Generative AI for Merchandising: A Practical Playbook

Last month, one of our clients at Creatuity faced a familiar problem. They’d just added 2,400 new SKUs to their catalog, and their merchandising team of four people needed product descriptions, category assignments, attribute tagging, and hero images for every single one of them. At their historical pace, that was eight weeks of work. They had three days before the site launch.

We used a generative AI workflow to handle the bulk of it. The team reviewed, edited, and approved the output. The whole thing took two days, and the quality scores from their internal review process came in higher than their manually-written baseline.

That’s not a story about AI replacing merchandisers. It’s a story about what happens when you pair good tools with people who know what they’re looking at.

What’s Actually Working Right Now

I want to skip the hype about what AI might do and talk about what we’re seeing work in practice across B2B e-commerce operations. Three areas are generating real results right now.

Product content at scale. This is the easiest win and the one I see teams adopt first. Large language models can take a spreadsheet of product attributes — dimensions, materials, specifications, use cases — and turn them into unique, SEO-friendly descriptions in seconds. One distributor we worked with processed 18,000 SKUs in four days. Their previous cycle for that volume was four months. The catch: you need clean attribute data going in. Garbage in, garbage out still applies, and it applies faster when AI is involved.

McKinsey published data last year showing that retailers using AI for product content generation saw a 30-40% lift in conversion rates on those pages compared to legacy content. The number makes sense when you consider that most B2B catalogs have thin or duplicated descriptions because nobody has time to write 10,000 unique blurbs.

Visual content creation. Generative image tools have gotten noticeably better in the past year. We’re seeing merchandising teams use them for lifestyle shots, background removal, and color variant generation without scheduling photography sessions. A industrial equipment client of ours started generating context photos — their products in use at job sites — because staging real shoots for thousands of SKUs was cost-prohibitive. The AI-generated images don’t replace professional photography for their top 50 sellers, but they’re perfectly adequate for the long tail of inventory that never justified a photo budget.

Assortment and space planning. This one’s newer and more interesting. Machine learning models can analyze your sales data, seasonality patterns, and customer segment behavior to recommend which products to feature, which to bundle, and where to position them on the page. Salesforce reported in their 2025 State of Commerce report that retailers using AI-driven assortment planning carried 15-20% less dead stock while maintaining or growing revenue. The mechanism is straightforward: the models spot demand signals that humans miss because there are too many variables to hold in your head at once.

Where People Get Stuck

The failures I’ve seen share a common pattern. A company buys an AI tool, points it at their messy data, and expects magic. When the output needs heavy editing, they conclude the technology doesn’t work.

The technology works fine. The problem is usually one of three things:

First, data quality. If your product attributes are inconsistent, incomplete, or trapped in PDFs and emails, no AI tool will save you. Clean your data first. It’s not glamorous work, but it’s the foundation everything else sits on.

Second, workflow integration. The teams that succeed don’t just generate AI content — they build review loops into their existing processes. A merchandiser generates descriptions in batch, reviews them in their PIM system, edits what needs editing, and publishes. If you make people log into a separate AI platform and copy-paste results around, adoption dies in a week.

Third, skill alignment. Your best merchandisers should be the ones training and refining the AI outputs, not some intern who got assigned “the AI project.” The people who understand your customers and your products are the ones who know whether an AI-generated description sounds right. That institutional knowledge is the ingredient the software can’t replicate.

The Human-in-the-Loop Reality

Here’s what I tell every executive who asks me about AI in merchandising: the goal isn’t full automation. The goal is giving your team superpowers.

A merchandiser who used to write 20 product descriptions a day can now review and refine 200. That means they spend their time on the strategic work — figuring out positioning, testing messaging, analyzing what converts — instead of typing the same basic details over and over. Their job gets more interesting, not less.

At Creatuity, we’ve built our AI services around this principle. We implement the automation, but we also train the teams using it, because the long-term value comes from people who know how to steer the tools, not from the tools themselves.

So Where Do You Start?

If you’re thinking about adding generative AI to your merchandising operation, here’s what I’d do:

1. Audit your product data. Pick one category and look at how complete and consistent your attributes are. Fix what’s broken before buying anything.
2. Start with product descriptions. It’s the lowest-risk entry point, the ROI is easy to measure, and you’ll learn how your team interacts with AI-generated content.
3. Build a review process before you scale. Decide who approves what, what your quality bar looks like, and how you handle edge cases.
4. Measure the right things. Don’t just track time saved. Track conversion rates, bounce rates on product pages, and internal quality scores. Those tell you whether the AI content actually performs.

The companies moving fastest on this aren’t the ones with the biggest budgets. They’re the ones willing to start small, learn from what doesn’t work, and keep iterating.

What’s your experience been with AI in merchandising? I’d like to hear what’s working and what isn’t — the real stuff, not the vendor demos.