"I Need AI My Supply Chain"—Where the Hype Meets the Reality Check

By Greg Mueller, VP Customer Success Strategy, Optilogic

‍

Intrigued, I asked about the biggest problems identified.  Pause, silence.  “My company has a large initiative to deploy AI across the enterprise.”  This is a great microcosm for the state of Artificial Intelligence in our world.  We all inherently know AI will have a big impact, however the placement, magnitude, and speed of the change is still unknown.  The hype is real, so where do we go?  

‍

We see the change around us as cars begin to self-drive, answers to any question at our fingertips, and phone agents automated, etc.  The capital investment is unprecedented:  An estimated $500+ billion will be deployed in 2025 alone between VC + big tech, ballooning to an estimated $2-5T (with a “T”!) by 2030.  Lots of bets are being placed, and the promise is that AI will automate almost everything and improve productivity, resulting in company bottom-line results.  But what do we know today?

‍

‍AI Is a Powerful tool—Not the Tool

‍

Many of the investment being thrown at AI today are infrastructure builds and not yet generating bottom-line return for customers and consumers of AI.  According to Gartner CSCO Insights for 2025, 78% of 102 CSCOs surveyed say that AI will impact less than 10% of COGS in the coming year.  AI is awesome, but is it humans that are not ready to let go?

‍

Many of us are comfortable in a Waymo self-driving taxi, but these are trained on relatively small geographical areas and travel mostly at lower speeds.  Would you feel comfortable putting your loved ones in a self-flying plane (no human, just the machine) yet?  

‍

Companies are the same. They will use AI and automation on relatively low-impact decisions, reserving human judgement for the high-impact decisions.  So how do we make AI a practical investment for the supply chain?

‍

Where AI Creates Value Today + What We Are Missing

1. Prediction: Demand, price, and leadtime forecasting using machine learning to more accurately predict outcomes.

2. Anomaly detection: models spotting leakage, quality outliers and processes that are out of specification.

3. LLM copilots: translating business questions into data queries, scenarios, or solver inputs using natural language, even when data is messy.

4. Visual Inspection: receiving, counting, quality inspection through computer-driven optics.

5. Routing & dispatch: learning from driver behavior, traffic, weather, service constraints.

6. Rootcause triage: decision tree, event streams, logs, and tickets to cut timetoresolution.

7. Automation:  Low-impact/high frequency workflows and decisions  

‍

AI Is Not the Only Way to Solve Supply Chain Problems

The modern supply chain is a decision factory. Great factories use the right machine for each job.  Optimization and simulation sit alongside AI as powerful tools to solve certain problems.  These tools, individually or together, accelerate insight, but most highstakes decisions still require alignment to strategy, service and cost promises, and risk limits with humans-in-the-loop.

• AI automates low value, augments high value, and predicts patterns.

• Optimization chooses the best set of actions under constraints.

• Simulation tests how the system behaves under variability and time.

• Humans define problems, set intent, risk tolerance, and ethics—and own the outcomes.

‍

Treat these as equals in a single decision stack, not as competitors.
‍

The Other Two Engines That Matter: Optimization and Simulation

Optimization (the choice engine)

‍

• When to use: network design, sourcing, inventory placement, mode mix, flow pathing, production planning.

• How it works: linear/mixedinteger programming, constraint programming, decomposition, metaheuristics.

• Strengths: provable optimality bounds, traceable tradeoffs, constraint fidelity.

• Cautions: requires clean data and explicit objectives; can be brittle if you ignore uncertainty.

‍

Simulation (the risk and test engine)

‍

• When to use: congestion, variability, service level trade-offs, labor/asset utilization, policy stress tests.

• How it works: discreteevent, agentbased, Monte Carlo; captures time, queues, and randomness.

• Strengths: shows how and why performance shifts; ideal for resilience and "whatif" shock tests.

• Cautions: results are scenariodependent; needs disciplined experiment design.

‍

Combine all three: Use AI to predict distributions, feed those into optimization to select actions, then simulate to test robustness before committing.

HumanintheLoop by Design

‍

• Intent and guardrails: Define objectives, business rules, service tiers, and risk tolerances upfront.

• Escalation paths: Autonomy for lowrisk moves; human approval for highimpact changes.

• Governance: Measure and monitor against baseline, develop versioned models, data lineage, bias checks, value tracking.

‍

Think back to the air travel example: pilots enable autopilot for cruise, but they still want a pilot and copilot for takeoff, landing, and turbulence.  Even on autopilot, humans are setting parameters and monitoring performance and taking over as escalations occur.

NearFuture: Agentic AI and LightsOut Optimization

What’s coming:  

• Agentic AI where agents are built to propose, run, and measure experiments, with increasing scope and autonomy as fidelity and trust improves; continuous monitoring and reoptimization against live twins; automated exception and escalation handling.

• Data Wrangling:  AI will unlock the ability to deal with data sets that are raw and unstructured, and shift from schema-on-write to schema-on-read; unlocks rapid prototyping, testing and simulation environments to detect patterns of potential value faster.

‍

‍

‍What won’t change: human accountability for cost, service, risk, safety, and compliance. Critical decisions warrant human accountability, and this is not being handed to the machines anytime soon.  The way to solve business problems is still to assemble the right team, executing (or overseeing) the optimal processes, supported by well-matched technology.

What Should Supply Chain Execs Do Now?

‍

What should we tell a person who wants to “AI my supply chain”?  AI is not an endgame, but another means to solve difficult problems, predict outcomes and automate certain tasks.  Let’s get back to first principles.  

‍

Start simply with identifying tractable problems to be solved, focusing on high-value, more difficult problems.  Often these problems are ones that traditional APS, TMS, WMS and MES applications are not suited to solve.  Find the cracks between the systems.  

‍

Don’t think about the solution or engine first, back into what methods and engines are needed when there is a clear definition of success, and KPIs.  Treat AI, optimization, and simulation as a team sport.  

‍

Sustainably solving problems requires a comprehensive approach.  Don’t overlook the foundation of strategy executed by people, processes and technical solutions coming together. Keep humans in the loop where stakes are highest. That’s how you scale decisions, capture value—and earn trust—the right way.

‍

‍