A Markdown Is a Decision, Not a Panic Button: How AI Can Guide Fashion Pricing

A fashion markdown often begins with an uncomfortable sentence: “This stock is not moving.” The rushed response is to cut the price and hope demand appears.

AI can improve that decision, but only if it is given a commercial question more precise than “What discount should we offer?” The useful question is: “What action gives this product a fair chance to sell while protecting margin and avoiding unnecessary discounting?”

Start with the reason the product is slow

Low sales do not automatically mean the price is wrong. A men’s shirt may be hard to find in search, unavailable in key sizes, poorly photographed, out of season or shown to the wrong audience. Reducing the price before checking those causes teaches the system the wrong lesson.

An AI-assisted review should bring together product age, available size depth, page views, add-to-cart behaviour, orders, returns, campaign exposure and contribution margin. It should mark missing data rather than filling gaps with assumptions.

Separate prediction from policy

The prediction layer

A model can estimate likely demand under several price scenarios. That estimate is uncertain, especially for a new style with little history. Similarity to older products can help, but “similar” must be defined carefully: colour, category, fit, season and price band may all matter.

The policy layer

The business decides the boundaries. Minimum margin, brand positioning, marketplace commitments, tax treatment and campaign rules should be explicit constraints. Shopify’s official guide explains how a sale price is represented using price and compare-at price; it also notes that legal requirements around sale pricing vary by region. See Shopify’s sale-pricing documentation.

AI can recommend within those boundaries. It should not quietly change them.

Ask for three actions, not one number

A strong decision note might offer:

• hold: keep the current price and improve discovery or creative;
• test: run a limited markdown for a defined audience or period;
• exit: use a deeper, time-bound action when stock age and demand evidence justify it.

Each option should show the evidence, expected trade-off and review date. For a product such as the Regal White Cotton Men's Shirt, the live product page remains the source of truth for current price, sizes and availability; the shirt is an example for the workflow, not proof that it requires a markdown.

Watch for unfair or unstable recommendations

If an algorithm learns mainly from promotion-heavy periods, it may recommend discounts too readily. If it treats different customer groups differently, the business needs to understand why. Google’s current AI principles call for rigorous testing, monitoring, safeguards, human oversight and action against unfair bias. Review the Google AI principles.

Test recommendations by category and size availability. Men’s shirts and boys’ kurtas should not share the same markdown logic simply because both are apparel. Children’s occasionwear has different purchase timing, size behaviour and family decision-making.

Measure what happened after the markdown

Do not judge the model only by units sold. Track gross margin, full-price cannibalisation, return rate, size sell-through and whether demand disappeared when the offer ended. Record what the team changed besides price.

The best outcome may be “do not discount yet.” That answer is valuable when it prevents a product-discovery problem from becoming a margin problem.

The TRYBUY.IN perspective

AI is most useful when it turns a vague markdown conversation into a reviewable decision with evidence, guardrails and a date to reconsider. Pricing authority should remain with the people accountable for the commercial result.

Explore TRYBUY.IN men’s shirts and check each live product page for current pricing and availability.

FAQ

Can AI set fashion prices automatically?

It can recommend actions, but automatic changes need strict policies, monitoring and human accountability.

Does slow stock always need a discount?

No. Search visibility, imagery, missing sizes or timing may be the real constraint.

How long should a markdown test run?

Long enough to gather meaningful evidence, with the period and success measures defined before launch.