Beyond the Hype: Making AI Work in Manufacturing with Sebastian Chedal

In this insightful and practical episode, Lisa Ryan welcomes Sebastian Chedal, founder of Fountain City and co-founder of TestFox.ai. Sebastian helps executives implement AI strategies that actually work, focusing on one critical question: How do you join the 20% of AI initiatives that succeed instead of the 80% that fail? With 60% of his work in manufacturing and industrial sectors, Sebastian brings a grounded, practical perspective where implementation matters more than hype.

A Journey Through Digital Transformation

Sebastian's journey began in 1998 when he started Fountain City in the Netherlands. Over more than two decades, his work has evolved through network security, website and app development, creative projects, and ultimately into digital transformation with a focus on AI implementation—predominantly in manufacturing.

As a self-described generalist at heart with diverse interests, Sebastian has founded five businesses total (two non-profits that didn't make it), giving him an entrepreneurial track record that includes both successes and failures. This real-world experience informs his practical, results-oriented approach to AI implementation. Fountain City has been the anchor and core of his professional life, adapting and evolving as technology has transformed over the past 26 years.

The Catalytic Moment: Why AI Is Different Now

Sebastian draws a powerful parallel between today's AI landscape and the mid-1990s internet era, when people would ask, "What's a website? I don't need a website. Why would I need a website?" People didn't understand the benefits, how it worked, or how much effort it would take to implement.

Like many technological innovations, AI has finally reached a threshold catalytic point where it becomes truly useful, effective, and mainstream. The real breakthrough with large language models (LLMs)—what most people refer to when discussing AI today—is the ability to create qualitative automations, not just deterministic ones.

The Fundamental Difference

Deterministic automation (traditional): If this number is above this number, do this thing—straightforward logic gates we've had for decades.

Qualitative automation (AI-powered): Integration of nuanced, context-dependent decisions into automation processes, opening entirely new categories of automation.

This capability works at multiple levels:

1. Workflow automation: Eliminating time-consuming, mundane work like data transformation and entry that used to require hours or intern labor
2. Strategic support: Brainstorming, strategic planning, code planning, and design patterns
3. Knowledge work: Tasks requiring judgment, context, and understanding rather than simple calculations

The last year in particular has brought proposals and curiosity from people wanting to understand what it actually takes to put these systems in place—but the hype also leads to overestimation of capabilities and underestimation of implementation effort.

Becoming AI-Ready: The Foundation for Success

Sebastian outlines several critical dimensions of AI readiness that organizations must address:

1. Management and Strategic Vision

The wrong approach: "We need to make sure 30% of our processes are run by AI by the end of the year."

This mandate isn't inspiring and doesn't give teams something meaningful to rally behind, even if it's the directive from stakeholders or management.

The right approach: Transform mandates into meaningful vision:

1. "We're bringing in AI to help you do less of the time-consuming work that distracts you from the real work you want to be doing"
2. "We're implementing AI to help with knowledge retention and dissemination so the experts' answers reach more people's hands"
3. Focus on removing bottlenecks and freeing up people's time

2. Clear Long-Term Goals with Measurable Steps

1. Define what you're trying to achieve with AI long-term
2. Break it down into concrete steps with measurable ROI
3. Ensure each step has an actual, achievable outcome
4. Keep the focus as narrow as possible—projects that try to do too many things with AI often fail

3. Data Infrastructure

Start before you even have a project: Capture as much data as possible everywhere you can:

1. Record calls
2. Transcribe videos and podcasts
3. Store blogs and written content
4. Organize existing documentation

Without good data, you end up with generic inputs for AI systems. With rich data, you can plug services directly into it and create genuinely useful, customized solutions. Data can be organized, synthesized, or analyzed—but you must have it first.

4. Process Documentation and Formalization

This is especially critical in manufacturing, where Sebastian frequently sees companies with ambitious AI project ideas but:

1. Processes exist only in someone's head
2. The CRM system is "Bob's phone with all his contacts"
3. Sales approaches vary completely by individual with no standardization
4. No formal documentation of workflows or decision trees

The three-tiered AI implementation process:

Tier 1 - Data: Identify what data is needed and ensure it's being captured

Tier 2 - Process: Document, create, and formalize processes (healthy for business regardless, supporting legacy, growth, and consistency)

Tier 3 - AI Integration: Once processes are well-defined (ideally as flowcharts or UML diagrams showing logic trees), integrate the actual LLM components

Critical insight: Don't just throw AI everywhere. Use traditional automation for deterministic tasks and AI only where you need qualitative assessment. Using AI for math or deterministic systems can lead to significant issues.

Solving the Knowledge Retention Crisis

One of the most popular AI applications Sebastian sees in manufacturing is knowledge retention and dissemination. When you have people who've been with the organization for 20, 30, or 40 years with critical knowledge in their heads—or in "Mikey's CRM on his phone"—losing them creates devastating gaps.

Knowledge Capture Strategies

The right approach depends on where the data currently exists:

Email archives: One manufacturing client is leveraging 20-25 years of email conversations, using AI to analyze and create a knowledge springboard

Data synthesis: Use AI with structured inputs to generate data, then have subject matter experts review, correct, and essentially train the system on what's accurate

Existing documentation: Documentation may already contain answers but be inaccessible or unsearchable—AI can make it queryable and useful

Structured interviews:

1. Formal interviews recorded on calls
2. Self-guided prompts where the expert records video or voice responses on their own schedule
3. Adapt the method to how each person thinks and works best—make the process as smooth as possible

The key mindset shift: AI isn't here to replace jobs—it's here to ensure that the work someone has been doing for 30 years can continue for the next 30 years, preserving institutional knowledge and expertise.

Addressing AI Fear and Resistance

Sebastian has studied resistance extensively and emphasizes that education is the most important antidote to fear.

The less people understand about how AI works and what it does, the more they glorify it and see it as a threat. The more they understand it, the more they realize how to control and use it as a tool—just like a calculator helps a scientist do faster math equations.

The Self-Fulfilling Prophecy

The dangerous pattern: People who refuse to learn AI out of fear are actually at the biggest risk of replacement. If you didn't want to use computers when they were being invented because you feared job loss, but then jobs required computer skills, you couldn't get hired—and the thing you feared came true because you created it.

The reality: Jobs of the future will be influenced and affected by AI. There is a need to adapt, learn, and shift in certain domains. But understanding AI greatly reduces the risk of flat-out replacement.

The people honestly at biggest threat are those who are afraid and don't want to learn.

Keys to AI Project Success

Project Sizing and Scope

Timeframe: Something between 3 months to 1 year is ideal for AI sub-projects. Rarely do projects over 2 years make sense in the current AI landscape.

Breaking down big goals: You might have a long-term vision (like creating a virtual sales engineer who knows everything about your B2B products and can speed up customized quotes), but break it into achievable steps:

Step 1 (3 months): System answers Q&A emails after webinars. Initially, humans answer while training the knowledge base.

Step 2 (3 months): Launch internal/external product replacement lookup tool for specifications—could be conversational or simple search.

Step 3 (6 months): Expand to training new engineers with best practices and accumulated knowledge.

Step 4 (1 year): Eventually reach the goal of quote assistance for customized solutions supervised by engineers.

Each step has a concrete outcome and measurable value, building toward the ultimate vision.

The Scaffolding Principle

Sebastian emphasizes that scaffolding is crucial whether you're writing code or improving sales and marketing processes. Think of it like painting:

First: Fill the canvas with broad strokes and broad fields of color Then: Work on the detail

Many projects fail because teams focus intensely on detail execution without doing the scaffolding first. Define the logic flow and overall structure before diving into specifics.

Important AI characteristic: AI is a "pleasing system"—it always wants to please you in the way it thinks will please you. If you ask it to criticize your work, it will criticize because it thinks that's what you want. You must have the right methodology and approach in both project planning and implementation steps to work effectively with AI systems.

Unexpected Risks and Challenges

AI Governance

The zero-tolerance trap: Some companies have zero AI tolerance, telling teams they can't use AI due to legal risks, copyright concerns, IP protection, or process concerns.

Why it's dangerous: People bypass the restrictions anyway—using different browsers, personal laptops, or working from home. The water flows around your barriers.

The solution: Establish at minimum a governance policy framework:

1. How should people use AI?
2. How can they use it safely?
3. What's allowed and not allowed?

Create control without being so tight that people work around you. Balance protection with practicality.

Change Management

AI projects are fundamentally transformation projects, which means they're change management projects. Several dynamics emerge:

Underlying issues surface: An AI project might be "the straw that breaks the camel's back" where a department that's felt ignored for two years suddenly expresses frustration—which has nothing to do with AI but is triggered by another major change.

Education and experience gaps: The difference in AI familiarity between upper management and the floor can create challenges. Sometimes upper management hasn't used AI yet (though they may be shy to admit it), while certain departments or individuals race ahead.

Assessment needs: Understanding where your business sits on the AI maturity spectrum is critical:

1. Pre-AI with no idea how it works?
2. Casual familiarity?
3. Advanced users in some departments?

The Weakest Link Principle

Sebastian recommends an AI readiness self-assessment where you rank yourself against different areas. Following the principle that "a chain is only as strong as its weakest link":

If your team has tremendous AI hunger and passion, racing ahead with learning and implementation, but you have zero governance in place—governance is your weakest link. Address it to catch up with where you're steaming ahead in other areas.

Critical insight: Even if you don't immediately address disparities, knowing they exist is valuable. You know what you don't know, instead of not knowing what you don't know—which is usually where projects fail.

Most project failures over 20+ years happen because someone didn't realize something they didn't know they needed to know.

Real-World Success Patterns

Sebastian emphasizes that 2025 saw knowledge retention as a particularly popular application, with multiple manufacturing companies working to:

1. Capture expertise before retirements
2. Distribute knowledge more widely for training
3. Remove bottlenecks created by having all answers in one person's head
4. Create systems that preserve institutional memory

The correlation Sebastian draws to social media adoption is instructive: companies initially had either zero tolerance or wild-west approaches until they developed policies and governance. Now it's part of the standard business lexicon. AI will follow a similar trajectory.

Getting Started: Sebastian's Recommendation

If you're at the very beginning: Focus on learning, awareness, and discussion. Recognize AI's importance and that it's not going away. Make time for this—it matters.

If you're further along: Conduct an AI readiness assessment to understand where you stand across all dimensions.

If you're ready to implement: Create a roadmap with proper planning, breaking projects into those 3-month to 1-year achievable milestones with clear outcomes.

Step one for everyone: Figure out where you're at, then have a good overview assessment of your business across all AI readiness dimensions.

Actionable Takeaways for Listeners

1. Start Capturing Data Everywhere Now
2. Before you even have an AI project, begin recording calls, transcribing videos, storing blogs, and organizing documentation. Data is the foundation—without it, you only get generic AI outputs.
3. Transform Mandates into Meaningful Vision
4. Don't just say "30% of processes must use AI by year-end." Frame it as "freeing you from time-consuming work" or "getting expert knowledge into everyone's hands."
5. Document Your Processes First
6. If your processes exist only in people's heads or vary by individual, formalize them before attempting AI implementation. Create flowcharts showing logic trees and workflows.
7. Keep Projects Narrow and Time-Bound
8. Target 3 months to 1 year for AI sub-projects. Break big visions into concrete, measurable steps with clear ROI. Avoid projects trying to do too many things at once.
9. Focus on Scaffolding Before Details
10. Like painting with broad strokes before adding detail, establish your overall structure and logic flow before diving into execution specifics.
11. Combat Fear with Education
12. The less people understand AI, the more they fear it. The more they understand, the more they see it as a controllable tool. Make education a priority across all levels.
13. Use AI Only Where You Need Qualitative Assessment
14. Don't throw AI everywhere. Use traditional automation for deterministic tasks (math, simple logic gates). Reserve AI for contexts requiring judgment, nuance, and qualitative decisions.
15. Establish AI Governance Early
16. Create clear policies on...