The Defense Logistics Agency — an organization responsible for supplying everything from spare parts to food and fuel — is turning to artificial intelligence and machine learning to fix a long-standing problem of predicting what the military needs on its shelves.

While demand planning accuracy currently hovers around 60%, DLA officials aim to push that baseline figure to 85% with the help of AI and ML tools. Improved forecasting will ensure the services have access to the right items exactly when they need them. 

“We are about 60% accurate on what the services ask us to buy and what we actually have on the shelf.  Part of that, then, is we are either overbuying in some capacity or we are under buying. That doesn’t help the readiness of our systems,” Maj. Gen. David Sanford, DLA director of logistics operations, said during the AFCEA NOVA Army IT Day event on Jan. 15.

Rather than relying mostly on historical purchase data, the models ingest a wide range of data that DLA has not previously used in forecasting. That includes supply consumption and maintenance data, operational data gleaned from wargames and exercises, as well as data that impacts storage locations, such as weather.

The models are tied to each weapon system and DLA evaluates and adjusts the models on a continuing basis as they learn. 

“We are using AI and ML to ingest data that we have just never looked at before. That’s now feeding our planning models. We are building individual models, we are letting them learn, and then those will be our forecasting models as we go forward,” Sanford said.

Some early results already show measurable improvements. Forecasting accuracy for the Army’s Bradley Infantry Fighting Vehicle, for example, has improved by about 12% over the last four months, a senior DLA official told Federal News Network.

The agency has made the most progress working with the Army and the Air Force and is addressing “some final data-interoperability issues” with the Navy. Work with the Marine Corps is also underway. 

“The Army has done a really nice job of ingesting a lot of their sustainment data into a platform called Army 360. We feed into that platform live data now, and then we are able to receive that live data. We are ingesting data now into our demand planning models not just for the Army. We’re on the path for the Navy, and then the Air Force is next. We got a little more work to do with Marines. We’re not as accurate as where we need to be, and so this is our path with each service to drive to that accuracy,” Sanford said.

Demand forecasting, however, varies widely across the services — the DLA official cautioned against directly comparing forecasting performance.

“When we compare services from a demand planning perspective, it’s not an apples-to-apples comparison.  Each service has different products, policies and complexities that influence planning variables and outcomes. Broadly speaking, DLA is in partnership with each service to make improvements to readiness and forecasting,” the DLA official said.

The agency is also using AI and machine learning to improve how it measures true administrative and production lead times. By analyzing years of historical data, the tools can identify how industry has actually performed — rather than how long deliveries were expected to take — and factor that into DLA stock levels.  

“When we put out requests, we need information back to us quickly. And then you got to hold us accountable to get information back to you too quickly. And then on the production lead times, they’re not as accurate as what they are. There’s something that’s advertised, but then there’s the reality of what we’re getting and is not meeting the target that that was initially contracted for,” Sanford said.

The post DLA turns to AI, ML to improve military supply forecasting first appeared on Federal News Network.