The Fifth Elephant Winter Edition 2026

Tracing supply chain disruption and checking AI recommendations with connected data and business rules

A fabric mill is late delivering an order for a seasonal collection. Which styles and sizes will be affected? Which stores and promotions are now at risk? Can the retailer change suppliers, move inventory between channels, or substitute another style without creating a different problem?

Answering these questions requires tracing dependencies across purchase orders, fabrics, garment vendors, products, inventory and launch plans. A recommendation also has to meet business constraints: an alternative supplier may lack the required certification, a store may have limited capacity, or replacement stock may arrive after the selling season.

This talk uses a fashion retail case study to examine how to structure and interrogate data for such decisions. Starting with the mill delay, I will walk through the entities, relationships and business rules needed to establish what is affected and assess possible responses. The case provides a concrete way to ask a broader question: what must we make explicit in our data before an AI agent can produce recommendations we can check?

The approach combines an enterprise knowledge graph with an LLM and rule-based validation—a form of neurosymbolic AI. The graph represents business dependencies; the LLM helps interpret questions, query the data and explain options; and rules and optimization tools check whether proposed actions meet the constraints represented in the system. I will explain how these roles fit together, where they help, and what they leave unresolved.

Building and maintaining this structure is part of the problem. I will discuss how LLMs can assist with extracting entities and relationships from business records, where human review remains necessary, and how changing data and incomplete rules affect the reliability of an answer.

An evaluation will compare answers to the same disruption questions using an LLM alone, retrieval over documents, and the graph-and-rules approach. We will examine whether answers identify the affected parts of the business, remain faithful to the available evidence, and recommend actions that satisfy the stated constraints.

The talk closes with lessons for recognizing similar problems in other domains, including manufacturing and finance: when connected data and explicit rules are useful, what it takes to maintain them, and how to evaluate whether they improve decisions.

What participants will take away

• How to work backwards from a decision question to identify the entities, relationships and constraints the data must represent.
• How to distinguish tracing an event’s consequences from checking whether a proposed response is feasible.
• How to evaluate an AI recommendation for supporting evidence, missed impacts and violations of business rules.
• How to recognize when this approach fits a problem, and where incomplete data, stale relationships or missing rules limit it.

Who this is for

Business analysts, operations leaders, data practitioners and people building AI decision-support systems. The talk uses retail as a worked example; no prior knowledge of knowledge graphs or neurosymbolic AI is assumed.

About me

Abhijith Neerkaje is a data science and AI leader with over 20 years of experience building AI products across retail, semiconductor manufacturing, and energy. He is the Co-founder of Beyond Vectors AI Pvt Ltd, where he helps mid-career professionals and engineering leaders build expertise in Generative AI and Agentic AI.

Previously, Abhijith led the Data Science & Analytics function at Falabella India, driving AI-powered search, recommendations, pricing, and seller intelligence for one of Latin America’s largest e-commerce platforms. Earlier, at Target, he built machine learning solutions for pricing, merchandising, and supply chain optimization.

Abhijith holds engineering degrees from PES University (formerly PESIT) and the Indian Institute of Science (IISc), Bangalore, and an MS in Engineering and Management from the Massachusetts Institute of Technology (MIT). He was an MIT Tata Fellow and a winner of the MIT Clean Energy Prize (Renewable Energy Track, 2013).

Areas I can help with

• I can help with: ideas / experience / technique / dataset / collaborator intro / critique / facilitation
• I’d like to meet: researchers / practitioners / domain experts / tool builders / dataset owners / facilitators}

Topic/domain tags

#DecisionIntelligence #SupplyChainAnalytics #KnowledgeGraphs #NeurosymbolicAI