Responsible AI starts with knowing what to do when something goes wrong

By Heidi J. Ellsworth

Clear processes and human accountability help roofing businesses put responsible AI into practice.

In my ongoing conversation with Brian Behe, chief technology officer of RIIG Technology, Inc., we continue to explore what practical AI adoption looks like in the roofing industry. One issue deserves more attention, and that is responsibility. As businesses decide where AI fits into their operations, they must also determine how they will respond when a tool produces an unexpected result.

Brian addressed this challenge directly. “Most people have a well-prepared answer for what responsible AI means. That is, at least until something goes wrong.” His observation raises an important question for roofing professionals. Does your team have a process for handling a problem or simply an expectation that the technology will work?

It is easy to discuss responsible use when everything is going smoothly. A company may agree that employees should review AI-generated information and protect customer data. But those expectations become meaningful only when employees know how to apply them during a busy workday—especially after an error has reached a customer or influenced a decision.

Brian’s perspective comes from experience with systems in which failures carry significant consequences. “I’ve spent a significant part of my career working on systems where failure isn’t just inconvenient; it matters in a critical way and directly impacts a person’s safety and security,” he wrote. That experience shapes his emphasis on establishing clear protocols before something goes wrong.

Consider an AI-assisted customer message that gives an incorrect project start date. The immediate question is who will catch the mistake and contact the customer. The broader question is how the company will trace the error to its source and prevent other customers from receiving the same inaccurate information.

An estimating workflow presents a different risk. If an AI tool helps prepare a proposal but omits information that affects the scope of work, employees need to know how to flag the omission and who has authority to approve a correction. A review requirement is effective only when the reviewer knows what to check and how to respond when information is incomplete.

These examples show why responsibility must become part of everyday operations. As Brian explained, “Responsible AI isn’t the values statement you write down. It’s what happens when something goes sideways. It’s how your organization responds under pressure.” That response must be clear enough for employees to act without guessing.

Responsibility begins with clear ownership. When an employee notices an unexpected result, a designated person should evaluate the issue and decide whether to pause the affected workflow. Employees also need a straightforward way to report concerns, particularly when they do not know how widely an error may have spread.

Brian drew a direct distinction between good intentions and an established process. “If those processes aren’t clear, you don’t really have governance. You have intent.” For roofing companies, governance can start with practical instructions that identify who reviews an AI output and who handles an exception.

Those instructions should reflect the work and its risks. An AI-generated marketing draft may require editorial review before publication. Information that influences a technical roofing decision should be reviewed by someone qualified to evaluate it. In each case, the level of oversight should match the consequences of an error.

Training should also cover what happens after an error is discovered. Employees need to know how to preserve the information required to investigate the problem and, when necessary, shift the affected work to an established manual process. Walking through a realistic scenario can expose gaps that are easy to miss in a written policy.

A company can learn from these situations by looking beyond the individual mistake. Was the source information inaccurate, or did the review process fail to catch a problem? Identifying what happened allows the team to improve the workflow before resuming normal use.

Brian’s message is that accountability must be built into how a system operates. “Because in reality, responsible AI isn’t about what you say you will do. It’s about how your system behaves when it actually matters.” For roofing businesses, the system includes not only the technology but also the people and processes around it.

As our conversation continues, contractors can use this issue to examine their own AI practices. Choose one workflow in which your team uses AI and walk through what would happen if the output were wrong. Make sure employees know whom to contact and how to keep the work moving responsibly. Then use what you learn to strengthen the process.