I’ve seen a lot of pitches. After 200+ angel investments in companies like Anthropic, OpenAI, and Scale AI, you get a pretty good sense of what’s real and what’s just smoke and mirrors. And right now, the AI space is filled with a whole lot of smoke.
Everyone’s talking about conversational AI. Every startup wants a chatbot. Every enterprise wants a voice assistant. The hype is deafening. But here’s the dirty little secret that no one wants to talk about: most of it is garbage.
I’m not just being cynical. The data backs me up. Depending on which report you read, somewhere between 80% and 95% of all AI projects fail. Think about that. The vast majority of time, money, and effort being poured into this space is going down the drain. It’s a staggering waste.
As someone who has built and sold two tech companies and now invests in the next generation of founders, this trend deeply concerns me. I’ve seen firsthand how the promise of AI can be a siren song, luring entrepreneurs onto the rocks of failed projects and empty bank accounts. But I’ve also seen what it takes to succeed. And I can tell you that the difference between success and failure is not what you think.
The Seduction of the Perfect Demo
It all starts with the demo. The perfectly polished, flawlessly executed demo. I’ve seen them a thousand times. A founder stands on stage, has a seemingly natural conversation with their AI, and the audience is blown away. The AI understands every nuance, every colloquialism. It’s witty, it’s helpful, it’s… perfect.
But here’s the thing about demos: they’re theater. They are meticulously scripted and rehearsed. The questions are carefully chosen to stay within the narrow confines of what the AI can actually handle. It’s a magic trick. And like any good magic trick, the illusion is shattered the moment you try to replicate it in the real world.
I remember a few years ago, a startup pitched me their revolutionary new customer service chatbot. The demo was incredible. The bot handled complex queries with ease. It even cracked a few jokes. I was impressed. I was this close to writing a check.
But then I asked to try it myself. The founder’s face went pale. He started making excuses. “It’s not really ready for public use yet.” “We’re still working out a few kinks.” I insisted. And what I found was a buggy, frustrating mess. The bot couldn’t even answer the most basic questions. The demo was a complete fabrication.
This experience taught me a valuable lesson: never trust the demo. The real test of a conversational AI is not how it performs in a controlled environment, but how it performs in the wild, with real users, in all their messy, unpredictable glory.
The Real Reasons for the 95% Failure Rate
So why do so many projects fail when they leave the lab? It’s not the technology. The models we have today, like the ones from my portfolio companies OpenAI and Anthropic, are incredibly powerful. The problem is the implementation. It’s the strategic and design choices that are made long before a single line of code is written.
Here are the biggest mistakes I see over and over again:
1. Trying to Boil the Ocean
This is the number one killer of AI projects. Founders get so enamored with the technology that they try to build a bot that can do everything. They want it to be a customer service agent, a salesperson, a therapist, and a stand-up comedian all rolled into one. The result is a bloated, unfocused product that does nothing well. It’s the classic “jack of all trades, master of none.”
When we were building RemoteTeam, which was later acquired by Gusto, we initially fell into this trap. We wanted to build a single bot to handle everything from onboarding to payroll to performance reviews. It was a disaster. The project was too complex, the user experience was confusing, and we were burning through cash with nothing to show for it.
We had to make a tough decision. We scrapped the “do everything” bot and focused on solving one problem: making it easier for remote teams to manage their HR. We built a simple, focused product that did one thing exceptionally well. And that’s when things started to take off.
2. The Tyranny of the Script
Most chatbots are not much more than glorified decision trees. They follow a rigid, predefined script. If the user says X, the bot says Y. If the user says Z, the bot says A. There’s no room for deviation. No room for real conversation.
This approach is doomed to fail because human conversation is not a script. It’s messy, it’s unpredictable, it’s a dance. A good conversational AI needs to be able to lead and follow. It needs to be able to handle unexpected turns in the conversation. It needs to be able to improvise.
I once invested in a company that was building a travel bot. The bot was supposed to help you plan your dream vacation. But it was so rigidly scripted that it was impossible to use. If you didn’t answer its questions in the exact order it expected, it would get confused and you’d have to start over. It was like trying to have a conversation with a brick wall. The company eventually pivoted, but it was a costly lesson in the importance of flexibility.
3. The Empathy Gap
This is a more subtle, but equally important, reason for failure. Many conversational AIs are designed by engineers who are brilliant at what they do, but who may not have a deep understanding of human psychology and emotion. They build bots that are technically proficient, but that lack empathy. They are cold, robotic, and impersonal.
Think about the last time you had a really bad customer service experience. It probably wasn’t because the agent didn’t have the information you needed. It was because they didn’t seem to care. They were just following a script. They didn’t listen to your concerns. They didn’t make you feel heard.
A good conversational AI needs to be able to do more than just provide information. It needs to be able to connect with the user on an emotional level. It needs to be able to show empathy. This is especially true in sensitive areas like healthcare or finance. A bot that is helping a user manage their diabetes or their retirement savings needs to be more than just a machine. It needs to be a trusted partner.
How to Build a Conversational AI That Doesn’t Suck
So how do you avoid these pitfalls? How do you build a conversational AI that is actually helpful and engaging? Here are a few principles that I’ve learned over the years:
- Start with a real problem. Don’t start with the technology. Start with a real, painful problem that your users are facing. What are they struggling with? What are their goals? How can you use conversational AI to make their lives easier?
- Be a scalpel, not a Swiss Army knife. Focus on solving one problem exceptionally well. Don’t try to be everything to everyone. It’s better to have a simple bot that is loved by a small group of users than a complex bot that is hated by everyone.
- Design for conversation, not for computers. Think about the flow of the conversation. How can you make it feel natural and intuitive? How can you guide the user without being too rigid? How can you inject personality and humor into the interaction?
- Always have a human in the loop. No matter how good your AI is, there will always be times when it fails. And when it does, you need to have a seamless way to hand off the conversation to a human. This is not a sign of failure. It’s a sign of good design. It shows that you care about the user experience more than you care about showing off your technology.
The Future is Human-Centered AI
The future of conversational AI is not about building bigger, more powerful models. It’s about building more human-centered experiences. It’s about using AI to augment human capabilities, not to replace them. It’s about creating a world where humans and machines can work together to solve the world’s most pressing problems.
I’m incredibly excited about this future. I’m investing in founders who share this vision. Founders who are not just building technology for technology’s sake, but who are building technology to make a real difference in people’s lives.
The road ahead will not be easy. There will be more failures. There will be more hype. But for those who are willing to do the hard work of building real, human-centered products, the rewards will be immense. The 95% failure rate is not a destiny. It’s a choice. And it’s a choice that we can, and must, change.
Frequently Asked Questions
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.