Unlike the free labour, AI cost may be higher than anticipated. At Tenato, we were surprised to learn this week that many leading companies are forbidding their staff from using AI on the job.  We’ve been selectively using AI for minor tasks at Tenato, but always with a great deal of hands-on involvement in reviewing and analyzing the results. Since hearing about this situation gave us pause, we asked our AI expert, Alan Osborne, to shed light on the issue.

AI carries real price tags

AI is becoming part of everyday operations, but many small and mid-sized businesses are still looking at it through the wrong lens. The issue is not simply whether AI can save time or reduce staff workload. The real issue is whether the business can use AI safely, accurately, and profitably without creating hidden costs in mistakes, security exposure, poor content quality, compliance risk, and damaged customer trust.

AI can be useful. It can help teams draft, summarize, research, classify, analyze, and automate repetitive work. But AI is not free labour, and it is not a direct replacement for trained people. It does not own the legal, operational, financial, or reputational consequences of a bad answer. That responsibility still sits with the business.

And while small and mid-sized companies are using it no-holds barred, some of the world’s leading companies either forbid staff from using it (the case with Apple) or put strict guidelines in place (the case with Microsoft).

For today’s businesses, the question is no longer, “How many people can AI replace?” but “Where can AI safely assist, versus where do we still need human oversight?”

AI is not free labour

Many businesses start by assuming that AI is a cheap replacement for labour.

AI can speed up first drafts, summarize long documents, support research, and help employees complete routine tasks faster. But it does not understand your customers the way your team does. It does not know your internal context unless that context is supplied. It does not know whether a claim is commercially safe, legally acceptable, or accurate enough to publish.

That means the real cost of AI is not just the subscription cost or token bill.

The hidden costs are often bigger:

• The cost of employee review and fact-checking.
• The cost of correcting bad outputs.
• The cost of legal, IT, and compliance review.
• The cost of security controls and approved platforms.
• The cost of poor customer experience when AI gives wrong or generic answers.
• The cost of low-quality AI content damaging search visibility.
• The cost of operational decisions based on incomplete or false AI outputs.

AI mistakes and hallucinations create business risk

AI systems can produce answers that sound confident but are wrong. This is often described as hallucination or confabulation. In a business context, that risk matters because a wrong answer can become a flawed report, a misleading sales email, an inaccurate web page, a poor legal assumption, a bad quote, or an incorrect customer response.

The danger is not just that AI makes mistakes. The bigger danger is that AI can make mistakes at scale. One unchecked human error may affect one customer, one document, or one decision. One unchecked AI workflow can create hundreds of inaccurate pages, emails, summaries, product descriptions, or recommendations before anyone notices.

This makes human review more important, not less. A trained employee can question the context, check the source, understand the customer, and recognize when something does not feel right. AI may produce a fluent answer even when the facts are incomplete, outdated, or invented.

• Correction cost: staff still need to review, edit, and repair AI-generated work.
• Trust cost: customers lose confidence when they see inaccurate or generic information.
• Compliance cost: regulated or sensitive industries may face legal exposure.
• Operational cost: bad AI outputs can move through departments and influence decisions based on false information.

That means the safest business use of AI is not “publish faster.” It is “draft faster, verify harder.”

The bigger security risk: staff uploading your business data into AI

Employees often upload sensitive information into public AI tools without thinking about data retention, logging, training use, vendor access, jurisdiction, or security controls. This may include:

• Contracts and legal documents.
• Pricing models and margin information.
• Internal strategy documents.
• Sales data and customer lists.
• HR issues and employee information.
• Financial reports and forecasts.
• Customer complaints or support transcripts.
• Unpublished product, service, or marketing plans.

This means the AI now knows far more about the business than it should; and it will make that data publicly available to everyone, even competitors who may run queries. Serious data leak? You bet.

A practical AI risk scorecard for business use

Businesses should create a simple approval framework before AI usage spreads uncontrolled. The scorecard does not need to be complicated. It needs to be documented, repeatable, and understood by department leaders. Something like the following:

What businesses should do now

The goal is not to slow the business down. The goal is to make AI useful without letting cost, risk, data leakage, or poor-quality output spread uncontrolled.

1. Create an approved AI tool list by department and use case.
2. Define which data types are allowed, restricted, or banned from AI tools (like the table example above).
3. Separate low-risk AI tasks from high-risk or client-sensitive workflows.
4. Require human review for customer-facing, legal, financial, medical, or compliance-sensitive outputs.
5. Measure ROI by workflow, not just vendor subscription or token cost.
6. Audit AI-generated content before it is published at scale.
7. Train staff on hallucinations, privacy, prompt safety, and data-handling rules.
8. Require legal, IT, or compliance review before adopting free, low-cost, or foreign-hosted tools.
9. Document who owns the final decision when AI output is used in the business.

Conclusion: AI needs governance, not blind adoption

AI is not the problem. Unmanaged AI is the problem.

The companies that benefit most from AI will not be the ones replacing the most staff or generating the most content. They will be the ones that use AI with clear rules, human review, protected data, measured ROI, and strong accountability.

AI should make good teams more effective. It should not become an excuse to remove the people. Otherwise, it’s pure, unmitigated risk.

At Tenato, we have been very cautious about how AI should be used for our work, and our clients’. If you would like to chat more with us, feel free to book a free consultation.