She Was Right About Chatbots

Last month I sat across from a senior enrollment leader — 35 years in higher education — on a discovery call.

A few minutes in, she stopped me and said something I hear constantly:

“I’ve had terrible experiences with chatbots. Every single time I interact with one, it’s never been good.”

Then she added the part most vendors hate to hear: “Vendors promise the world and rarely deliver.”

I didn’t push back.

I told her she was right.

Chatbots Were Never Built to Help You

Here’s the history most people don’t know.

Chatbots were created by big tech companies to solve a big tech problem: too many support tickets, not enough support staff. So they put a widget on the website to create the illusion of support.

You know the experience, because you’ve lived it.

You click the bubble. It asks which of three options you need help with. None of them are your problem. You pick the closest one anyway. Three menus later, you hit a dead end — or the digital equivalent of a shrug: “We’ve created a ticket for you. Someone will respond by email.”

That’s not support. That’s deflection with a smiley face.

And it’s not a flaw in the design. It is the design. Chatbots were built to keep you away from a human, not to get you to an answer. The industry metric they’re optimized for is literally called “deflection rate.” I’m not making that up.

So when a leader tells me every chatbot experience she’s ever had has been bad, she isn’t being cynical. She’s describing a product working exactly as intended.

A Different Species in the Same Skin

Here’s where it gets confusing: conversational AI looks like a chatbot. Same little window. Same text box. Same skin, same wrapper.

It’s not even close to the same thing.

A conversational AI agent, that's what we built at FRANSiS™, we call them AI Powered Helpers is different in kind, not in degree:

1. It knows your institution, not a script. It’s fed as much of your organization’s knowledge as possible — programs, policies, deadlines, the answers your best staff member would give.
2. It reasons. Because it’s powered by a large language model, it doesn’t match keywords to a decision tree. It actually understands the question — including the messy, three-part, didn’t-fit-any-menu version.
3. It holds a real conversation. Ask a follow-up. Change direction mid-thread. Ask the thing that isn’t on any menu. It keeps up, in real time.
4. It’s built to reach resolution, not create a ticket. The goal is that the person walks away with their answer — at 2 AM, on the device already in their hand.
5. It measures success by resolution, not deflection. The exact opposite incentive of the thing it resembles.

Don’t Take My Word for It. Run the Test.

Back to that leader’s deeper objection: vendors promise the world and rarely deliver.

She’s right about that too. So don’t trust the label, including ours. Test it:

• Ask it a question that isn’t on the menu.
• If the conversation ends in a ticket, it’s a chatbot.
• If it ends in a resolution, it’s conversational AI.

That’s the whole test. It takes thirty seconds, and no vendor can fake it.

By the end of our call, the same leader who opened with “I’ve never had a good chatbot experience” told me: “It is absolutely what we need.”

She didn’t change her mind about chatbots. She was right all along. She’d just only ever been shown the wrong species.

Your turn: think about the last time you used a chatbot. Did it solve your problem — or file it?

Learn about FRANSiS™ Open Door →

‍