AI in Supply Chain Planning: Where the Real Value Lies

AI is opening up new possibilities for Supply Chain Planning. But where can it really make a difference? Nick Verstraete, CTO at Solventure, explores what has changed, what becomes possible within and beyond existing planning systems, and why the real value lies in better planning decisions, not the technology itself.

When I talk to Supply Chain leaders at manufacturing companies, most tell me they have either not started with AI in Supply Chain Planning or are still exploring. Only a handful have a pilot running, and even fewer have something in production. At the same time, the pressure from the board to "do something with AI" is real. The result is a lot of activity, and surprisingly little impact on planning decisions.

In this blog, we explore how recent advances in AI are changing what's possible in Supply Chain Planning, both within and beyond APS systems, and how to identify and validate use cases with the potential to create business value.

How has AI evolved

The real innovation is not in ‘AI’ as a domain – it is in large language models (or LLMs for short). Artificial Intelligence (or ‘Super Intelligence’ as a certain president wants to name it) has always been a broad field: machine learning, heuristics, search algorithms, and the linear and mixed-integer optimization that APS systems like Arkieva, Kinaxis and OMP have been built on for decades. None of that is new.

What changed is natural language models. Since the breakthroughs of 2017, these models can be trained on vastly more text, and today's LLMs can understand, reason, act and build in ways software never could. But let’s be clear about what they do: they generate text that fits the context you give them. That is why the evolution (see image below) went from:

• prompt engineering: asking questions such as "recommend how to allocate constrained inventory"
• to context engineering: giving the model access to orders, inventory and customer priorities;
• and now to harness engineering: where the model retrieves, analyzes, checks constraints and forms a recommendation.

What becomes possible within APS?

Within Advanced Planning and Scheduling (or APS) systems, AI mainly improves planner productivity and effectiveness – and that has value, but it is not where the value stops. All major APS vendors are releasing or roadmapping Model Context Protocol (or MCP) connections, which let an LLM interact directly with the planning system. Here, I see three maturity steps:

1. Ask an LLM to read and explain: "Show me the production plan", "Explain how this month's forecast differs from last month's", "Generate the S&OP deck". Work planners spent hours on, done in minutes.
2. Update data through natural language: give Sales a chat interface to enrich their forecast instead of a planner workbench – a new user experience on the same planning data.
3. Generate and evaluate scenarios: scenario management has existed for years, but building a sensible scenario was cumbersome. With LLMs, scenarios can be generated and compared almost automatically.

But which AI opportunities matter most to supply chain professionals? During the webinar: AI in Supply Chain Planning: What’s Changing and What Really Matters? we asked participants which AI opportunities they would prioritize in supply chain planning.
Scenario generation and decision simulation emerged as the clear top priority, followed by disruption, exceptions and root cause analysis, and balancing service levels, inventory and working capital.

Where does AI add value beyond APS?

The bigger opportunity lies in broadening the Supply Chain decision context. Instead of keeping Supply Chain planning and Finance planning in separate silos, you can build decision support that spans APS, ERP and FP&A data together.

As my colleague Bram Desmet describes in his book Rethinking Supply Chain, Integrated Value Planning means moving beyond planning Supply Chain and Finance separately. It is about making more integrated decisions across the Supply Chain Triangle, balancing service, cost and capital employed.

AI makes those cross-functional use cases practical: detecting disruptions and translating them into supply planning exceptions, balancing service levels against inventory and working capital, or supporting service-cost-cash trade-off decisions in your S&OP or IBP cycle. The value is in better trade-offs, not faster spreadsheets.

Why is your data foundation the real bottleneck?

If you want good results from LLMs and agents, the quality of the context you give them is everything and that context is still mostly implicit. When I ask companies what is stopping them, the answers are consistent: data availability, data quality, fragmentation, skills and executive commitment.

What does ‘available inventory’ mean in your company? What is your definition of service level, of capacity, of margin? Which make-to-order and make-to-stock policies apply? Today these definitions live in planners' heads.

For AI they need to be explicit and centralized: a clear planning data model, a semantic layer, and documented decision logic on your enterprise data platform, whether that is Azure, Databricks, Snowflake or Palantir Foundry. You don't need a supply chain knowledge graph on day one. You do need to understand the layers and build them out gradually.

How does AI prototyping change the way you validate use cases?

AI prototyping brings validation forward. In Solventure's 7-step transformation methodology, the envision, engage and design phases used to mean analysis, requirements and blueprints. Now, instead of only talking about requirements, we build a prototype and validate three things fast: does it solve the problem, will planners adopt it, and does it create bottom-line value. Only then do you take the industrialization decision – a production pilot, or a sharper definition of what the APS needs to cover.

This is made possible by a new type of role: the forward deployed engineer, combining Supply Chain planning expertise, data & cloud skills, AI engineering and classic data science. Forecasting algorithms and safety stock calculations are not replaced by LLMs – they remain essential. 

A recent example: in two months, one of our forward deployed engineers worked alongside a customer's planning team to develop a prototype combining rough-cut capacity planning, demand sensing and shift-based operator staffing with a small optimization engine.

One engineer, one planning team, and several fast iterations to explore an alternative to an Excel-based planning process.

What really matters as AI and Supply Chain Planning evolve?

AI is still software. It is fundamentally more capable software, but the hype should not overtake that fact. Start from high-value decisions, augment your existing enterprise technology rather than replace it, and look for value in better trade-offs between service, cost and capital employed.

The goal is not to predict every AI use case. It is continuous discovery: list the planning decisions your organization is poorly supported in, estimate the value at stake, and only then explore solutions. AI changes what planning software can do. But the real value still comes from improving the planning decision.