Stop doing isolated AI: How multiagent processes transform the supply chain

Do you know what every organization has in common? Lots of ideas about how to use AI.

While the CEO is talking about AI in broad terms, the innovation team is already running proofs of concept and some business functions have started pilots. Demonstrations, vendor conversations and internal experiments are all happening at the same time.

Amid all this activity, proof is emerging that AI adds value. Yet I keep hearing the same thing: We don’t know what to do next.

My answer is simple: You probably don’t need another use case. You need to connect the ones you already have.

That means your next step should be to implement multiagent systems, where multiple AI agents work together across a business process to interpret what is happening, coordinate the response and take action.

More ideas don’t automatically create more value

Naturally, you can’t start connecting AI use cases until you know what they are. Once you start mapping them, you’ll be surprised at how many you find.

When a manufacturer wanted to define and implement agentic intelligence throughout their value chain, we ran 14 workshops for their customer service, production, logistics and warehousing departments and spoke with more than 60 people. By the end of the process, we had cataloged 129 improvement opportunities.

Finding that many opportunities isn’t unusual. It reflects the way most organizations have adopted AI so far — they’ve introduced capabilities function by function, launched pilot programs and encouraged employees to experiment with AI tools.

This has led to the discovery of many small use cases that save individual teams time and effort. But for the organization as a whole to benefit, those use cases must be organized and connected.

Multiagent systems connect several activities into a process

In supply chain management, AI is already improving forecast accuracy, flagging changes in demand patterns, surfacing supplier risk earlier and optimizing transportation routes.

These gains are worth keeping, but each use case enhances only a single activity. A human is still required to pass the separate pieces of intelligence on, even when they all relate to the same supply chain, and this limits the value each activity creates. Organizations realize the most value from AI when they stop thinking in use cases and start learning how to connect processes.

Say an isolated AI tool reads and interprets incoming customer orders. It’s useful but self-contained. In a multiagent environment, the system reads the order, checks the quantity against expected demand, identifies anomalies, evaluates stock-out risk over a two-week horizon, recommends a revised quantity, updates the planning system and notifies the people who need to know.

For the manufacturer I mentioned earlier, following a multiagent approach enabled us to group the 129 individual improvement opportunities into just 36 prioritized AI use cases.

At the moment, multiagent systems are mostly used in software, but they will increasingly extend into the physical world through autonomous handling equipment, vision systems and connected assets on plant floors. It’s still early, but the processes you connect today could eventually be executed by physical AI, so it makes sense to prepare for that transition.

• ALSO READ: From manufacturing to retail, physical AI makes modernization real

Two problems, one process

When you look at manufacturing, logistics and customer fulfillment as one connected system rather than separate functions, you start to see that apparently unrelated improvement opportunities are actually the same problem, just viewed from different angles.

Let’s take a real-world example. For one of our projects, warehouse managers were worried about staffing. Some shifts needed 10 frontline operators, others 4. Managing the peaks and valleys was difficult. Separately, the logistics team told us that the arrival pattern of trucks at the docks was a constant headache: they turned up early, late or in clusters, and nobody knew which it would be in advance.

At first glance, these looked like two separate problems. In reality, they were one use case.

Because truck arrival patterns drove much of the staffing volatility, solving each challenge separately meant solving neither particularly well. Looking at them together, however, has allowed us to synchronize operator availability with inbound and outbound truck movements.

If you can predict that an important truck will be two hours late, you can move employees around before the gap opens up — for instance, placing fewer in loading but more in preparing customer orders — instead of leaving them standing around waiting or scrambling once the truck finally arrives.

A logistics specialist thinks about transportation, and a warehouse manager thinks about staffing. AI creates more value when you look at the process connecting the two. That’s how you turn a long list of possible use cases to a much shorter list of AI opportunities.

• ALSO READ: NTT DATA and Microsoft simplify multiagent AI systems at scale

How to decide what to do first

Start with the value chain rather than a list of use cases, and look for problems that appear in different forms in different functions. The priorities will soon become clearer.

We define our clients’ AI journeys by working through six elements:

1. Set the ambition: Decide what the business needs to achieve before thinking about the technology.
2. Identify improvement opportunities: Speak to your employees about their pain points.
3. Define the use cases: Group improvement opportunities and determine which capabilities are worth building.
4. Prioritize the use cases: Use clear criteria to decide which use cases to tackle first.
5. Choose your technical architecture: Ensure that you have the data, platforms and integrations needed to support the use cases you’ve prioritized.
6. Build a roadmap: Settle on a phased approach to help your organization introduce and scale the new capabilities.

Most of the discussions, and arguments, are about prioritizing use cases. We weigh two things: the value each use case could create and whether your organization is ready to deliver it. Do you have the data availability, process maturity and systems integration to make it an effective AI use case?

Find a partner that brings together strategy and technology

Plenty of firms will help you think. Some partners specialize in strategy, while others are strongest at implementation.

What’s harder to find is a partner that can do both: help you decide where AI can create the most value and then build the systems that make it happen.

That’s where NTT DATA is different. We combine deep industry and process knowledge with AI strategy, consulting, engineering and implementation expertise. It means we can look across your value chain to identify where AI can make the biggest difference, connect opportunities that might otherwise be treated in isolation and then design, build and integrate the multiagent systems needed to put those ideas into practice.

What to do next

Request a half-day workshop to identify where AI will create value in your organization and turn a long list of possibilities into a practical plan. This is where your journey to AI success starts.