AI-Driven Facility Closures: Lessons from J&J Snack Foods

J&J Snack Foods says it will close three manufacturing plants (Atlanta, GA; Holly Ridge, NC; and Colton, CA) as part of “Project Apollo,” a transformation program targeting at least $20 million in annualized operating income in 2026. About $15 million of that is expected to come from the closures once they’re complete by Q2 2026, and the company has already seen distribution costs drop 8.3% year over year in Q4 2025 thanks to fewer internal transfers and better truck utilization.

Those numbers are the headline. The more interesting story is what’s underneath: companies aren’t closing sites because they “feel like it.” They’re doing it because the economics of manufacturing networks have changed—demand signals are noisier, transportation is still volatile, labor is tight in pockets, and service expectations keep rising.

In this installment of our AI in Supply Chain & Procurement series, I’m going to take a clear stance: facility closures and network consolidation are often the right move—if (and only if) you can prove the decision with data and manage the second-order effects. AI doesn’t make the decision for you, but it can make the trade-offs visible early enough to act.

What J&J’s plant closures really signal (beyond cost cutting)

Answer first: J&J’s move signals a shift from “owning capacity everywhere” to running a simpler network with higher utilization, supported by targeted modernization and a distribution center strategy.

Plant closures are usually framed as a blunt cost reduction tactic. In practice, they’re more like a network rebalancing move—you’re trading fixed-cost footprint for flexibility, throughput concentration, and better logistics flow.

J&J’s leadership described the closures as “the next logical step” enabled by investments to modernize plants, expand capacity for core products, and build out regional distribution centers. That phrasing matters. A closure is easiest to execute when:

• Remaining sites can absorb volume without service degradation
• Core SKUs are standardized enough to move production
• Warehousing and transportation can be redesigned to reduce “non-value miles”

If you’ve ever tried to consolidate production and then discovered your warehouse network can’t handle the new lane mix, you know how quickly “savings” turn into expedite spend.

The part most teams underestimate: second-order costs

Closures don’t just change where you make product. They change:

• Inbound freight patterns (supplier lanes, MOQ pressure, lead times)
• Inventory positioning (safety stock location, cycle stock levels)
• Changeover dynamics (if you’re consolidating more SKUs into fewer lines)
• Labor and overtime curves (especially during seasonal peaks)

In snack foods, seasonality and promotional spikes are real—December into early Q1 is often planning-heavy for the year ahead. If you consolidate the wrong lines, you can “save” on overhead and then pay it back in service failures when demand pops.

Where AI fits: predicting which facilities are truly redundant

Answer first: AI helps identify closure candidates by combining demand forecasting, cost-to-serve, capacity constraints, and risk signals into a model that can be stress-tested—fast.

Traditional network design is already data-driven, but it’s often episodic: run a study, pick a direction, then don’t revisit for 18–24 months. That cadence is too slow when transportation rates, regional demand, and supplier reliability can shift in a quarter.

AI becomes useful when it turns network design into a living process rather than a one-off project.

A practical “closure candidate” scoring model

If you’re trying to decide whether a plant should be expanded, repurposed, or shut down, an AI-supported scorecard typically blends:

1. Forecasted demand by region and channel (including promotions and cannibalization)
2. True cost-to-serve (manufacturing + warehousing + transportation + handling)
3. Capacity and OEE trends (line performance, downtime patterns, yield)
4. SKU complexity and changeovers (schedule fragility is a hidden tax)
5. Service impact simulation (fill rate and OTIF risk if volume is moved)
6. Supplier and raw material constraints (where inputs can realistically flow)
7. Risk and resilience factors (single points of failure, weather, labor availability)

The key is that AI doesn’t replace network optimization math. It improves the inputs (forecasting, anomaly detection, risk signals) and makes scenario testing less painful.

A closure decision is rarely about one plant being “bad.” It’s about the network working better without it.

The myth: “AI is only for forecasting”

Most teams stop at demand forecasting because it’s easy to explain. The bigger value in restructures comes from AI applied to operational constraint discovery:

• Detecting chronic micro-stoppages driving hidden capacity loss
• Predicting changeover overruns that break schedules
• Flagging which SKUs cause the most downstream picking and transport inefficiency

That’s where you find out whether a facility is underperforming because of fixable execution issues—or because it’s structurally the wrong node in the network.

Distribution optimization: why J&J’s 8.3% drop matters

Answer first: The reported 8.3% reduction in distribution expenses suggests J&J is attacking “waste miles” and internal handling—exactly where AI-driven logistics optimization tends to pay off.

The article notes the improvement came mainly from fewer internal transfers and better truck utilization. That’s a strong signal that the company is simplifying flows: fewer touches, fewer repositions, fewer “move it twice because the network isn’t aligned.”

AI use cases that directly reduce internal transfers

Internal transfers often happen because inventory is in the wrong place relative to demand, or because manufacturing and warehousing are planned in separate tools.

AI can reduce this through:

• Multi-echelon inventory optimization (MEIO): reposition safety stock to the right echelons and nodes
• Probabilistic ETA and lead time modeling: plan replenishment with realistic variability, not averages
• Order promising tied to real constraints: avoid committing to ship-from locations that trigger later transfers

If you’re running a regional distribution center model (as J&J is), MEIO is where you stop guessing and start quantifying the inventory you need at each node to protect service.

AI use cases that improve truck utilization (without breaking service)

Better utilization can mean a lot of things. The version that holds up is when you increase cube/weight utilization while maintaining OTIF.

AI typically supports this via:

• Dynamic load building: choose which orders ship together based on promised date, lane, and cube
• Shipment consolidation recommendations: identify when two partials can become one full without delaying
• Carrier selection with performance learning: pick the carrier that actually performs on that lane, not the one that looks best on a rate sheet

The pitfall is easy: optimize utilization too aggressively and you create late shipments. Good models explicitly price the trade-off between utilization and service.

Procurement’s role in a manufacturing consolidation (it’s bigger than renegotiating)

Answer first: Procurement can make or break a consolidation by managing supplier lane redesign, contract risk, and input availability—not just unit price.

When plants close, procurement has to rewire the physical reality of supply. That includes qualifying alternative raw material sources, renegotiating freight terms, and changing order patterns that may increase supplier costs.

Here’s what I’ve found works when procurement is pulled in early:

The “supplier-lane map” exercise

Before you finalize closures, build a lane map that shows:

• Current supplier-to-plant lanes and spend
• Proposed lanes after consolidation
• Lead time changes and variability
• Incoterms and who controls freight
• Single-source exposures created by the new design

Then use AI (or simpler analytics, if that’s what you have) to identify which supplier changes drive the biggest total landed cost swings.

Watch for hidden MOQ and packaging constraints

Consolidation often increases batch sizes at the remaining plants. That can trigger:

• MOQs that inflate inventory
• Packaging or ingredient constraints at a single location
• Longer production runs that reduce responsiveness to demand changes

This is where AI-enhanced demand sensing can help. The more confident you are in near-term demand signals, the less you overproduce “just in case.”

A restructuring playbook you can apply in Q1 planning

Answer first: The fastest path to smart consolidation is to pair network scenarios with AI-supported operational reality checks—then implement in controlled waves.

A lot of restructures fail because the plan looks perfect in a network model, but execution collapses in scheduling, warehousing, and transportation.

Here’s a pragmatic approach (and yes, it’s doable without a massive tech overhaul):

1. Create 3–5 network scenarios (baseline, consolidation A/B, outsource option, DC redesign)
2. Quantify cost-to-serve per customer segment (retail, foodservice, DTC if applicable)
3. Stress-test service using variability, not averages (lead time distributions matter)
4. Run a “capacity truth” audit at the remaining sites (OEE, downtime reasons, labor limits)
5. Sequence changes in waves (move stable SKUs first; leave volatile/promotional SKUs for later)
6. Set leading indicators (internal transfers, expedite spend, schedule adherence, OTIF)

If you only track financial savings after the fact, you’ll miss the early warning signals that tell you the network is starting to fray.

People Also Ask: “Can AI tell you which plant to close?”

Yes, but not by itself. AI can rank closure candidates and simulate outcomes, but leadership still has to weigh strategic factors: customer commitments, labor relations, regulatory constraints, and long-term portfolio direction.

The best teams treat AI as the decision-support layer that makes assumptions visible and debatable.

People Also Ask: “What data do we need to do this well?”

Start with what you already have:

• Shipment history by lane, cost, and service performance
• SKU-level demand history with promo markers
• Plant production data (rates, yields, downtime)
• Inventory snapshots by node
• Supplier lead times and OTIF/quality performance

Perfect data isn’t required. Consistent definitions are.

What to do next if you’re considering consolidation

J&J Snack Foods’ Project Apollo is a clean example of a broader trend: manufacturers are simplifying networks and tightening distribution execution to protect margins. The numbers—$20M annualized operating income target, $15M expected from closures, 8.3% distribution expense reduction—are a reminder that savings come from coordinated moves across manufacturing and logistics, not isolated cuts.

If you’re in supply chain or procurement leadership, the next step is straightforward: treat network design as an always-on capability, not a once-every-two-years project. AI makes that practical by improving forecasts, exposing operational constraints, and helping teams run more scenarios with less effort.

If you’re mapping your 2026 plan right now, ask yourself one forward-looking question: what would your network decision look like if you had to defend it with weekly data instead of a one-time study?