Every commerce platform vendor is suddenly an "AI company." If you've sat through a vendor pitch lately, you've heard the promises: autonomous shopping agents handling 30% of transactions by 2030, conversational AI replacing traditional browsing, agentic commerce revolutionizing the customer journey.

The pitch decks are impressive. The revenue numbers are more sobering. 1.5% of retail sales run through AI platforms in 2026, and only 34% of consumers are willing to let AI make purchases on their behalf. Product recommendations, search, and AI-assisted content have been driving measurable results for merchandisers for years and rarely get mentioned in those pitches.

What Agentic Commerce Looks Like From the Catalog Side

Agentic commerce is the dominant AI narrative in enterprise commerce right now. AI agents autonomously browse catalogs, compare products, complete transactions, and handle post-purchase workflows without human involvement. Controlled demos work. Some platforms processed their first autonomous transactions in late 2025.

40% of shoppers are frustrated by the lack of human support in AI experiences, and 21% distrust AI recommendations outright. Amazon Rufus generates $10B in annualized sales -- less than 2% of Amazon's total revenue after years of investment and deep catalog integration.

For merchandisers, the more immediate problem is catalog readiness. AI agents need structured, accurate, complete product data to function. Inconsistent attributes, thin descriptions, and messy taxonomy produce wrong products, mismatched recommendations, and broken experiences. Getting the catalog into shape has value regardless of whether autonomous purchasing takes off -- that same data quality improves every other AI application on this list.

Where the Revenue Actually Is

Product Recommendations

Up to 31% of ecommerce revenue flows through recommendation engines -- and for most merchandisers, the biggest untapped opportunity isn't adding recommendations, it's extending them further into the journey. Homepage personalization and product page cross-sells are table stakes. Category page sorting is where the real lift hides. When AI surfaces products based on individual browsing and purchase behavior rather than manual ranking rules, you stop leaving conversion on the table for every customer who doesn't shop the way you merchandised the page.

The operational benefit is real too. Less time maintaining sort order rules means more time on higher-leverage work. Most commerce platforms support this natively or through established vendor integrations, and deployment is typically weeks, not a multi-quarter project.

Search

Site search users convert at 2-3x the rate of general browsers, which makes it one of the highest-ROI places to invest in the entire storefront. Most keyword-based implementations are leaving that conversion on the table. AI search delivers 3-10% lift by handling natural language, misspellings, and synonyms -- but the compounding benefit is that it learns. Every search session makes the next one more accurate.

A customer searching for "warm coat for skiing" on a keyword system gets exact-match results. An AI system understands they want insulated alpine outerwear and surfaces the right products regardless of how the descriptions are worded. For merchandisers, the bigger day-to-day win is what you stop doing: fewer synonym lists to maintain, fewer redirect rules to build, fewer zero-results reports to investigate every Monday morning.

AI-Assisted Product Content

Thin or missing product descriptions are a quiet revenue problem. They suppress search visibility, hurt on-page conversion, and are almost impossible to fix at scale with a human content team alone. Generative AI produces structured first drafts from product attributes that merchandisers review, adjust for brand voice, and approve. What used to take a copywriter a full day now takes an hour, sometimes less.

For commodity and mid-range products, this works well out of the gate. For premium or highly differentiated products, the review step matters more -- a flat, generic description undercuts the brand positioning that justifies the price point. The workflow that tends to stick is AI handling the volume across the long tail, with closer human review reserved for the products where voice actually matters.

Visual Search

AI visual search lets customers upload photos to find similar products. For fashion, home goods, and lifestyle retailers with comprehensive imagery and strong attribute tagging, it's worth testing in 2026. Customers who can't locate something through keywords can often find it visually, recovering revenue that would otherwise leave the site.

The limiting factor is always catalog coverage. Inconsistent photography or incomplete attribute data produces poor matches, and poor matches erode trust faster than having no visual search at all. For most other verticals, the ROI case is still thin -- worth watching, not worth funding yet.

Questions Worth Asking Vendors

The fastest way to separate a production-ready capability from a roadmap promise is to ask for three customer references running it at full scale -- not a pilot, not a controlled beta, but real transactions. If a vendor can't name three, move on.

Beyond references, push for specific numbers. "Improved product discovery" is a direction. "2.8% conversion lift over a 90-day A/B test" is something you can take to a budget conversation. Also ask what the implementation actually requires from your team. Tools that need significant attribute cleanup or taxonomy restructuring before they work aren't fast wins, and vendors don't always volunteer that upfront. Finally, ask what data the model trains on -- AI underperforms on messy or fragmented data regardless of how capable the underlying model is, and if your product data isn't clean, that's your problem to solve first.

Sequencing the Investment

If you're allocating budget for 2026, recommendations and search are the clearest starting points. Both have fast payback, established vendor ecosystems, and don't require organizational change to implement. If your current search is keyword-based, that's a Q1 project, not a roadmap item.

AI-assisted content is next, starting with the highest-volume categories where thin descriptions are actively suppressing discoverability. Visual search is worth a test if you're in fashion, home, or lifestyle and your catalog imagery is strong -- run it on one category before expanding.

Agentic commerce, conversational shopping assistants, and fully autonomous dynamic pricing are real technologies. They're also ones where the customer adoption curve hasn't caught up to the vendor enthusiasm. Monitor them, but the budget case isn't there yet.

A Note on Platform Architecture

Most commerce platforms are retrofitting AI capabilities onto architectures that weren't designed for the integrations merchandisers actually need. Connecting a recommendation engine, an AI search vendor, and a content tool often means three separate IT projects with custom API work for each.

Broadleaf Commerce was built API-first, which means AI applications access product data, catalog structures, pricing, and inventory through standard integrations. For merchandisers, that means faster vendor deployment, cleaner data flow, and less dependency on engineering to connect tools that should work out of the box. When agentic commerce or visual search reaches meaningful adoption, the architecture supports it without a platform migration.

The Bottom Line

Recommendations, search, and AI-assisted content are proven, deployable, and delivering returns right now. Visual search is approaching viability in the right verticals. Agentic commerce is real technology with a customer adoption problem that investing early doesn't solve.

The merchandisers getting the most out of AI in 2026 aren't the ones chasing the most ambitious roadmap. They got their catalog data clean, picked two or three proven tools, and executed. That's it.