Most manufacturers know AI matters. The harder questions are where to start, what their data can support, and which projects will actually pay for themselves.
I look at readiness across the whole product lifecycle: concept and engineering, prototyping and testing, new product introduction, the supply chain, production and the business systems that tie them together. Here is how I approach it and what I have learned taking products from zero to scale.
Can engineering, PLM, ERP, production, quality and service data be accessed, trusted and combined? Is anyone building it into data products AI can use?
Where do small errors, delays and manual work add up across the lifecycle: design and testing, new product introduction, quoting and order entry, scheduling, quality and maintenance?
Do teams trust the data and have the skills to use AI tools, and who will own them once the project ends?
Which opportunities pay off fastest, what they cost, and how the return will be measured.
Begin with the customer and the business: the problem worth solving, the constraints, and what success should look like.
Follow the work from engineering to production, interview the people doing it, and review systems, data and workflows firsthand.
Rank use cases by value, feasibility and risk, and identify the data and skills gaps to close first.
Quick wins, pilots and longer-term projects, with owners, expected returns and a plan for the first 90 days.
Most AI efforts fail because no one is accountable for a business outcome. They start with a tool and hope for value. I start with the result that matters, such as fewer errors, faster launches or lower cost, and name who owns it.
The model is rarely the hard part. The work is in the data: getting it out of engineering, ERP and production systems, fixing its quality, combining sources, and shaping it into reliable data products that feed AI. That is where value is won or lost.
The fastest returns often come from unglamorous work in the handoffs between engineering, sales and operations. My team deployed AI agents that catch ERP order-entry errors before they reach the shop floor.
AI in daily use needs upkeep and an owner. People act on its recommendations only when they can see why and know when to override. Prototype to production means solving a real customer need, proving it with a prototype, then earning trust.