Over the last few years, one challenge has consistently come up in conversations with customers reliant on their supply chain insights: how do you introduce intelligent AI capabilities into existing ecosystems in a cost-effective way whilst delivering genuine business value?

For these organizations, resilience depends not only on the volume of data they collect, but on how rapidly they can act on it. Disruptions can ripple across inventory, demand, manufacturing, and logistics rapidly, often before the full picture can be fully communicated or anticipated. AI can help unlock faster insight, giving teams the window to act, but only when built on a trusted data foundation containing data which is ready for reporting.

Fragmented supply chain data

Many supply chain businesses have spent years developing reliable planning processes. Historically, long-term forecasts, annual budgets, and stable logistics networks provided sufficient certainty for decision making. But recent times have been characterized by continual disruption and volatility in global logistics.

A disruption in one part of the supply chain can quickly trigger a chain reaction across inventory, production, and fulfilment. To protect customer commitments, teams often find themselves making expensive last-minute decisions, such as switching transport modes which have premium freight rates.

The challenge is that the data needed to make these decisions is often scattered across SAP ERP (Enterprise Resource Planning) systems, logistics providers, and planning platforms. Without a connected view, organizations struggle to understand downstream impacts until they’ve already affected performance.

Business users are often limited to predefined reports and dashboards. When new questions arise, they must wait for their data teams to build additional analysis. In a fast-moving environment, that delay can mean missing the opportunity to act.

We recently worked with a global pharmaceutical manufacturer facing exactly this challenge. Their experience highlights why strong data foundations are essential for unlocking AI value.

Both ends into the middle

At Thorogood, we use a framework we call “both ends into the middle” to bridge the gap between business goals and data reality.

We start by understanding the business problem or opportunity. In this case, our client wanted the ability to query data on demand, receive proactive insights and access recommendations they could act on immediately to improve their supply chain efficiency.

We then examine the data landscape. Like many organizations, they had valuable information spread across multiple systems and providers. The temptation can be to place a natural language chatbot directly on top of these data sources, but an AI solution is only as effective as the information it can access. If the underlying data is fragmented or inconsistent, the outputs will be too.

The solution is to create a robust semantic layer: a trusted data foundation that contains cleaned, reporting-ready data which has been unified from multiple sources.

Building the foundation for effective AI use

Creating an effective AI solution to get insights into your supply chain starts with establishing a single source of truth.

For our client, this meant bringing together SAP and logistics provider data into a Lakehouse with a medallion architecture. The immediate benefit was that disconnected datasets are unified, creating a common foundation for analysis.

We then developed a semantic model that defined key business logic, relationships between data sets and metric calculations needed to gain valuable insights into their supply chain. This created a trusted, reusable foundation for reporting and AI initiatives.

With this layer in place, we could build analytical reporting for day-to-day operations in Power BI and introduce AI agents using Databricks Genie workspaces. Databricks Genie is an AI offering which allows a user to ask a question in plain language about their data. Users receive instant answers (this can be a text response but can also be supported by relevant visualizations) without writing code or submitting requests to data teams.

Users can also go a step further by using the Agent mode of Databricks Genie. Whilst Chat mode is optimized for quick answers, Agent mode acts more like an analyst by creating a research plan, running multiple queries, testing hypotheses, iterating on findings and producing detailed explanations or reports. This deeper insight allows a user to see not just the current problems, but trends and patterns, alongside future recommendations which can allow them to manage issues proactively in that crucial window when action will have the most impact.

A roadmap to effectively using AI to gain full insight into your supply chain

Our incremental approach to AI solutions ensures that, at each stage of the process, our client unlocked business value and had the right foundations in place to support the next phase.

To recap, in the first phase, we consolidated the data into a unified platform. And in the second, we established the semantic layer, creating reusable business logic and consistency. The third phase introduced analytical reporting and the fourth implemented natural language querying and the adoption of AI agents using Databricks Genie.

So, what comes next?

So far, we have implemented a dedicated logistics agent, which covers a subset of the full supply chain. The plan is to replicate this success across the wider network, building agents for sourcing, manufacturing, warehousing, distribution and commercial. But the real value will come from interconnecting these agents and making them work together.

To achieve this, we’re building a cohesive, multi-agent agentic framework to answer multi-step problems across the supply chain, with the shared business context to provide root cause analysis and to flag problems with other agents as they arise.

The foundation comes first

AI has enormous potential to improve supply chain performance, but success does not start with AI itself, it starts with the data foundation.

Organizations that invest in strong data foundations, unified platforms, trusted semantic models, and robust governance place themselves in the best position to unlock the next generation of intelligent capabilities.