When the Usual Advice Doesn't Fit
Automation advice often pushes for maximum AI involvement. But for Thales Serra Xavier, a developer working with Koube, a Brazilian company specializing in automotive chemical products, that approach created a safety risk.
He initially built a customer service automation using n8n that relied heavily on AI. The idea was to have an AI assistant collect customer information and answer frequently asked questions using the company's knowledge base. During development, the team noticed the AI was sometimes generating inaccurate information. Since the company deals with chemical products used in vehicles, even a small mistake in product recommendations or instructions could potentially create safety risks or damage a customer's vehicle.
"We decided to change our approach," Thales said. "Instead of having AI handle the entire conversation, we moved to the SuperSAC platform and implemented a more structured chatbot, using AI primarily to identify customer needs and direct them to the appropriate human representative."
That experience taught him a key lesson: "Successful automation isn't about using as much AI as possible. It's about knowing where AI adds value and where human expertise is still essential."
What the How-To Guides Leave Out
Thales points to a gap between building an automation and making it work in a real business. "Today, you can find plenty of tutorials teaching the technical side of automation, such as how to build workflows, connect different tools, and troubleshoot errors," he said. "But knowing how to make an automation work doesn't necessarily mean knowing how to integrate it into a company's daily operations or ensure that it actually improves an existing process."
Another challenge is the accuracy of information. When developing an automation, you often depend on information provided by different departments or employees. Sometimes that information is incomplete, outdated, or simply incorrect, which can cause problems in the automation itself. In customer service, for example, incorrect information about how to handle a specific request or transfer a customer to the right department can affect the entire workflow.
Maintenance also gets too little attention. "Tutorials usually focus on getting an automation up and running, but that's only part of the job," Thales said. "Automations can break, business processes change, and unexpected situations come up. That's why it's essential to have someone who understands the system and knows how to maintain, troubleshoot, and improve it over time."
To him, building an automation is just the beginning. "The real challenge is making sure it works reliably within the business and continues to deliver value over time."
Decisions Owners Make Too Quickly
One common mistake business owners make is assuming that AI should replace as much human work as possible. "They often want to automate everything at once, but that's not always realistic or even the best approach," Thales said. "Some processes still require human judgment, and I believe AI should be used to support employees rather than completely replace them."
Another mistake is underestimating how much time successful automation takes. Before implementing anything, it's important to understand how the business operates, identify its internal processes, and determine where automation can actually add value. From there, Thales prefers a gradual approach: implement a solution, test it, see how employees adapt to it, make adjustments, and then move on to the next process.
He also sees owners getting excited about powerful AI tools and rushing to implement them without clearly understanding what problem they're trying to solve. "Just because a technology is impressive doesn't mean it's the right solution for every business," he said.
For Thales, successful automation isn't about doing everything as quickly as possible. "It's about making the right improvements, at the right pace, while keeping both the business and its employees in mind."
Practical Takeaways
- Limit AI where errors carry real-world risk. In safety-sensitive industries, use AI to route and identify needs, not to deliver final recommendations.
- Treat implementation as a separate skill from building. Connecting tools is only part of the job; integrating with daily operations and verifying source information matters just as much.
- Plan for maintenance from day one. Automations break and processes change, so someone needs to own troubleshooting and improvement.
- Start small and iterate. Implement one solution, test it, observe how employees adapt, adjust, and then move to the next process.
- Define the problem before choosing the tool. An impressive AI tool is not a substitute for a clear understanding of what needs fixing.
More of Thales Serra Xavier's work can be found at thalesxavierwork.com.