After 50+ client conversations about AI, I learned that what clients ask for and what they actually need are rarely the same thing.
In the last year, I have had over fifty conversations with potential clients about AI. Business owners, executives, operations managers. Different industries, different sizes, different levels of technical sophistication.
Almost all of them asked for the wrong thing. Not because they are uninformed. Because the AI industry has done a terrible job of setting realistic expectations.
Here is what clients say they want, what they actually need, and how to bridge the gap.
“We want a chatbot.” This is the number one request. Some version of “we need an AI chatbot for our customer service / sales / internal help desk.” They have seen ChatGPT and they want one that knows about their business.
“We want to automate everything.” The second most common request. Total automation of some business function. Usually expressed as “we want AI to handle our entire [marketing / customer support / data entry] process.”
“We want to save money on headcount.” Sometimes said directly, sometimes implied. The expectation is that AI replaces people and the savings go straight to the bottom line.
“We want what [competitor] has.” They saw a competitor announce an AI initiative and they want to match it. Usually without knowing what the competitor actually built or whether it is working.
These are understandable requests. They also lead to disappointing outcomes when taken at face value.
The exact architecture, memory layers, and delegation patterns I use to run 50 agents across two businesses.
Get the AI Agent Blueprint →The chatbot request misses the mark because chatbots solve a narrow problem: answering questions that have known answers. If the client’s real problem is that their customer service team is overwhelmed, a chatbot handles the easy queries. But the hard queries, the ones that actually frustrate customers and consume staff time, still need humans. The chatbot reduces volume by 20 to 30 percent while creating a new problem: maintaining and updating the chatbot’s knowledge base.
The “automate everything” request misses because total automation of any complex business function does not exist yet. What exists is automating specific, well-defined steps within a process. The gap between “automate our marketing” and “automate the generation of our weekly email newsletter” is the gap between fantasy and deliverable.
The headcount reduction request misses because AI rarely replaces entire roles. It replaces tasks within roles. The person who spent four hours per day on data entry now spends one hour. They still need to handle the exceptions, verify the output, and manage the system. The saving is three hours of that person’s time redeployed to higher-value work, not one fewer salary on payroll.
The competitive matching request misses because the competitor’s announcement and the competitor’s reality are rarely the same thing. Most AI announcements describe ambition, not achievement. Building to match an announcement means targeting a moving goalpost that might not exist.
After the initial conversation, after we get past the buzzwords and the LinkedIn articles they read, the actual need usually falls into one of three categories.
They need visibility. The business generates data that nobody has time to analyze. Sales numbers, customer feedback, operational metrics, financial reports. The data exists. The insight does not. What they need is not AI in the flashy sense. They need automated analysis that turns data into decisions.
One client asked for a “predictive AI for sales forecasting.” What they actually needed was a weekly report that combined their CRM data with their pipeline data and highlighted deals at risk of stalling. No prediction model required. Just structured analysis that nobody had time to do manually.
They need consistency. Processes that depend on individual people vary based on who is doing them and how busy they are. Client onboarding is different depending on which team member handles it. Report formatting changes depending on who wrote the report. Follow-up timing varies depending on workload.
What they need is standardized execution of processes that should be the same every time. AI agents are good at this. Not because they are intelligent. Because they do not get tired, distracted, or creative with processes that should not be creative.
They need time. This is the most common underlying need. The client’s team is competent. They know what to do. They just do not have enough hours to do all of it. Manual tasks eat into time that should go to strategy, client relationships, or product development.
The solution is surgical automation of the most time-consuming manual tasks. Not “automate everything.” Identify the three tasks that consume the most time relative to their complexity, and automate those specifically.
My discovery process has changed significantly based on these patterns.
First call: no solutions. The entire first conversation is about understanding the problem. I ask about the client’s day. What takes the most time? What keeps falling through the cracks? What would they do with an extra five hours per week? What has been tried before and why did it fail?
I do not mention AI during this call unless the client brings it up. The goal is to understand the operational reality, not to sell a technology.
Second call: reframe. Based on what I learned, I reframe the problem in operational terms. “You said your team spends eight hours per week compiling the management report. Here is what that process looks like automated.” This is specific, grounded, and measurable. The client can immediately evaluate whether the reframed problem matches their experience.
Third call: proposal. The proposal addresses the reframed problem, not the original request. It includes the specific process to be automated, the expected time savings, the implementation timeline, and the cost. No buzzwords. No “AI-powered intelligent automation.” Just: this is what will change, this is how long it takes, this is what it costs.
The close rate on reframed proposals is significantly higher than when I used to take the original request at face value. Clients feel understood because the proposal describes their actual situation, not a generic AI pitch.
The turning point in most client conversations is when I say: “Based on what you have described, you do not need a chatbot. What you need is your weekly reporting automated so your team gets three hours back every Monday.”
Clients react to this in one of two ways.
Some are relieved. They did not actually want a chatbot. They wanted to solve a problem and assumed a chatbot was the answer because that is what they kept hearing about. When someone offers a simpler, more targeted solution, they appreciate the honesty.
Others are disappointed. They wanted the impressive-sounding project. The AI chatbot they could demo to their board. Telling them they need basic process automation is not what they came to hear.
The first group becomes good clients. The second group would have been bad clients regardless of what I sold them. The disappointment would just have arrived later, after the chatbot was built and failed to solve their real problem.
Being honest about what clients need instead of selling what they ask for is the single most effective sales strategy I have found for AI services. It loses some deals in the short term. It builds a reputation that wins more deals in the long term.
The newsletter covers the client discovery process in detail, including the specific questions I ask and why.
The AI industry sells capability. Clients buy outcomes. The gap between the two is where most AI projects fail. Close that gap by understanding the outcome first, then choosing the capability that delivers it.
Subscribe to The Signal
One issue per week. The biggest AI build of the week, the top five videos worth your time, and an operator's read on what actually matters. For builders who ship, not dreamers who scroll.
No spam. Unsubscribe anytime. Just The Signal.