explanation points to a different correction.

Consider an unverified statement about backup power appearing in draft marketing. Removing it from the advertisement addresses the immediate issue. The team should also establish whether the statement came from an outdated brochure, an ambiguous note or an unsupported generated claim.

If the source record remains misleading, the same statement may return in the next draft. If similar material has already been distributed, the business needs to identify the affected versions and correct the communication appropriately.

Sokolic argues that responsibility should remain with the business. A client has engaged a property adviser. The fact that a system assisted with preparing an answer does not make an inaccurate answer less consequential to the person relying on it.

“We should be able to explain what we checked and what still needed verification,” he says. “That is part of being accountable for the information we provide.”

He proposes keeping a practical record of recurring errors. The purpose is to identify patterns that can improve the workflow. Repeated confusion around availability may indicate a verification problem. Repeatedly omitted requirements may point to the structure of the client brief.

Staff should also find it easy to flag uncertainty without feeling pressured to accept a polished output. A confident tone is not evidence that a statement is correct. The review process needs access to the source information and time to examine material details.

The lessons should feed into training, templates and decisions about which tasks remain suitable for automation. Some outputs may need closer review than others because the consequences of an error differ.

For Sokolic, a mature approach to AI includes the ability to recognise its limits in ordinary work.

“Trust is strengthened when a business catches problems, corrects them properly and learns from them,” he says. “That is the standard our systems should support.”

First published on sokolic.co.za by Sokolic Property Group.

Get commercial property news by email

Free alerts when we publish — pick a focus or get everything, unsubscribe in one click. Subscribe »