AI Inventory Optimization Lessons from Liquor Stockpiles

U.S. spirits exports to Canada fell 85% year-over-year in Q2 2025—a cliff, not a dip. And now, at least four Canadian provinces are clearing out remaining stockpiles of American liquor that were pulled from shelves earlier this year.

If you lead supply chain or procurement, don’t get distracted by the headline. This isn’t “just” a trade dispute story about whiskey and tariffs. It’s a clean example of what happens when policy risk meets inventory decisions, and the organization doesn’t have a fast, data-backed way to re-plan.

I’m using this story as part of our AI in Supply Chain & Procurement series because it highlights something most teams underinvest in: AI-driven inventory optimization that can re-forecast demand, quantify risk, and choose the least-bad exit when you’re already holding surplus.

What really happened (and why it matters to supply chain teams)

The immediate point: several Canadian provinces halted sales of U.S. liquor in March 2025 in protest of announced U.S. tariffs and broader political tensions. Months later, provinces including Nova Scotia, Prince Edward Island, Manitoba, and Newfoundland reported plans to sell remaining inventory—often for a limited window—and in some cases donate proceeds to charities.

The supply chain point: when a category is suddenly removed from shelves, you don’t just lose those sales. You create second-order effects:

• Inventory carrying costs rise (warehousing, insurance, shrink, write-down risk).
• Demand shifts to substitutes (Canadian brands, other categories), and that shift can “stick.”
• Supplier relationships strain, especially if the vendor’s production planning assumed steady cross-border flow.
• Forecasting signals break because the shelf is no longer a truthful read of consumer preference—policy is now a demand driver.

The numbers in the source story underline the scale: the Spirits Canada coalition cited a 12.8% dip in total spirits sales from March to end of April 2025, calling out the ripple effects when cross-border trade breaks down. In other words: the impact didn’t stay neatly within “U.S. products.”

This is exactly the kind of scenario where AI can help—not by predicting politics perfectly, but by making your inventory decisions more resilient and your response faster.

Why surplus stockpiles happen: three failures AI can reduce

Surplus isn’t usually a single mistake. It’s a chain of reasonable decisions that become unreasonable once conditions change.

1) Forecasts that assume “the future looks like last year”

Classic demand planning struggles with structural breaks—events that make historical demand less relevant. A policy-driven boycott is the definition of a structural break.

AI demand forecasting won’t eliminate shocks, but it can adapt faster by:

• weighting recent signals more intelligently,
• detecting abrupt regime changes (e.g., “sales dropped to near-zero for non-price reasons”),
• incorporating external variables (policy changes, news sentiment indices, border friction signals) as features, not anecdotes.

A practical stance: if your planning system can’t represent “this demand went to zero because we removed it,” your model is learning the wrong lesson.

2) Inventory policies that don’t price risk correctly

Many organizations treat inventory targets as operational hygiene (service levels, safety stock) rather than a risk-priced financial position.

When geopolitics and regulation are live variables, your safety stock is also a bet. AI-supported inventory optimization can quantify that bet in dollars by simulating scenarios:

• If the category is blocked for 30/60/90 days, what’s the cash impact?
• What’s the expected write-down risk by SKU (age, velocity, margin)?
• What’s the cost curve for holding vs. clearing (storage + capital + obsolescence)?

You don’t need perfect predictions. You need a range and a policy that doesn’t pretend the range is narrow.

3) Slow decision cycles between procurement, legal, and operations

In stories like this, the operational question isn’t “Should we sell it?” It’s “When, where, and how do we clear inventory without creating the next problem?”

AI helps when it’s paired with clear governance:

• procurement owns supplier and contract constraints,
• finance owns write-down thresholds and margin guardrails,
• operations owns storage, distribution, and shelf execution,
• legal/compliance owns what can be sold, relabeled, or transferred.

When those rules are codified, AI can optimize within them—instead of waiting for a dozen meetings to happen first.

How AI would approach a stockpile clearance (step-by-step)

Clearing surplus isn’t a one-lane road called “discount it.” The provinces in the story effectively used a time-bound clearance window; some tied it to a cause (charity) to manage reputational risk. That’s smart human judgment.

Here’s what an AI-assisted clearance playbook looks like in a commercial setting.

Step 1: Build a SKU-level “sell-through reality” view

The first win is accuracy: many teams don’t have a clean, shared picture of what’s in stock, where, and in what condition.

An AI-enabled inventory control tower typically normalizes:

• SKU attributes (pack size, alcohol %, origin, shelf-life constraints)
• inventory age and dwell time
• on-hand vs. in-transit
• store/DC capacity constraints

If your data is messy, don’t start by dreaming up sophisticated models. Start by making the “how much do we actually have?” question boring.

Step 2: Segment inventory by the right objective (not one blanket rule)

A common failure is treating all surplus the same. AI-driven segmentation can classify inventory into buckets like:

• Protect margin: high-demand, low-risk SKUs; clear slowly.
• Protect cash: medium-demand; clear with targeted promos.
• Protect space: bulky/slow-moving; clear aggressively.
• Protect compliance: SKUs with regulatory or labeling constraints; prioritize legal pathways.

This matters because your best clearance strategy is rarely uniform across the category.

Step 3: Choose the clearance channel mix

AI can recommend the lowest-cost set of actions based on constraints:

• targeted store reallocation (ship to locations with better sell-through)
• time-boxed promotions with price floors
• bundles or mixed-case offers for licensed trade
• returns/credits negotiations where contracts allow
• secondary markets where legally permitted

The important part: AI can evaluate combinations. Humans tend to pick one lever. Real-world clearance often needs three.

Step 4: Optimize timing with seasonality and shelf competition

December matters. Holiday demand lifts spirits categories, and shelf space competition is intense. Provinces choosing to sell through the season isn’t random—it’s inventory math.

AI planning systems can explicitly model:

• holiday lift by region and channel
• promotional elasticity (how much volume responds to discounting)
• cannibalization (clearing U.S. spirits may reduce sales of substitutes)
• labor and distribution constraints during peak season

A blunt clearance in December can boost cash but also distort your January baseline. AI can forecast the post-promo trough and help set expectations upstream.

Step 5: Measure outcomes in dollars, not anecdotes

Clearance programs succeed or fail on a few metrics:

• days of inventory on hand (DIO) reduction
• gross margin impact vs. baseline
• working capital released
• waste/write-down avoided
• service level recovery in unaffected categories (space and labor freed)

AI supports this by creating a “counterfactual” view: what would have happened without the action? That’s how you learn—not just react.

The procurement angle: risk-aware sourcing beats reactive clearing

The clearance story is the visible end of the pipeline. Procurement decisions happen earlier, quieter, and with fewer headlines.

Here’s the procurement lesson I’d take from this case: if your category is exposed to cross-border policy risk, you should treat sourcing and inventory as one integrated decision.

What AI can do for risk-aware procurement

AI-driven procurement analytics can:

• score suppliers and lanes for geopolitical and regulatory exposure
• model dual-sourcing trade-offs (cost vs. resilience)
• recommend contract structures (flex volume bands, postponement, buy-back clauses)
• detect early signals from shipment delays, customs friction, and market pricing

One stance I’ll defend: If you’re still doing supplier risk scoring once a year in a spreadsheet, you’re operating at the speed of the last crisis.

A practical “policy shock” checklist (you can use next week)

If a category faces sudden restrictions (tariffs, bans, labeling changes), run this checklist:

1. Inventory truth: What’s on hand, where, and how old?
2. Constraint map: What can’t we do (legal, contract, channel rules)?
3. Demand reset: What is demand under restriction vs. after restriction?
4. Exit options: Which levers exist (reallocate, discount, bundle, return, secondary)?
5. Financial guardrails: Minimum margin, maximum write-down, cash target.
6. Decision cadence: Daily/weekly control tower until normalized.

AI makes this faster and more consistent. But the checklist is the foundation.

“People also ask” style answers (the ones stakeholders will throw at you)

Can AI prevent overstock entirely?

No. It can’t prevent a tariff announcement or a boycott. What it can do is reduce how often you’re surprised, and shrink the dollar value of the surprise by adapting forecasts and inventory targets faster.

Is clearance optimization just pricing optimization?

No. Pricing is one lever. The real clearance problem is a network optimization problem: inventory location, capacity, labor, channel rules, and timing.

What data do you need to start?

You can start with:

• sales history by SKU/location
• current inventory by node (store/DC)
• inbound POs/ETAs
• basic product attributes (case size, margin, constraints)

Add external risk signals over time. The first milestone is a reliable internal dataset.

Next steps: turning this headline into a capability

The Canadian provinces’ sell-off is a reminder that inventory is a decision made under uncertainty, not a static asset. The teams that will win in 2026 are the ones that can re-plan quickly when uncertainty becomes real.

If you’re building your roadmap for AI in supply chain and procurement, prioritize two capabilities:

• AI demand forecasting that detects structural breaks (policy, channel shifts, product removals)
• AI inventory optimization that recommends actions (reallocate, clear, hold) with dollar-based trade-offs

The final question to bring to your next S&OP meeting: If a top category went to near-zero demand tomorrow for non-price reasons, how many days would it take us to decide—and how many dollars would that delay cost?