In a previous post, I laid out a crawl, walk, run framework for putting AI large language models to work in your business. The opportunity is real, and the tools are more accessible than ever for small and mid-sized businesses.
As you proceed down that path, consider this quote from noted quality expert and statistician W. Edwards Deming: “A bad system will beat a good person every time.”
Most IT decisions are also process decisions. What role will AI play in fixing or improving your business processes? Does simply giving everyone a productivity tool solve all your problems? When’s the last time you switched systems, and everything went perfectly? The fundamentals are still true. Process changes need to be managed so they don’t create short-term chaos and long-term systemic problems. Here are a couple of things to watch out for.
Garbage In, Garbage Out – Faster
ChatGPT and similar tools use generative AI. Very simply put, it uses your input and its data to predict a good answer. The output will always look clean, professional, and confident even when it’s wrong.
Incomplete or wrong information early in the process creates errors downstream. In a small business only one or two people might carry the critical data in their heads. Your institutional knowledge has quietly been doing the heavy lifting to get work done. Those gaps do not disappear when you add AI.
A realistic scenario: a team uses AI to generate client-facing summaries faster. The summaries look polished. But the inputs feeding the process were inconsistent to begin with, and the AI has no way to flag that. The team produces more summaries, faster, with the same underlying gaps. The problem scales before anyone notices.
How to fix it:
Before you add AI to a process, map the process. Identify where the real inputs, handoffs, and quality standards live. AI works best when it is given a well-defined job inside a well-understood process.
New Tool, New Failure Points
When you change the process, you might be creating new opportunities for mistakes.
The most common new failure point is over-trust. AI tools write well, summarize quickly, and sound authoritative. They generate plausible responses based on patterns in data. When the output is wrong, the only warning label is the ubiquitous footer at the bottom of the chat window. “LLMs may produce inaccurate information about people, places, and facts.” That’s a pretty broad disclaimer.
A realistic scenario: an employee uses AI to draft a response to a customer question. The output looks thorough and professional. They send it without checking. The answer is wrong. The customer notices before they do. Just put “ridiculous AI disaster” into your favorite search engine if you need more examples.
How to fix it:
Ensure the right verification exists in the workflow before you scale. Generative AI output is a first draft, not a final answer. Define explicitly where human review is required and make that expectation part of the process, not an afterthought.
Before You Scale AI Use: A Short Checklist
- Do we understand the process AI is supporting? If you cannot describe the steps clearly, you are not ready to automate any of them.
- Have we defined what “good output” looks like? Your team needs a standard to verify against, not just a prompt to copy.
- Is there a verification step before AI output reaches a customer, partner, or decision maker? If not, build one.
- Do we know what data our team is putting into these tools? Public AI tools are not the place for client data, financials, or sensitive internal information. If your team does not know the line, draw it.
- Are we measuring whether AI is actually saving time? Adoption is not the goal. Useful adoption is.
What This Means in Practice
AI is not going away, and the businesses that learn to use it well will have a real advantage. But “using it” and “using it well” are different things. The gap between the two is usually not the technology. It is the process, the training, and the guardrails around it. The businesses that navigate this well usually have someone in their corner who understands both the technology and the process it supports.
If your team is already using AI tools and you are not sure how, what for, or whether the output is being verified, that is worth a conversation with your IT partner before it becomes a problem.
