AI You Can Bet Your Business On: How Mars' Context-Lake Builds the Trust Bridge Between People and AI
Can you truly bet your business on AI? For most, the answer is no. LLMs excel at stand-alone tasks, but their "black-box" guessing is a liability for core processes. Our experience at Mars proved this: in a complex supply chain, 98% accuracy per step still leads to systemic failure. This creates a conflict where employees don't want an AI that guesses, they need an AI that knows by learning from their expertise.
This talk reveals the architectural shift required to create AI you can bet your business on. We introduce the "Context-Lake," our innovative architecture, demonstrated through a Mars case study, that builds a "Trust Bridge" between people and machines. By capturing explicit business logic in a knowledge graph, we move beyond the limits of RAG. This approach de-risks enterprise AI , transforming it from a probabilistic liability into a deterministic asset that enhances employee collaboration and delivers reliable outcomes (opens the black box and kills halucinations).
Intended Audience an key takeaway:
For leaders: wonder why AI is so hard to embrace in business while it is so successful for consumers? most AI is actually great guesswork - this is not the language most large businesses speak. By moving AI form a black box to a glass box and putting your company context into a glass box too, you can eliminate the fear factor of "what if it guesses wrong" thus AI progress becomes as intuitive and symbiotic as google maps.
A framework to articulate the business-impact limitations of LLMs to senior leadership.
Enterprise Architects: a key peace to unlock AI is the new concept of a contextlake. A place where company context is stored, validated by employees and readable by AI. With AI maintenance the once complex graphical database can do this job for you.
company AI developers: Agents do not just feed on data and context windows they need to be connected to the context of the company. This is what a graphical database can do for you with the right ontology. It also avoids hallucination and ensures that in the big context of a big company the right answer is found for the right employee.
project managers: the real business value of AI is not taking over a task by beating the employees at their own game but by mutual development where AI learns from the human and the human from the AI. This builds trust, efficiency and a lasting insight on how to be successful.
Also practical first steps to get there are shared.
My goal is to take you on a journey from a common, high-stakes problem of not trusting the guessing black box of AI in my day to day work to a practical PoC that proofs a solution that we are ready to scale at MARS.
Part 1: The High-Stakes Reality
The Hook: I'll start with a direct question: "We all agree AI is the future, but are you willing to bet your business on it?" This immediately frames the real-world stakes.
My Credibility: You'll learn I lead digital transformation for 500 planners at Mars and before that globally at Unilever supply chain. This isn't a lab experiment; it's about applying AI to a complex global business where mistakes have consequences.
The "Aha!" Moment: I'll introduce the "2% Problem"—showing how 98% accuracy per step is a recipe for failure in a complex ecosystem. This reveals the hidden fragility of most AI strategies and a key reason in my experience why AI is not trusted at the workspace in many companies.
Part 2: The Core Conflict
Today's AI: We'll explore the "brilliant guesser" model of AI and how current practices (RAG, bigger models, contextengineering, ... ) are just improving the guesswork, not eliminating it.
The Bottleneck: I'll distinguish between simple tasks where guesswork is fine and core processes where it's dangerous, letting you see your own company's challenges.
The Human Problem: We'll confront the conflict: AI competing with your experts, not collaborating. This creates a slow, painful adoption path where employees have no incentive to help an AI that might replace them. In this scenario AI has to work very hard to outsmart the employee by just guessing how things are done based on the data in the datalake and some unstructured documents on a sharepoint.
Part 3: A Proven Solution
The New Concept: I'll introduce the "Context-Lake"—a structured knowledge reservoir that both humans and AI can read, understand, and collaborate on. I will share the high level architecture here as well how we are combiing: agents, datalake, a graphical database and intelligent process tools like Celonis, Sygnavio and different workflow tools.
The "How-To": You'll learn the simple principle behind it: capturing the Why, Who, How, and What of any process in a mindmap (for humans) that becomes a knowledge graph (for AI).
The "Why Now": I'll explain how AI makes this practical today, turning the knowledge from meetings and interviews into a "digital twin" of your operational brain incl a high level design of our solution combining Gemini technology with a seperate graph database.
The Trust Bridge in Action: You'll see a real Mars example (carton thickness) where we captured the logic of two different experts. I'll show how making the how of our processes explicit in this "glass box" builds trust and delivers business value like speed, quality, efficient human connects,...
Part 4: Your Actionable Takeaways
Clear Lessons Learned: You'll get three concrete principles from our journey:
-Trust & Consistency is at the foundation of doing business - guesswork is not the ideal in that context
-Contextlake is the essential visual Bridge between the current reality of company employees and the AI support of the future. (the mindmap-to-graph pipeline)
-Experts are no subjects - Empower Experts as Teachers through fast low cost and intuitive feedback loops.
Conclusion & Q&A:
20 years of digital end to end supply chain experience at Mars, Unilever and a direct to consumer company.
Currently: Center of Excellence Planning (Europe) Director at Mars (M&M, Bounty, Twix, Celebrations,....) responsible for the digital transformation of our planning processes from supplier to customer.