Where AI is actually working in supply chain

Advertisement

Pittsburgh-based Allegheny Health Network estimated its clinicians were spending at least 112,000 hours a year chasing supply orders. Now a bot follows up with suppliers first.

“A buyer doesn’t touch an open order until after the bot does,” said Scott Meiser, vice president of supply chain operations at the system, during a panel at Becker’s 14th Annual CEO + CFO Roundtable in Chicago.

Supply chain leaders at the event were skeptical of much of the AI on the market. But several described tools already running in daily operations, beyond the pilot stage. Here’s where they said AI is earning its keep and where it isn’t yet.

Procure-to-pay

Pittsburgh-based Allegheny Health Network uses automation across procurement operations: following up on open orders, working match exceptions and disputing discrepancies with suppliers. Mr. Meiser said much of that work wasn’t happening before. Staff waited for clinicians to ask about missing supplies, then followed up.

The change let AHN shift six buyers into inventory planning roles without adding headcount as its distribution center more than doubled in size. “Nobody in here is adding headcount. Let’s be clear,” Mr. Meiser said.

To build trust, AHN started with tasks it was confident could run without a human in the loop. Mr. Meiser said the system’s procurement operations director began as a skeptic. A few months after the first workflow went live, the director came back asking to automate another one.

Detroit-based Henry Ford Health has built about 15 AI agents in recent months for tasks such as contract validation and reading statements, said Bill Moir, senior vice president of care delivery support services operations. He said the system expects to reduce its procure-to-pay staff by about 20% next year.

Forecasting

Rochester, Minn.-based Mayo Clinic forecasts supply demand using upcoming case schedules and past usage rather than past purchases alone, said Dave Marcelletti, division chair of supply chain operations. Three AI agents collect the data, forecast 30 days out and recommend what to stock where. The model refreshes daily.

The tool is still rolling out in procedural areas and has been running for less than a year. Mr. Marcelletti said monthly purchase orders have been cut in half and stat orders in cardiology have dropped to almost nothing. The time staff spend managing supplies has fallen fivefold, he said, but that wasn’t the goal. “The ROI was never to save time, save FTE, save transactions, save freight,” he said. “It was always how can we serve the patients better, quicker and never run out of anything and never cancel a case.”

Inventory cameras

San Francisco-based UCSF Health is live in 19 rooms with a computer vision system that counts and replenishes supplies, said Jake Limbert, system director of supply chain. The first phase covers 155 rooms. He said the inventory return ranges from about $600 to thousands of dollars per room, depending on acuity.

UCSF is a union shop, so the labor savings come from reallocating staff and not backfilling positions, not from layoffs, Mr. Limbert said. Staff who used to count bins with handhelds now monitor several rooms from a terminal, which he compared to air traffic control. Staff worried less about being on camera than about whether the system recorded audio, he said.

Mr. Meiser cautioned against building a camera business case around labor. At AHN, materials staff spend only about 20% of their time counting bins, he said. The bigger payoff is keeping nurses from leaving their floor to find a syringe.

Disruption monitoring

Houston-based MD Anderson Cancer Center uses a disruption monitoring tool linked to its purchase order data. It flags global events that could affect its supplies, said Janney Cheeran, director of finance and analytics. Supply chain and finance are also testing AI agents that pull in other inputs, such as weather. The program was approved only in the last couple of months, and MD Anderson is keeping access limited until it has a structured plan.

Group purchasing organization HealthTrust Performance Group uses AI to predict where markets are headed after major events and to map suppliers’ second- and third-tier manufacturing sources, said Eric Swaim, senior vice president of strategic sourcing. It also analyzes order confirmation data to monitor fill rates and back orders.

Where it falls short

When a fire destroyed UCSF’s primary distributor’s warehouse in June, Mr. Limbert said, the system’s supply chain “basically evaporated in an hour and a half.” AI played no role in the response. “We got on the phone, we had Zooms,” he said. “You set up a control center, right? And you just, you manage through it.”

Mr. Limbert said many vendors pitching AI lack specificity about which healthcare supply chain problem they solve. He called much of the hype “innovation theater.” Mr. Meiser said too many tools deliver insights without telling staff what to do next. AHN has built some of its own automation in its integration layer instead.

Leaders also said AI is only as good as the data underneath it. “That’s basic blocking and tackling,” Mr. Moir said. “You got to do that before you start building on top of a good framework for AI.”

Mr. Meiser’s advice was to start small. “Don’t wait to get started,” he said. “Go find a use case and plug it in. It doesn’t have to be overly complex.”

Advertisement

Next Up in Supply Chain

Advertisement