As I start having conversations with clients about AI implementation, I’m finding a significant gap between what companies say they want and what they actually need.
The disconnect became clear after speaking with an AI implementation company recently. Their pitch centered on building chatbots to help employees find information faster. Search your documents. Query your knowledge base. Ask questions in natural language.
That’s a valid use case. But it’s not what most of my clients actually need.
The Meeting That Changed My Perspective
A few weeks ago, I sat down with a client who was convinced they needed an AI solution. They’d seen the demos. They’d read the articles. They were ready to talk to their documents.
“We want our team to be able to ask questions and get answers from our systems,” they told me.
So I asked a simple question: “What questions are they asking?”
Turns out, the questions weren’t philosophical. They weren’t complex analytical queries that required synthesizing information across dozens of documents. The questions were things like “Did this invoice get approved?” and “What’s the status of this ticket?” and “Has anyone followed up with this customer?”
They didn’t need AI. They needed their systems to actually talk to each other. They needed a dashboard. Maybe a few automated notifications.
I felt a little like a doctor telling someone who came in asking for surgery that they probably just need to drink more water. Less exciting, but considerably more effective.
The Irony of My Own AI Usage
Here’s what I find interesting about my own work. I use AI constantly, but rarely in the way most people imagine when they hear “AI implementation.”
I’m not deploying chatbots for my clients to interact with. I’m using AI to build the automations themselves.
When I need to create a webhook connection between two systems, Claude helps me get the solution right on the first try instead of spending hours debugging. When I’m building an n8n workflow that requires complex logic, AI assists in working through the edge cases. The automation that runs afterward? It’s often just straightforward conditional logic and API calls. No AI required in production.
The AI is the tool I use to create. It’s not necessarily the tool my clients interact with daily.
What Clients Say vs. What They Need
When a client says “we want to implement AI,” what they usually mean is one of these:
- “We want to stop copying data between systems manually”
- “We want approvals to happen automatically instead of sitting in someone’s inbox”
- “We want our systems to talk to each other without human intervention”
- “We want to reduce the busywork so our team can focus on higher-value tasks”
None of these require a chatbot. None of these require large language models running in production. They require well-designed automation: triggers, conditions, actions, and integrations.
The Chatbot Question
This isn’t to say chatbots have no place. They can be valuable when:
- Your team genuinely spends significant time searching for information across fragmented systems
- You have a large knowledge base that’s difficult to navigate
- Customer-facing support needs to scale beyond your current team capacity
But before jumping to “we need an AI chatbot,” it’s worth asking: would a better-organized knowledge base solve this? Would improved search functionality be enough? Would documentation that actually gets maintained address the root problem?
Sometimes the answer is yes, you need AI-powered search and retrieval. Often, the answer is that you need better processes and connected systems.
Where AI Actually Fits
In my experience, AI delivers the most value in three areas:
- Building automations faster – Using AI assistants to write code, design workflows, and troubleshoot integrations
- Processing unstructured data – When you genuinely need to extract meaning from documents, emails, or other text that doesn’t fit neatly into forms
- Decision support for complex scenarios – When the logic tree is too complex for traditional automation rules
For everything else, traditional automation often does the job better, costs less, and runs more reliably.
The Practical Takeaway
Before your next conversation about “implementing AI,” try reframing the question. Instead of “how can we use AI?” ask “what manual work is slowing us down?”
The solution might involve AI. But more often, it involves connecting the systems you already have, automating the workflows you’re doing manually, and eliminating the friction points that eat up your team’s time.
That’s less exciting than a chatbot demo. It’s also more likely to deliver real ROI.
And hey, if it turns out you actually do need a chatbot after all that? At least you’ll know exactly what questions it needs to answer.