Markdown Optimization: How AI Cuts Inventory Carry Without Killing Margin

Why markdown is broken at most retailers

Most multi-store retailers run markdowns the same way they did in 1995. Buyers and category managers set markdown plans on a calendar: week 8 of the season, week 12, end of season. Discounts step up in fixed percentages: 25% off, then 40%, then 60%. The plan rarely accounts for what's actually happening at the SKU-store level.

The result is two structural failures playing out simultaneously across thousands of items:

• Late markdowns. Slow movers sit on shelf for weeks past the point where the discount needed to start. By the time the calendar markdown hits, the inventory is stale, the season is shifting, and the margin recovery is much worse than it would have been with an earlier, smaller cut.
• Deep markdowns. When the calendar markdown finally arrives, it's set at the discount level needed for items that are months past their sell-through window. Items that would have moved at 20% off get marked 40% off because the calendar is uniform. The over-discounting compounds: customers learn the markdown cadence and wait for it.

The combined cost of these two failures, across a typical specialty retailer, is 3-7 percentage points of gross margin per year. On $200M in revenue, that's $6M-$14M in recoverable margin sitting in the markdown calendar.

What markdown optimization actually does

Markdown optimization is the use of demand modeling and constraint solvers to set the right markdown level for the right SKU in the right store at the right time. The output is a markdown recommendation per SKU per store per week, not a uniform calendar.

Three things change when AI is doing this work instead of a planning calendar:

1. Markdowns trigger on sell-through, not the calendar. The system tracks weekly sell-through against the planned trajectory. When an SKU is trending behind, a small markdown gets recommended early. The earlier markdown is smaller (10-15% vs. 40%) and produces equivalent or better margin recovery because the inventory hasn't aged yet.

2. Markdowns are store-specific. The same SKU sells at different velocities in different stores. A blanket "30% off" applied across the chain over-discounts where demand is healthy and under-discounts where demand is weak. Per-store markdowns let the high-demand stores keep more margin while the slow stores clear inventory faster.

3. Markdowns adapt to what actually happens. When a markdown lands and the sell-through response is faster or slower than predicted, the next-period recommendation adjusts. The system learns from each cycle.

The math of earlier, smaller markdowns

Why earlier and smaller wins, in concrete terms. Take an SKU stocked at 1,000 units across 50 stores, $40 retail, $20 cost. Selling at 80 units/week chain-wide, expected to sell through in 12.5 weeks if velocity holds.

By week 6, actual sell-through is at 60% of plan: 280 units sold instead of 480. The traditional calendar says wait for week 8 markdown. Here's the difference between two paths:

Path A: Calendar markdown. Wait until week 8. Apply 30% off ($28 retail) chain-wide. Velocity recovers to 65 units/week (a 90% lift on the discounted price). Selling continues for 11 more weeks. Final clearance at week 19 with 80 units left, marked 60% off. Total margin: $11,200.

Path B: Optimization-driven markdown. At week 6, apply 12% off ($35.20 retail) at the 30 lowest-velocity stores. Velocity at those stores recovers to 50% of plan. Apply nothing at the 20 healthy-velocity stores. By week 9, velocity is on track. Total markdown depth: averaged across the chain, ~6%. Sell-through completes by week 11 with 30 units left, marked 30% off for clearance. Total margin: $14,800.

Path B produced $3,600 more margin (32% improvement) on a single SKU. Multiply across 8,000-15,000 markdownable SKUs at a typical specialty retailer and the annual recovery is in the millions.

What it takes to deploy

The infrastructure needed for store-level markdown optimization is mostly things multi-store retailers already have. The gaps are usually integration, not capability.

Required:

• POS data with SKU-store-day granularity (most retailers have this)
• Inventory position by SKU-store (most have this, often with delay)
• Original retail and cost by SKU (always available)
• Some history of past markdowns and the response to them (most have this in spreadsheets)

Helpful but not required:

• Store cluster / format metadata (used to bootstrap the demand model)
• Promotional calendar (lets the system avoid markdown-on-promo conflicts)
• Customer purchase history (improves prediction at customer-loyalty-program retailers)

The actual deployment shape: connect the data sources, run a 4-6 week baseline period, pilot on 2-3 categories with controlled comparison, then scale. Time to first margin-positive markdown decision is typically 8-10 weeks.

How to measure markdown rate, and the number underneath it

Markdown rate is markdown dollars divided by original retail dollars, times 100. A category that rang $2M against $2.4M of original retail took a 16.7% markdown rate. Simple, and reported almost everywhere.

The reported number is rarely the useful one, for two reasons.

Permanent versus promotional. A promotional discount is a demand-generation decision. A permanent markdown is an admission that the buy was wrong. Blending them into one rate hides the second inside the first. Split them, and the markdown line stops being a pricing metric and starts being a buying metric.

Rate versus timing. Two categories can post identical 22% markdown rates and be in completely different health. One took 12% in week 6 and 10% at season end. The other took nothing until week 14 and then 22% at once. The second one destroyed far more margin and both look the same on the scorecard. What separates them is sell-through trajectory, not markdown depth.

The single best diagnostic here is average weeks-to-first-markdown against season length. If your first markdown consistently lands after 60% of the season has elapsed, you are running the late-and-deep pattern regardless of what the rate says.

Markdown rate benchmarks by vertical

Markdown rate is one of the most format-dependent numbers in retail. A 30% markdown rate is a crisis in grocery and unremarkable in fashion.

Read the last column with the second. A fashion retailer at a 38% markdown rate whose first cut lands in week 6 is running a normal business. The same 38% arriving entirely in week 11 means the buy was wrong and the calendar hid it for two months.

Where markdown optimization doesn't help

Three categories where the standard markdown optimization approach underperforms or doesn't fit:

• Grocery and consumables. The category structure is replenishment-driven, not markdown-driven. Slow-moving items get cycled out of assortment, not marked down to clear.
• Pure-fashion seasonal launches with no historical analog. Brand-new product lines or limited-edition drops without comparable history don't have the data foundation. Optimizers default to flat calendar markdowns until comparable patterns emerge.
• Retailers with severely fragmented inventory data. If your POS, ERP, and inventory systems disagree about how much of an SKU exists at which store, the optimizer makes decisions on bad data and produces bad outcomes.

Ward's role in the markdown stack

Ward isn't a markdown optimization platform. Those tools (Revionics, dunnhumby, Rapid Response) handle the optimization itself. What Ward does is monitor whether markdowns are producing the predicted outcome.

When a markdown lands and sell-through response is materially off prediction, too fast (under-priced) or too slow (under-discounted), Ward surfaces the deviation, attributes it to a probable cause (demand shift, competitive promo, weather, regional inventory issue), and routes the finding to the merchandising team.

For retailers running markdown optimization without the observability layer, the failure mode is consistent: the optimizer makes a decision, the outcome misses, and nobody notices for 2-4 weeks until the next planning cycle. By then, the margin opportunity is gone. Ward closes that loop in 24-72 hours.

Key takeaways

• Calendar-driven markdowns fail two ways at once: too late, so inventory ages past its recovery window, and too deep, because the uniform discount is sized for the worst items in the bucket.
• The combined cost runs 3 to 7 points of gross margin a year at a typical specialty retailer, which is $6M to $14M on $200M in revenue.
• Earlier and smaller beats later and deeper. A 12% cut in week 6 at the slowest 30 stores outperformed a chain-wide 30% cut in week 8 by 32% on margin in the worked example.
• Markdown rate is markdown dollars over original retail. Split permanent from promotional, because a permanent markdown is a buying error and a promotional discount is a demand decision.
• Rate alone is not health. Two categories at 22% are in different shape if one took it in stages from week 6 and the other took it all at week 14. Track weeks-to-first-markdown against season length.
• Benchmarks are format-specific: 5 to 10% grocery, 20 to 40% fashion, 12 to 20% furniture. A 30% rate is a crisis in one and routine in another.
• Markdown optimization does not fit replenishment-driven grocery, brand-new lines with no historical analog, or retailers whose POS and inventory systems disagree about what exists where.
• Ward is not a markdown optimizer. It watches whether the markdowns you set produce the predicted sell-through and flags the misses in 24 to 72 hours instead of at the next planning cycle.