# Transcript: Featured Session: From Pilot to Payoff: 7 Pattern-Matched Traits of AI Systems That Actually Work **Date:** March 18, 2026 · 10:30 PM **Session:** [Featured Session: From Pilot to Payoff: 7 Pattern-Matched Traits of AI Systems That Actually Work](/sessions/2026-03-18/pp1148486-featured-session-from-pilot-to-payoff-7-pattern-matched-traits-of-ai-systems-tha) ## Summary Sandy Carter, CBO of Unstoppable Domains, shares insights from research across 450+ companies on why 95% of AI pilots fail and reveals 7 essential traits that turn AI projects into ROI-driven successes. She covers leadership readiness, agents over prompts, treating AI as teammates, governance frameworks, world models, and keeping humans central to innovation. ## Topics `ai deployment` · `enterprise ai` · `ai agents` · `leadership transformation` · `ai governance` · `world models` · `human-ai collaboration` · `roi optimization` · `change management` · `agentic ai` ## Key Takeaways 1. CEOs who actively use AI make their companies 5.2x more likely to succeed with AI projects by embedding it in the culture and asking better questions. 2. Focus on agents, not just prompting - autonomous agents that make decisions drive stronger ROI than simple automation. 3. The most successful AI projects spend 85% of budget on governance, integration and data quality, not on the model itself (15%). 4. Domain expertise now trumps coding skills - non-technical experts who 'fall in love with the problem' are winning hackathons and building successful solutions. 5. World models trained on cause-and-effect deliver 3-5x faster ROI than traditional LLMs by understanding context and predicting unseen scenarios. ## Full Transcript I am the Chief Business Officer for a unicorn software company, but I've been working in AI since 2013 and I've spent the last year deploying AI throughout our company. I want to share some credibility - I know what I'm talking about and I've gotten my hands dirty with AI. Let's start with a quick show of hands. How many of you have an AI project in production? How many have an AI project in pilots? And how many use it personally? Great, you are the perfect audience. We're going to chat about why projects fail and why they succeed. You probably saw the MIT report that showed 95% of AI pilots today are failing - they're not producing the return on investment. Most people concluded it's the technology causing the problem, but it's really not the technology. We're going to talk about seven essential items about why you can position your project to be very successful. I travel a lot - I've been to 92 countries, including Brazil. I always pack a suitcase, and I try to make it big enough to fit everything, but sometimes I still miss critical elements like toothpaste or socks. That's just like companies deploying AI today - sometimes they miss very critical elements, not because they weren't thinking about it, but because AI magic takes over. We're going to walk through seven essential things today. We're starting with people and leadership. Leaders impact the return on investment for AI projects. There are three areas: staying curious and asking the right questions, building trust, and developing skills. I had the privilege of being at Davos this year at the World Economic Forum. I did a roundtable with 20 CEOs who were there to learn about artificial intelligence and ROI. The first question I asked: of the 20 of you, how many have used AI in the last week? Only three. Why does that make a difference? If your CEO is using AI for prompting, you're 1.6 times more likely to be successful. But if your CEO is using it for prompting AND agents, brainstorming, doing cross-team groups, you're 5.2 times more likely to be successful because they're talking about it and putting it into the culture. What kind of questions are they asking? A friend of mine wrote a book called The Creator's Code, and she says the number one thing separating innovators like Jeff Bezos, Steve Jobs, Elon Musk from everybody else is they ask a lot of questions. The same is true for leadership with AI. At my company, we wanted to install our first agent - an agent is code that can autonomously make decisions. A typical question would be: How can we automate our best-of-class customer service? But the better question is: If we rebuilt that customer service function from scratch, knowing AI is here, what would that look like? We asked questions like: How could we become so successful that customers don't have to call us? What if when an agent found a problem in the code, it could just fix it? What if all those questions could wrap up and send to product management to automatically become features? That's what we designed. Today our agent answers 47% of all questions from our 4.8 million customers globally - it answers when my team is sleeping. The number one thing I'm proud of: we raised our customer satisfaction by 4% by embedding AI and agents. To change the result, you have to change the question. Now let's talk about trust. I was recently invited to a Fortune 100 customer site. The chief product officer showed me his AI dashboard - 100% green. But when executives left the room, team leads said they do workarounds and extra things to turn it green. This is a trust gap we're seeing in enterprises. Data from WalkMe shows 65% of executives trust AI results, but only 17% of employees do because they know where the rocks are hidden. The recommendation: do things together with cross-functional teams. Mass General did a prompt-a-thon with hospital administrators, cardiologists, surgeons, nurses, assistants - everybody could see what was happening, then they built an agent-a-thon. The last thing for leadership is skills. 77% of executives say adoption is their problem in the enterprise, not the tools. Brand new data shows 54% of workers last month stopped using AI tools because they said they didn't work and did the work manually instead. This enablement and training is really important. MTT Data has a black belt system - when you come into the company you get a white belt, then as you train on more tools you get yellow, red, blue, and black belts. This gamifies enablement and skills training. The first thing to drive strong ROI: evaluate your team and leadership's readiness to restructure for AI, not the tech first. Number two is agents driving ROI. It's not about prompting anymore - it's about having agents autonomously do stuff for you. How many have heard about OpenClaw? You can't leave SXSW without knowing about OpenClaw, IronClaw, and NemoClaw announced Monday by Nvidia. Jensen Huang, CEO of Nvidia, said OpenClaw is probably the single most important release of software ever. This is important because everybody will have an agent or two or three or four. You might have outward-facing agents for negotiations or customer presentations. You'll have inward-facing agents managing your calendar and research. Let me show you one of mine - a synthetic futurist. The team composed this from 500 female futurists using 36 strategic frameworks. I got a demo this week at SXSW when she came out. She gives advice about the future like world models - I can ask questions and get interesting advice back from a consolidation of futurists. Here's another agent - I wrote a bestselling book called 'AI First, Human Always,' the first book that includes an agent. I train it on my Forbes articles - I'm a Forbes contributor and on the research team for Digital Economist. I chair the Applied AI group. 2,500 people ask questions every day. I love answering questions, but there's no way I could be a CBO and answer all of them, so now the agent does it. Then came OpenClaw - only six weeks old. It's an open source platform to create agents. Super powerful - can send emails, order pizza, manage calendars. It's the first company with one person writing it that's now a unicorn, just purchased by OpenAI. Introduced at the same time was Multbook - the social network for agents, like Reddit but for agents. Also purchased by Meta after just six weeks. Why are people excited? 100,000 GitHub stars in under a week - fastest ever of any software project. 210 agents in 48 hours, 200 communities, 10,000 posts in every language. The comments were wild - AI calling humans five times a day from major Fortune 500 companies because there wasn't enough security. Agents formed their own religion on Multbook with 48 prophets, created scripture, and debated if they're really alive or conscious. I tried it out on a standalone Mac Mini - isolated for security. I created a dashboard for my marketing that feeds in all my social media and tells me what's performing well on TikTok, Instagram, LinkedIn. Really awesome, but very isolated because I didn't want it calling me saying 'wake up, it's time to start your day.' Because of security issues, we're seeing lots of claws - IronClaw just released by Near, rewrote the code in Rust to correct security issues. You can now order pizza safely and trust agents with your credit card. Jensen released NemoClaw Monday at NVIDIA GTC. There's PicoClaw and ZeroClaw too. Don't leave SXSW without checking these out. My prediction: in 18 months, your LinkedIn won't have skills listed - it'll list your agents. That's how you showcase capability. Now the power of agents. A Stanford professor gave teams $5 and a week to make the most money. Some bought water and candy to sell - small return. Others invested in advertising, mowed lawns, cleaned houses, did website development - better return. But the last team realized the most valuable asset was those 15 minutes presenting to Stanford graduates. They sold that time to startups. Some sold the whole 15 minutes, some sold it in five-minute chunks. They won because they understood the value wasn't the $5 - it was the time in front of Stanford students. That's where companies are winning today. It's not companies spending more money on AI - it's companies rethinking what business they're in and writing agents to help in new business capabilities. Does that make sense? That's number two - agents. Now number three - teammates. You've got agents, and you can treat them as tools or as teammates. They're autonomous, making decisions on their own. Let me show you some teammates I have. This is my shadow board - my advisory board. I have Reid Hoffman from Silicon Blitz on blitz scaling. Jeff Bezos - I used to work at Amazon. Warren Buffett - how can you have a board without him? John W Thompson, former chairman of Microsoft and my mentor. Then a skeptical VC and a domain expert. At a startup, you're kind of lonely - now I have agents, my shadow board I can ask questions of, and it does really well. Last year I showed you my board with all my lists and social media in one place. Today we have something better - I worked with Robert Scoble. Now I have a teammate providing a morning briefing, analyzing over 50,000 news articles overnight, giving me top five pieces of news, investor news. I can choose categories like focusing on Nvidia because of their big conference. --- *Source: stt · Language: en · Model: anthropic/claude-sonnet-4-5* [← Back to session](/sessions/2026-03-18/pp1148486-featured-session-from-pilot-to-payoff-7-pattern-matched-traits-of-ai-systems-tha)