AI Demand Forecasting to Prevent Inventory Gluts

Victoria’s Secret is staring at a very modern supply chain problem: inventory that arrives cheaply, but at the wrong time.

The company’s shift from air freight to ocean shipping to offset tariffs is logical on paper—ocean is typically far cheaper per unit. But the operational side effect is brutal: you own the inventory earlier, you carry it longer, and you risk ending the season with more product than demand can absorb. In their case, leadership signaled a mid-teens percentage increase in total inventory in fiscal Q4, while also expecting a $90 million net tariff impact for fiscal 2025 (with $65 million hitting Q4).

This isn’t just a Victoria’s Secret story. It’s a case study in what happens when cost-saving moves (modal shifts, tariff workarounds, vendor negotiations) aren’t tightly connected to demand forecasting, supplier coordination, and inventory planning. In the “AI in Supply Chain & Procurement” series, I keep coming back to one stance: inventory is where strategy goes to either compound value or quietly bleed cash. AI doesn’t fix strategy, but it can prevent strategy from turning into an inventory hangover.

The modal shift trap: cheaper freight, costlier inventory

Switching from air to ocean tends to reduce transportation cost per unit, but it also changes the physics of planning. You add weeks of transit and port variability, which means you must commit earlier. That earlier commitment increases the odds you’ll be wrong about demand, promotions, weather, social trends, and competitor pricing.

Here’s the simple trade:

• Air freight: later commitment, higher cost, lower inventory risk
• Ocean freight: earlier commitment, lower cost, higher inventory risk

The problem is that many retailers evaluate modal shifts primarily as a freight line-item decision. It’s not. It’s an enterprise working-capital decision.

When you move volume to ocean, you typically trigger:

• Higher average inventory on hand (because product is in the pipeline longer)
• Earlier ownership transfer (depending on Incoterms and supplier agreements)
• More safety stock (because variability is harder to buffer)
• More markdown risk (because demand error compounds over longer horizons)

In Victoria’s Secret’s case, this shows up as a “merchandise glut” dynamic: inventory builds not necessarily because the business wants it, but because the network now requires it.

Tariffs change the playbook, but they don’t forgive bad forecasting

Tariffs are a procurement and finance headache, but they often become a planning problem too. Victoria’s Secret described a multi-pronged mitigation approach that will sound familiar to any sourcing leader:

• Optimize vendor costs
• Diversify sourcing
• Raise prices
• Adjust the air vs. ocean freight mix

I’ll take a stance here: price increases and freight optimization are often the fastest levers, but they’re the easiest to mis-execute because they don’t automatically align demand and supply.

If you raise prices while extending lead times, demand becomes harder to predict. If you shift sourcing, supplier performance and variability may change. If you shift modes, your planning calendar and replenishment logic must change.

So the question becomes: how do you make tariff-driven moves without creating a bigger downstream cost—markdowns, storage, expediting, or obsolete product?

Answer: you need tighter decision-making loops between procurement, logistics, and merchandising. That’s exactly where AI in supply chain planning earns its keep.

How AI reduces overstock risk when lead times get longer

AI doesn’t eliminate uncertainty. It does something more practical: it quantifies uncertainty faster and updates plans more often.

Traditional planning often relies on monthly cycles, static forecasts, and manual overrides. That’s fine when lead times are short and variability is modest. When you stretch lead times by shifting to ocean, the planning system must be able to:

1. Forecast demand at a more granular level (SKU-store-channel-week)
2. Sense demand changes earlier (signals, not just sales history)
3. Simulate multiple supply scenarios (port delays, supplier misses, tariff changes)
4. Recommend inventory and purchase adjustments without waiting for the next planning meeting

Demand sensing: stop treating sales history as the only “truth”

Retail demand is shaped by far more than last year’s sales. AI demand forecasting systems can incorporate:

• Near-real-time sales and returns
• Web traffic and conversion rate shifts
• Promotion calendars and price elasticity patterns
• Regional weather anomalies (critical for apparel seasonality)
• Social trend signals (especially for fashion-driven categories)
• Competitive pricing and assortment changes

The practical result is earlier detection of demand drift.

If your lead time is 45–60 days, you can’t wait 30 days to admit the forecast is wrong.

Demand sensing helps shorten that “denial window,” which is where overstock starts.

Probabilistic forecasting: planning for a range, not a single number

Most companies still plan to a point forecast. The better approach is to plan to a distribution: what’s the expected demand, and what’s the confidence range?

AI models can produce probabilistic forecasts that make inventory decisions more explicit:

• What’s the probability we’ll exceed demand by 15%?
• What’s the probability we’ll stock out if we cut orders by 10%?
• Where should we hold buffer: at origin, at DC, or in-store?

When you’re making tariff-mitigation moves, that clarity matters. It lets leaders decide, consciously:

• “We’ll accept some stockout risk to protect cash.”
• or “We’ll accept higher inventory because margin is strong and demand is stable.”

What kills businesses is not picking a trade-off. It’s accidentally picking one.

Supplier coordination: AI for procurement is about timing, not just price

Victoria’s Secret also noted sourcing diversification across countries such as Vietnam, Sri Lanka, Mexico, Indonesia, India, Egypt, China, Cambodia, and Bangladesh. Diversification reduces tariff and concentration risk, but it increases coordination complexity—different calendars, factories, transit lanes, compliance requirements, and payment terms.

AI in procurement and supplier management helps in three concrete ways.

1) Earlier risk detection in POs and production milestones

Overstock and stockouts both start as small misses: a raw material delay, a production slip, a booking issue.

AI can monitor signals across supplier OTIF performance, factory capacity indicators, and shipment status to flag:

• POs that are likely to miss ship dates
• Styles likely to arrive after peak selling windows
• Lanes with rising delay variability

That gives teams time to intervene with realistic options:

• Re-allocate inventory to higher-demand regions
• Adjust the promotion calendar
• Convert some volume back to air selectively
• Cancel or defer production where margin risk is high

2) Smarter allocation of “expedite budget”

When companies say, “We’re shifting to ocean,” they rarely mean 100%. They mean “mostly ocean, with exceptions.”

AI can recommend which SKUs deserve exception treatment based on:

• Margin and sell-through velocity
• Seasonality deadlines
• Substitute availability
• Customer service impact

A useful rule I’ve seen work: air is for protecting revenue, not fixing planning errors. AI helps enforce that discipline.

3) Negotiation leverage through total cost clarity

Tariff mitigation tends to focus on landed cost, but the real number procurement and finance should care about is:

Total landed cost + carrying cost + expected markdown cost

AI-assisted cost-to-serve models make that visible. When you can show that a 4% unit cost reduction is likely to create a 10% markdown exposure because of lead time and timing, the negotiation strategy changes. You stop “winning” the price and start winning the profit.

An AI operating model that prevents the “glut” outcome

If you’re a retailer or brand watching this story and thinking, “That could be us,” you’re right. The fix isn’t a single tool. It’s a tighter operating model.

Here’s a practical blueprint I recommend when lead times are increasing (ocean shift, sourcing shift, tariff disruption).

Step 1: Create a single forecast that merchandising, supply chain, and procurement trust

Most inventory gluts come from forecast fragmentation:

• Merch has one view
• Supply chain has another
• Finance has a third
• Procurement commits based on a fourth (usually last month’s plan)

Start with one demand forecast and a clear override process.

Step 2: Plan in weekly increments, not monthly

Ocean freight variability doesn’t respect your monthly S&OP calendar.

Adopt a cadence where you:

• Update demand sensing weekly
• Recompute inventory targets weekly
• Review exceptions (not everything) weekly

Step 3: Use scenario planning for tariffs and transit variability

For tariff and lane decisions, build 3–5 scenarios you can actually act on:

• Base case (expected demand, expected transit)
• High demand / normal transit
• Normal demand / delayed transit
• Low demand / normal transit
• Low demand / delayed transit (the dangerous one)

Then attach decisions to triggers: “If sell-through is below X by week Y, we cut receipts by Z.”

Step 4: Make inventory a shared KPI with procurement

This is where many companies get political. Procurement is rewarded on cost; inventory is “someone else’s problem.” That’s how you end up with cost wins and margin losses.

Shared KPIs that work:

• Forecast accuracy at commit date
• Receipt timing vs. seasonal window
• Inventory turns by category
• Markdown rate tied to late-arriving inventory

When procurement feels the outcome, coordination improves quickly.

People also ask: “Isn’t more inventory safer during uncertainty?”

Sometimes, yes—but only when it’s the right inventory.

Extra inventory is protective when:

• Demand is stable and predictable
• Product has long shelf life (or low obsolescence risk)
• Margin can absorb storage and capital costs

Extra inventory is dangerous when:

• Fashion cycles are fast
• Returns are high
• Lead times are long and variable
• Price sensitivity is increasing (common in inflation-fatigued consumers)

For apparel retailers in late December, the timing angle matters even more. Holiday demand has already peaked, and the market is heading into post-holiday promotions. Owning too much inventory going into January invites markdown pressure.

Next steps: if you’re planning a freight or sourcing shift in 2026

Victoria’s Secret’s experience puts a spotlight on an uncomfortable truth: cost-saving moves create new planning failure modes. The fix is not “try harder.” It’s to build planning and procurement processes that can update faster than the variability.

If you’re considering a shift from air to ocean to offset tariffs—or diversifying sourcing across countries with different lead times—start with two actions:

1. Audit your commit points (when you lock demand assumptions into POs, production, and bookings). If you can’t clearly explain your commit timeline by category, you’re guessing.
2. Pilot AI demand forecasting and exception-based planning in one category where seasonality and margin matter. Prove it with sell-through, forecast error, and markdown rate.

The real question for 2026 planning isn’t “How do we avoid tariffs?” It’s: How do we make big structural shifts without paying for them later in inventory?