From Idea to Production: What It Takes to Build Enterprise AI Programs That Actually Work
Interview with Melvin Manchau, Former Salesforce Enterprise Strategist and Founder of Broadwalk AI
When I talk to executives about AI, the pattern is clear: hype fatigue is setting in. Teams are tired of proof-of-concepts that go nowhere, consultants who promise too much, and tools that don’t scale.
So I sat down with someone who’s been in the trenches.
Melvin Manchau is the former head of Enterprise Strategy and Transformation at Salesforce. His work focuses on AI readiness, data governance, and change management for large, regulated enterprises. He's now the founder of Broadwalk AI, a venture focused on building enterprise-grade AI capabilities with measurable ROI.
We spoke about the silent failure points inside enterprise AI programs—and what it really takes to go from shiny pilot to production deployment.
The Problem: AI Programs That Never Leave the Sandbox
Esther Katz: What brought you to Real Talk AI today?
Melvin Manchau: I’ve been having the same conversation over and over: Why are enterprise AI programs not working? Everyone’s stuck in POC [proof of concept] land. They start strong, hire a vendor, run a workshop. And then nothing gets deployed.
The gap between vision and execution is massive—and it’s not a technical problem. It’s structural.
AI Readiness Is Not About Tools—It’s About Alignment
Esther: That’s something we see a lot: companies assuming they can "add AI" like a software plugin. What’s missing?
Melvin: The main problem is the way companies define “readiness.” They think it’s about buying infrastructure or hiring data scientists. But real readiness is about organizational alignment.
You need to answer: Who owns the outcomes? Who makes the decisions? Who’s accountable if things go wrong?
When those questions aren’t answered, you get siloed initiatives. One department tries something with an LLM; another team experiments with automation. But there's no central architecture. No consistent success criteria.
Esther: So it fails in governance, not engineering?
Melvin: Exactly. Most companies don’t have a proper AI strategy office. Or they confuse governance with compliance. Good governance is operational—it helps teams move faster, not slower.
From PowerPoint to Production: What the Best Teams Do Differently
Esther: You’ve worked on real deployments inside complex organizations. What differentiates the companies that succeed?
Melvin: The best teams operate like startups inside the enterprise. They move in 90-day cycles. They ship things, learn, adapt. But that only works if you have executive sponsorship and cross-functional trust.
Also: the data foundation matters. If your data is a mess, don’t expect AI to clean it up for you. That’s magical thinking.
Esther: Can you give an example of a successful AI deployment?
Melvin: One I led at Salesforce involved a global health client. We built an early-warning system to identify patient churn. Sounds simple, but it took four months just to align on data definitions. Then we ran iterative experiments—tight feedback loops with operations, not just IT.
The key success factor? A cross-functional “mission team” with legal, data, operations, and frontline staff in the room from day one.
Why AI Pilot Programs Fail (Even When the Tech Works)
Esther: Why do you think so many AI pilot programs fail to scale?
Melvin: Three reasons.
No ownership. AI projects float in limbo. IT owns the infrastructure, but the business doesn’t own the outcome.
Poor integration. You build a beautiful model that no one can use because it doesn’t plug into workflows.
No incentive alignment. If KPIs aren’t impacted, the business doesn’t care.
People underestimate how hard it is to integrate AI into live systems. You’re not just launching software—you’re changing behavior. That’s harder.
From POC Theater to Operational Maturity
Esther: You’ve said before that AI maturity isn’t about how advanced your models are—it’s about how much value they deliver. How do you measure that?
Melvin: Every enterprise needs to ask: What is the “AI operating model” inside our business?
Are we building one-off use cases, or do we have reusable capabilities?
At Broadwalk AI, we help companies move from "AI as a tool" to "AI as an operational function." That includes setting up governance structures, capability maps, and funding models.
If you don’t change how decisions are made, the technology will never get used.
What’s Next for Broadwalk AI?
Esther: Tell us more about your new venture. What problem are you solving with Broadwalk AI?
Melvin: We’re helping enterprises go from experimentation to adoption. That means building the architecture, policies, and training needed to scale AI responsibly. We’re working with organizations that don’t need more pilots—they need outcomes.
We’re building playbooks, frameworks, and accelerators that make AI operational. Less demo, more delivery.
Final Word: Advice for AI Leaders in 2025
Esther: What advice do you have for leaders trying to bring AI into their organizations this year?
Melvin: Stop looking for magic tools. Build cross-functional teams. Define ownership. Measure outcomes.
AI is not a project. It’s a capability.
The companies that win won’t be the ones with the most advanced models. They’ll be the ones who learn how to execute.
Esther Katz is a deep tech go-to-market strategist who takes breakthrough products from 0 to 1. She hosts Real Talk AI, where builders speak plainly about what it takes to implement AI in the real world. Subscribe for weekly conversations with the people driving real change in business and technology.