Second in a series on building stronger AI-driven commerce with Rithum

At a glance  

• AI shopping agents evaluate your product data before they evaluate your brand. Incomplete or poorly structured catalogs get excluded from recommendations before a shopper sees them.  
• Most brands still rely on AI platforms scraping their product pages. Direct, structured feeds give you control over how products are represented inside AI environments.  
• Reaching an AI response and earning the recommendation are different problems. Without visibility into AI platform performance, improving placement becomes guesswork.  
• Payment infrastructure inside AI shopping environments is being built now. Brands that address the earlier stages will be positioned to capture those transactions as they scale.  

AI agents are already shaping what consumers see and buy. 70% of consumers have used an LLM to shop in the last three months, and 19% are now purchasing from brands they’d never encountered before those recommendations. But when it comes to making sure your products appear in those results, you’re operating against a black box. Are consumers seeing your products? Are they clicking? How will they purchase in an AI-driven environment? 

Demystifying that black box comes down to four stages: AI-ready product data, LLM connection, monitoring and optimization for AI engines, and in-LLM payments. Each one feeds the next. A gap at any stage means a leaky bucket for everything that follows. 

With Rithum, you can address each stage of the agentic commerce funnel today and prepare for where the space is heading next. 

Bad catalog data keeps products out of AI recommendations

AI shopping agents match product attributes against a shopper’s query. Missing specifications, inconsistent formatting, or outdated inventory signals knock products out of results entirely. 

When an AI assistant returns incorrect product information, shoppers blame your brand. Catalog quality becomes a brand trust issue, not just an operational one. Find out exactly what it’s costing you. 

The work on catalog structure, attribute coverage, and category alignment usually happens earlier in this process. Tools like Catalog Assist and Magic Mapper focus on those areas, handling attribute gaps and cross-channel categorization so product data is usable across AI-driven environments. With your catalog complete, structured, and current, you can tackle the remaining stages of the funnel with reliable inputs. 

Scraped data adds risk you cannot control  

AI platforms still rely heavily on crawling websites, marketplace listings, and third-party sources to assemble product information. Inconsistencies follow. Pricing, availability, and product descriptions can all drift away from the current state of your catalog.  

When that happens, the AI response reflects whatever information it was able to gather, not the current reality of your inventory. Very few shoppers click through to verify those details elsewhere, which turns the AI output into the primary version of the product. 

Rithum replaces that with direct, structured feeds into AI platforms. Rithum’s ChatGPT and Perplexity Feeds get your product data live and accurate on LLMs in three steps: your data is compiled into a feed, that feed is optimized for LLMs, then delivered directly for ingestion. Your brand owns its presence on LLMs instead of leaving it to crawlers. 

Rithum’s Stripe partnership extends this further by allowing brands to connect once and distribute product data across multiple AI platforms as they come online. Instead of building new integrations for each new surface, you can test across an assortment of LLMs and understand the ROI, all while keeping your product data updated and aligned. 

Getting into the system is not the same as getting selected  

AI-generated responses return a limited set of recommendations. Products compete for inclusion in that shortlist, and small differences in product data, relevance, or confidence signals can determine which products appear. 

Your feed is not a set-it-and-forget-it deliverable. You need to understand how your products are ranking across AI platforms and how those rankings shift over time.  

Our upcoming GEO (generative engine optimization) capabilities provide a way to track how products appear, move, and compare within AI-driven results. 

But monitoring only addresses half of the equation. Once you understand how you’re ranking, you need a way to improve those rankings. 

Rithum’s upcoming Performance Lab translates those signals into specific optimizations to improve how products appear in LLMs. Between GEO and Performance Lab, brands can move from “live but invisible in recommendations” to earning placement where it actually drives revenue. 

Because Rithum connects monitoring and catalog management in one place, you can act on performance signals directly. No exporting data, no cross-referencing tools, no guessing what to fix. There are plenty of myths about how agentic AI actually works. One of the most costly is assuming that presence alone drives results. 

Agentic checkout infrastructure is taking shape  

Payment is the final stage of the funnel: a shopper completes a purchase inside the same environment where they found the product. 

The current state of in-LLM checkout is uneven. Some platforms are testing in-conversation transactions, others are still building toward it, and some LLMs, including ChatGPT, have moved away from native in-chat checkout toward third-party app integrations instead. 

Rithum’s Stripe partnership provides the infrastructure for this layer. Product data flows from Rithum into AI platforms. When a transaction occurs, Stripe processes the payment while Rithum handles the necessary inventory updates and order orchestration. The brand stays the merchant of record and retains control of the post-purchase experience. 

Google’s UCP, available through Rithum’s Google Shopping feeds, opens another route, allowing brands to opt products into agentic checkout through AI-enabled search and Google Shopping, with support for loyalty programs and order management.  

Checkout only functions when the upstream stages are already working. Catalog data, platform access, and product-level performance all shape whether a shopper reaches the point of transaction. Getting those right now is the most direct path to capturing agentic commerce revenue as it grows. 

Rithum connects the full agentic commerce funnel  

Agentic commerce does not reward partial readiness. Every stage you leave unaddressed leaks value from the funnel. The teams gaining ground are the ones connecting these stages rather than running them as separate workstreams. 

Rithum and its upcoming agentic commerce capabilities connect these stages inside a single platform. Catalog improvements flow into AI feeds. Feed performance is measured. Optimization updates feed back into the catalog. When transactions occur, performance signals inform the next set of improvements. 

That full loop runs on one of the largest commerce datasets available: $50B+ in annual GMV, billions of SKU updates, and 3 out of 4 AI-driven optimizations accepted by clients. The scale and breadth to cover the full funnel from product data to AI-driven sale, in one system. 

The post How Rithum supports every stage of the agentic commerce funnel appeared first on Rithum.

First in a series of articles on how to incorporate stronger AI-driven commerce solutions

Nearly every commerce team we talk to has a version of this very same catalog problem: The catalog is technically live, and products are listed. The channel feeds seem to be running smoothly. But somewhere in the gap between “technically live” and “performing well,” a lot of margin is quietly disappearing.

Just as most commerce teams experience this, they also usually have the same root cause: bad product data. Things like missing attributes, miscategorized SKUs, incomplete descriptions, and values that don’t meet channel requirements.

It’s a common problem, but fixing product data at scale is often relentless manual work. And the teams responsible for it have other things to do.

That’s the problem RithumIQ’s was built to solve.

Your product data is a revenue problem

Your instinct may be to treat catalog hygiene as an ops issue, or something that gets cleaned up in a quarterly data audit.

But every SKU with a missing product weight creates a shipping estimation problem. Every product in the wrong category reduces agentic discoverability. Every listing suppressed for missing attributes is a hurt product sale. Multiply those issues across hundreds of thousands of SKUs and dozens of channels, and what looks like a data maintenance task starts showing up in your P&L.

Solving this manually doesn’t scale. You can’t audit 400,000 SKUs in any reasonable timeframe, and even if you could, the channels are constantly changing their requirements. It’s an unwinnable race.

Two features from Rithum’s one AI-driven solution

RithumIQ, the AI engine at the heart of Rithum, addresses the unwinnable race issue through two features that work on different ends of the same problem.

Catalog Assist handles the content side. It uses AI to detect missing or invalid product attributes and generates suggestions to correct them, based on your existing product data and channel-specific rules, not generic recommendations. For a product with incomplete color values, an incorrect size designation, or a description that fails a channel’s character count threshold, Catalog Assist surfaces the issue and proposes a fix. Crucially, it keeps humans in the loop: every suggestion requires approval before it goes live, and teams can edit directly within the Rithum platform. The goal is to remove the part of their job that’s purely manual, so your team focus on the decisions that actually require judgment.

For retailers, Catalog Assist solves a specific version of this problem in the supplier onboarding flow. New supplier products often arrive with catalog gaps that stall activation. Catalog Assist generates suggestions to fill those gaps and speed up the path to live listings by compressing a process that normally takes days into something that can happen in hours.

Magic Mapper handles the categorization side. Getting products into the right category on each channel sounds like a solved problem. It isn’t. Different channels have different taxonomies, different hierarchy depths, and different naming conventions. A brand managing presence across 20 marketplaces is essentially doing 20 categorization jobs. And the volume of SKUs means the work scales with the catalog, not with the team.

Magic Mapper automatically categorizes products across multiple channels simultaneously, and at scale. Teams can review recommendations, filter results, and selectively approve changes in bulk before anything goes live. It replaces hours of manual categorization work with a review step that takes minutes, giving you and your teams the confidence that comes from knowing nothing was pushed live without human sign-off.

The power of RithumIQ

Both capabilities are part of RithumIQ, Rithum’s AI intelligence engine, which was built on one of the world’s richest and most comprehensive commerce datasets: more than $50B in GMV annually, 4B+ SKU transformations daily, and 500M+ listing updates every day.

That scale is what gives us the best basis to ensure your product data quality is clean and actionable for all your automations. Catalog Assist and Magic Mapper aren’t drawing on a generic product knowledge base. They’re drawing on what correct attribute formats actually look like for a given category and what categorization patterns work on which channels. These recommendations are informed by how ecommerce actually operates at scale.

Rithum IQ’s suggestions are grounded in data that your team doesn’t have access to on their own. Three out of four Rithum clients accept RithumIQ’s recommendations 99% of the time.

The bigger picture of product data

The emergence of AI-powered shopping, or agentic commerce, is raising the stakes on clean, accurate, complete product content in a new way.

Rithum and Retail Dive’s recent consumer research found that 58% of shoppers blame the brand when an AI assistant gives them wrong product information. Simply put, if your product data is incomplete or inaccurate, it doesn’t just hurt you in search terms. It hurts you every time a shopper asks ChatGPT, Gemini, Perplexity, Claude, or any other agentic commerce tool to find them the best in your category. And it especially hurts you when AI answers give your shoppers a competitor’s product instead of yours, simply because their data is cleaner.

Catalog Assist and Magic Mapper are tools for today’s operational problems. They’re also the infrastructure layer for what comes next. An agentic shopping environment requires catalogs that are already in good shape, with accurate attributes, correct categories, complete content, and channel-compliant data across the board. The work to get there can get done faster with RithumIQ. It’s not AI as a feature announcement, but AI as the foundational layer behind your commerce initiatives.

Explore RithumIQ →

The post Your underperforming listings have a data problem. RithumIQ fixes it. appeared first on Rithum.

At a glance: 

• Shoppers who get a clear explanation of why an AI recommended a product are nearly 2x more likely to buy without verifying anywhere else. 
• Only 32% of shoppers named accuracy as the top trust-builder in AI recommendations. 49% chose a clear explanation of why a product was selected. 
• 47% of 28-to-43-year-olds say AI makes them faster decision-makers. Shoppers “very familiar” with AI tools are 3x more likely to purchase without verification. 
• When product data is wrong or incomplete, 80% of shoppers stay in the AI channel and ask again. 

A shopper asks ChatGPT for noise-canceling headphones for an open office, under $300. One result explains why it fits: isolates low-frequency hum, weighs less than comparable models, includes a transparency mode for conversations. Another lists a name, a rating, and a price. 

In a Rithum and Retail Dive survey of 1,046 U.S. and U.K. online shoppers, 49% named a clear explanation of why a product was chosen as the top trust-builder in an AI recommendation. Always-accurate information came in at 32%. And shoppers who get that explanation are nearly twice as likely to buy without checking anywhere else. 

Why shoppers value AI explanation over accuracy in product recommendations 

Shoppers expect AI to get the basics right. 67% named price as the top detail AI needs to be accurate on, followed by reviews and availability. But when asked what would most increase their trust, they reached past accuracy. 49% chose a clear explanation of why a product was selected. Always-accurate information came in at 32%. 

Any ecommerce team has seen this on a product detail page. Accurate price and clean specs keep a listing live. Rich attributes are what make it sell. The same applies to AI. An LLM builds its explanation from whatever product data it can find. If your listing includes driver size, noise cancellation type, and a note about comfort for all-day wear, the AI has something specific to say. If it doesn’t, the AI defaults to price. 

A jacket listed with fabric composition, weight, care instructions, and a note that it runs slim through the shoulders gives AI something to work with. A jacket listed as “men’s jacket, blue, available in S-XL” gives AI a price to compare. 

Newer brands with complete, attribute-rich product data already use this to their advantage, earning more persuasive recommendations than established names running on thin listings. When product data is wrong or incomplete, 80% of shoppers stay in the AI channel and ask again. The next answer is built on whatever data is available at that point. 

AI-powered shoppers buy faster and verify less 

47% of 28-to-43-year-olds say AI makes them faster decision-makers, compared to 21% of shoppers 60 and older. Shoppers who are “very familiar” with AI tools are 3x more likely to purchase without verification.   

For these shoppers, the explanation in the recommendation has to do the work that a product page, a review site, or a friend’s opinion used to handle. When the explanation falls short, the shopper moves to the next option in the response. There is no second visit, no follow-up search. The sale goes to whichever product explained itself best.   

How to optimize product content for AI recommendations 

Product content built for explanation earns stronger AI recommendations than content built only for visibility.   

• Enrich product attributes beyond the minimum required fields. Include use cases, compatibility notes, and sizing context. 

• Keep pricing and availability current across every channel where AI pulls data. 

• Monitor where shoppers verify AI recommendations and ensure your data holds up in those places too. Shoppers turn to search engines first at 28%, followed by online reviews at 19% and friends and family at 17%. 

• Test your own visibility: ask an LLM about your product category and evaluate whether your products appear with a clear, specific reason attached. 

• Prioritize data hygiene: validate and standardize titles, attributes, categories, and inventory/pricing sync so AI doesn’t amplify broken inputs across channels. 

Prioritize data hygiene: validate and standardize titles, attributes, categories, and inventory/pricing sync. AI can’t fix bad data. It can only move faster with whatever you give it, and when the inputs are off, that speed works against you. 

For a full breakdown of the data, download The New Discovery Engine report. 

The post Shoppers trust AI recommendations more when they explain why appeared first on Rithum.

At a glance: 

• 53% of shoppers trust AI tools, including an AI shopping assistant, as much as brand websites, according to a Rithum and Retail Dive survey. This trust is reshaping AI shopping verification behavior and expectations for AI in retail. 
• When shoppers verify an AI recommendation, retailer and brand sites rank near the bottom at 5%, showing how AI shopping verification is happening away from owned channels. 
• 64% of shoppers ages 18 to 27 say they’re likely to purchase based on an AI recommendation without verifying it anywhere else. 
• AI-referred visitors now convert 42% higher than non-AI traffic.
• Retailers like Walmart, Target, and Instacart are enabling purchases directly inside AI conversations, accelerating a future where the entire shopping journey happens in one place. 

A year ago, shoppers arriving through AI tools browsed but left without buying. A year later, those same shoppers are 42% more likely to buy than shoppers arriving through traditional channels¹. In the same month, Walmart deployed its AI shopping agent inside ChatGPT², joining Target, Instacart, and DoorDash in letting shoppers browse, compare, and buy products directly inside the conversation. 

In Rithum and Retail Dive’s survey of 1,046 online shoppers across the U.S. and U.K., 53% already trust AI tools as much as brand websites. And when they want a second opinion on what AI told them, they’re going everywhere except the brand to get it. 

When shoppers double-check, they go everywhere but the brand site 

When shoppers verify an AI recommendation, they’re choosing channels outside the brand’s control. Search engines are the top destination at 28%. Online reviews come next at 19%, followed by friends and family at 17%. Retailer and brand sites rank near the bottom at 5%. 

A brand’s own product page, no matter how thorough, is still the brand talking about itself. Shoppers want independent voices, and they’re finding them everywhere else. 

That puts more pressure on the product data traveling through those channels. If the search engine is the second stop after AI, the data you’re pushing into Google, Bing, and other platforms needs to be accurate and complete. If a shopper pulls up a review site and finds specs that conflict with what the AI told them, the brand absorbs that cost. In the same survey, 58% of shoppers said trust in the brand decreases when an AI recommendation contains incorrect product information, and 16% abandon the purchase entirely. 

Brands have the answers, but shoppers are asking somewhere else 

A shopper asks an AI tool to recommend a running shoe for flat feet under $150. Three options come back. The shopper likes one but wants to confirm the arch support claim before buying. 

They type the product name into Google. They scan a couple of review sites. They text a friend who runs. The brand’s product page may have the most detailed answer to their question, but the shopper has already moved on to other sources. 

Product information accuracy across your entire distribution footprint now carries more weight than the quality of your own site experience. Feeds, marketplace listings, third-party retailer pages, and structured data that AI tools can parse all shape what the shopper encounters during verification. The brands investing in that full footprint are the ones staying in the consideration set. The ones focused primarily on their own site are building for a shopping journey that fewer customers follow. 

A growing share of shoppers skip verification entirely 

Among shoppers ages 18 to 27, 64% say they’re likely to purchase based on an AI recommendation without verifying it anywhere else. Higher-income shoppers are twice as likely to trust AI without visiting another site. And across all demographics, 32% say they spend less time browsing other sites after using an LLM. 

For these buyers, the AI recommendation from an AI shopping assistant is the decision. The brand site is largely absent from it. And shoppers who verify through search and reviews encounter a broader set of options than they would on a single brand’s site, giving unfamiliar brands a real opening to enter the consideration set with accurate, well-distributed product data. Shoppers who skip verification altogether are relying entirely on whatever the AI already knows about your product. 

The verification step itself is disappearing 

64% of shoppers already take AI at its word. The platforms coming next are built to make that feel even more natural.

AI agents could mediate $3 trillion to $5 trillion of global consumer commerce by 2030, according to McKinsey and Co. Two competing open protocols are already live and processing transactions end to end: OpenAI and Stripe’s Agentic Commerce Protocol and Google’s Universal Commerce Protocol. The AI agent handles product discovery, comparison, checkout in one place. 

Now picture that same running shoe shopper six months from now. They ask ChatGPT the same question. A product card appears with an image, price, and a “Buy” button. They tap it, confirm their saved payment method, and the order ships. The entire transaction happened inside a single conversation. 

The shoppers who still verify aren’t going back to the brand to do it. They’re checking reviews, search results, other people. And as AI agents take on more of that process, the brand’s window to influence the answer gets smaller. The data has to be right before the question is ever asked.”

Your product data is now your pitch to an AI buyer that will never visit your homepage 

Getting product data right across every channel is the minimum. It’s expected. The question is where that data lives: search engines, review platforms, marketplace listings, and the structured data feeds that AI agents read when they decide what to recommend. AI-readable product content needs to be complete, consistent, and built for machines to parse. For a growing number of shoppers, that content is the only version of your brand they’ll ever see. 

The distance between discovery and purchase is collapsing. Sometimes it’s a single conversation with an AI agent. The brands feeding that conversation with accurate, well-distributed product data are the ones the agent recommends. 

For a full breakdown of the data, download The New Discovery Engine report. 

Sources:
1: https://www.retailtouchpoints.com/features/the-agentic-commerce-paradox-its-already-here-and-its-also-still-evolving/618945/  
2: https://www.cbsnews.com/news/ai-agentic-artificial-inteligence-what-is-it/

The post Shoppers are verifying elsewhere (away from brands and retailer sites)  appeared first on Rithum.

TL;DR 

• Product journeys, from discovery to decision, are shifting to AI. Among AI-active shoppers, 90%+ use LLMs to research products and compare options, and 53% use them to decide where to buy. 
• AI is diminishing buyer loyalty. 19% say they now buy from brands or products they had not heard about before, and 13% say they are more likely to switch retailers or products after using an LLM. 
• Brand and retailer sites have less time to influence the decision. 32% of shoppers spend less time browsing other sites after using LLM tools, and only 5% verify AI shopping information on a retailer or brand site. 
• Product information now plays a bigger role in trust. 49% say a clear explanation would do the most to increase trust in an AI recommendation, 67% say price is the most important detail to get right, and 58% say trust in the brand drops when an LLM provides incorrect product information. 

Rithum’s new report, The new discovery engine: How consumers are using AI to find, trust, and choose brands, and what’s at risk for those they never see, has a clear message for retailers and brands: the shopping journey is no longer confined to shelves, search results, category pages, or product detail pages. 

Based on a survey of 1,046 online shoppers in the U.S. and U.K., the report shows how LLMs have become the entire shopping journey and a one-stop shop where products get researched, compared, narrowed down, and chosen. Among AI-active shoppers, more than 90% use LLMs to research product information and compare options, while 53% use them to decide where to buy. By the time a shopper lands on a product page, the filtering may already be over. 

That shift to AI is creating room for brands that were not previously in the mix. In the survey, 19% of shoppers say they now buy from brands or products they had not heard about before. Established brands are facing a tougher version of the same market. Recognition still helps, but it has less force when AI is doing more of the sorting, ranking, and explaining before the click. 

The shortlist is forming earlier  

AI shopping adoption is the starting point for 80% of shoppers ages 18-to-27 and 80% among shoppers ages 28-to-43. Among households earning $100,000 to $150,000, it reached 84% adoption. These are commercially important shoppers, and they are already weaving LLMs into their purchase journey. 

The filtering that used to happen across tabs, retailer sites, and review pages is now happening inside LLMs. A product that appears high in the response moves ahead. One that doesn’t can disappear before the shopper has seriously considered it. 

ChatGPT’s product comparison feature adds to that shift. Shoppers can compare products side by side inside the chat, with price, features, reviews, and other details presented in one place instead of scattered across multiple retailer tabs. 

More than half of shoppers already trust AI tools as much as brand websites, and among high-income households, confidence in AI accuracy climbs as high as 80%. That trust gives LLM recommendations real weight early in the decision process, according to the report. 

Retailers now have more riding on how products are represented on their page, even before a shopper lands on the site. Brands face the same pressure, but with fewer natural intervention points. A retailer may still appear as the place to buy, even if a brand is filtered out earlier. If a brand’s product information is incomplete, inconsistent, or hard for AI to explain, it can be dropped from consideration before its own product page or brand story has a chance to influence the decision. 

New brands are finding room where familiar brands once had an easier ride 

You can see the LLM effect far beyond just initial research, with ripples into what shoppers buy. Nineteen percent of shoppers say they are more likely to buy from brands or products they had not heard about before if an LLM suggests it. Another 13% say they are more likely to switch retailers or products after using an LLM. Together, those numbers create a shopping environment where familiar brands have less room to rely on recognition alone. 

That creates an opening for challenger brands. A newer brand does not need years of broad recognition to get in front of a shopper. It needs usable, consistentproduct information and enough context for AI to present it clearly and convincingly. 

Established brands have less room to lean on familiarity alone. Customer loyalty still helps, but it no longer ensures that they go to your site first. Nearly half of shoppers say a clear explanation of why a product or brand was chosen would do the most to increase trust in an LLM recommendation. What carries weight here in an LLM recommendation is not name recognition but whether the recommendation feels specific, informed, and ready to act on. 

Brand-owned sites get fewer chances to influence the outcome 

The shopping journey used to leave more room for second thoughts. A shopper could open a few tabs, compare prices, read reviews, leave, come back, then change course. LLMs have shortened that process.  

In the survey, 32% of consumers say they spend less time browsing other sites when using LLM tools to shop. Another 36% say they make faster decisions, while 34% say they feel more confident about their purchases.  

These three stats don’t live in a vacuum. They indicate a continual trust-building experience for the shopper: they’re saving time, they’re finding what they need faster, and they feel better about their purchases. Why would they leave that experience to go back to a retailers website? 

The same pattern appears in how people verify what they see. Shoppers who double-check an LLM recommendation rarely begin with looking for confirmation on a brand or retailer site. Twenty-eight percent turn to search engines (which is likely also relying on AI tools), 19% specifically look for online reviews, 17% ask friends and family, and only 5% go to a retailer or brand website. A beautiful, brand-forward website won’t convince them to buy your product. They won’t even see it. But a PDP with in-depth specifications, GEO-optimized keywords, and highly relevant descriptions will impact consumers’ decisions, even if they don’t see the page. 

The recommendation is only as strong as the product story behind it 

Ask an LLM why it chooses one product over another, and it has to build that answer from the product facts it can find: materials, dimensions, compatibility, intended use, etc. The recommendation that an LLM givesis assembled from those pieces in real time. 

For brands, that raises the standard for product content. Copy, attributes, use cases, and supporting details are no longer sitting off to the side as content maintenance. They are becoming part of the recommendation itself. When the product story is thin, generic, or inconsistent, the answer reads that way too. 

Retailers feel the same pressure across the assortment. Pricing, inventory, attribute completeness, and feed quality all shape how products are represented before a shopper ever reaches the site. Anyone who has spent time inside a catalog has seen how quickly that can start to fray. A bad price, a missing dimension, or stale availability can make a solid product look less reliable than it is. 

The harder question is whether the product story still holds together everywhere that LLMs are pulling from. This includes product content, syndication, pricing, availability, and the systems that keep those details aligned. It also includes sources brands and retailers cannot fully control, such as reviews, forums, and social discussion. When those external signals surface alongside structured product data, inconsistencies become more visible. That makes it even more important for the information you do control to be accurate, complete, and easy for AI to explain. LLMs only give recommendations they can trust, based on the information that holds it together. 

Trust is moving closer to the data itself 

The survey leaves little ambiguity on price. In an AI shopping recommendation, 67% say it is the most important detail to get right. Reviews, availability, where to buy, and technical specifications all come after it in the list of prioritization 

That order will feel familiar to anyone who has watched shoppers abandon a cart over a mismatch or lose confidence over a number that does not look right. A wrong price or stale detail does not stay in the background. It becomes part of the recommendation, which means it becomes part of the shopper’s impression of your brand. 

The report puts numbers behind that. When an LLM provides incorrect product information, 58% say trust in the product or brand decreases, and 16% say they leave the purchase altogether. 

At that point, the issue is no longer confined to data quality. The recommendation may come from the model, but shoppers are not spending time sorting out where the error began. They decide whether the information feels reliable, and the brand lives with the result. 

The next phase is close enough to shape decisions now 

The report also looks ahead to a shopping flow where the model takes on more of the decision itself. More than 25% of AI-active shoppers say they are already very likely to hand purchasing decisions to AI, and another 39% say they are somewhat likely to consider doing this, if and when it’s available. Among the most AI-active shoppers, 65% say they are very or somewhat likely to use an AI agent that would buy for them. 

What AI sees already shapes what shoppers buy. Thin product content, stale pricing, patchy attributes, and inconsistent availability all weaken the recommendation before the shopper has done anything beyond type in a prompt. 

The priorities are clear. Keep your product story consistent. Keep pricing accurate. Keep availability current. Make products easier to compare, easier to explain, and less likely to be misread. New brands already have more room to enter the conversation. Established brands have less room for weak information, stale details, or missing context. 

For more details on the survey and a full breakdown of the results, download the report here.  

Methodology 

Rithum’s 2026 report is based on a survey of 1,046 online shoppers in the U.S. and U.K. Some questions look at behavior in the last 3 months, some category questions use the last 6 months, and some trust and behavior questions are broader and are not tied to a single recall window. 

The post A brand your customer had never heard of just won the sale appeared first on Rithum.

A deep dive into the machine learning models behind more accurate ETA predictions 

TL;DR 

• Many retailers still build delivery promises by combining a processing window, a carrier transit estimate, and a buffer. 
• That approach starts to break down in supplier-fulfilled ecommerce, where processing times vary by warehouse, backlog, fill rate, and current operating conditions. 
• Rithum’s Delivery Promise uses machine learning to predict processing time and transit time separately, which produces a more realistic ETA. 
• The advantage comes from the data behind the prediction, especially supplier-warehouse visibility across the network. 
• Package Predictor is separate from Delivery Promise, but it improves shipping-cost decisions by predicting package weight and dimensions more accurately. 

For a lot of retailers, delivery promise still starts with simple math: three days to process, five days in transit (call it eight) and move on. That kind of estimate can work for a while, especially when the fulfillment network is relatively predictable. But it gets shaky fast in supplier-fulfilled ecommerce, where one order may ship from a warehouse running normally, while the next may come from a location dealing with backlog, fill rate issues, or a completely different operating rhythm. 

Carrier performance adds another layer of variability, changing by service level, lane, and time of year. So, while a static estimate may look clean in the system, it can end up far removed from how an order is actually likely to travel. 

That is the problem that machine learning is helping solve in Rithum’s Delivery Promise. Instead of relying on one rule that tries to cover everything, Delivery Promise can use historical and real-time data to make a better estimate of how a given order will progress through fulfillment and transit. And because Rithum sits across a broad supplier network, the model can work from a fuller picture than most retailers have on their own. 

Why do static ETA estimates break down in supplier-fulfilled ecommerce networks?

Static promise logic assumes fulfillment behaves like a fixed process. In supplier-fulfilled ecommerce, it rarely does. 

If you are relying on a standard processing window and an average transit estimate, you are treating every order like it moves the same way when it doesn’t. Supplier performance is rarely consistent across the network; one warehouse may be operating normally while another is slowed by backlog, labor constraints, fill rate issues, or the type of orders coming through. 

You may not know ahead of time which warehouse will handle the order. That alone makes it tough to pin processing time to one standard number. 

A simple estimate is easy to put in place. Keeping it useful is another story once the network gets bigger and more complex. 

How does machine learning improve delivery promise accuracy?

Rithum’s approach starts by separating two questions that many systems treat as one. 

Delivery Promise uses predictive machine learning models to predict how long an order is likely to take to process before it ships and how long it is likely to take in transit after it leaves the warehouse. Those two steps are connected, but they are not driven by the same conditions. 

Processing time depends on what is happening inside the supplier’s operation. Transit time depends on what happens once the package is in the carrier network. Treating them separately gives you a more realistic ETA than rolling everything into one estimate. 

Rather than forcing every shipment through the same assumption, the system can use historical performance and current conditions to make a better prediction for the specific order in front of it. 

Why does Rithum’s network give the model a clearer view of ETA risk?

The model only gets you part of the way. ETA accuracy also depends on how much of the fulfillment picture the system can actually see. 

In supplier-fulfilled commerce, retailers are often working with gaps. They may know the supplier, but not the warehouse that will ship the order. Even when they know the likely location, they may not have a current view of backlog, fill rate, or how that warehouse has been performing under similar conditions. 

Rithum works from a broader set of signals across its network, including where inventory sits, which warehouse is likely to fulfill the order, how that location has performed in similar situations, and what current conditions look like in real time. 

That broader view is the real advantage. A retailer may know its own order history. Rithum can pair that with network-level visibility into supplier warehouses, which gives the model a stronger read on where risk is building and where a promise is more likely to hold. 

Why do more accurate delivery promises help at checkout?

At checkout, the estimate has to hold up. When the date is built from a simple estimate, retailers usually have to play it one of two ways: pad it to be safe, or tighten it and hope the order moves the way the system expects. Neither is a great option. 

With a better prediction behind it, the system can generate a date based on how that order is likely to move through fulfillment and transit under current conditions, rather than applying one broad assumption across the board. 

That gives retailers a better shot at posting a date that can hold up without pushing it farther out than necessary—protecting checkout conversion rates while safeguarding brand trust. 

What is Package Predictor, and how does it connect to Delivery Promise?

Package Predictor is related to Delivery Promise, but it is not doing the same job. 

Delivery Promise is trying to predict timing. Package Predictor is trying to predict how the shipment will actually be packaged, especially when it comes to weight and dimensions. 

That is a different problem, and it affects a different part of the shipping decision. The size of the box usually is not what determines how fast something moves through the network, but it does affect shipping cost and service selection. 

That is where things get messy in dropship. You may have catalog data for an item, but not enough detail to know how a real order will be packaged, especially when multiple items ship together. And when that data is coming from a broad supplier base, it’s often incomplete, inconsistent, or both. 

Package Predictor gives the system a better way to work through that uncertainty. It looks at historical shipment behavior and uses those patterns to make a better estimate than a manual default can. 

How does Package Predictor improve shipping-cost decisions?

Package Predictor gives the rate estimate better information to work from. If the estimated weight and dimensions are wrong, the estimated shipping cost is wrong. And once the cost estimate is off, it becomes much easier to choose the wrong service or make a fulfillment decision that costs more than expected. 

Package Predictor improves both accuracy and coverage, reducing the number of cases where the system has to fall back to broad supplier-level or retailer-level defaults. 

Better package estimates sharpen carrier-rate accuracy—and that’s what drives smarter shipping decisions. 

Why rising shipping complexity puts more pressure on ETA accuracy and package data

Shipping has gotten more expensive in more complicated ways. Carrier agreements are more layered than they used to be, dimensional-weight charges continue to hit certain shipments harder, and small inaccuracies in the data can create bigger downstream problems than they once did. 

That puts Delivery Promise and Package Predictor under a brighter light. One is trying to generate a delivery date that holds up in a more variable network. The other is trying to improve the package data behind the rate estimate, so the shipping decision is built on something more reliable than a rough default. 

When costs tighten and variability increases, the quality of those inputs is what ultimately protects your profit margins. 

What retailers should do next if static delivery estimates are starting to fall short

The old approach gets harder to trust as fulfillment spreads across more suppliers, more warehouses, and more moving parts. 

If you are trying to improve ETA accuracy in supplier-fulfilled ecommerce, broad averages only get you so far. Better predictions come from data that reflects how fulfillment and transit are actually performing. 

Machine learning becomes useful when it can model that day-to-day variation instead of smoothing it over with one broad rule. That is especially true in supplier networks, where the operating conditions behind an order can change from one warehouse to the next. 

The prediction is only as strong as the visibility behind it. The clearer the view into supplier warehouses, fulfillment conditions, and transit performance, the stronger the delivery promise becomes. 

To learn more about how Rithum supports delivery promise accuracy and shipping-cost decisions, schedule a demo for a closer look at Delivery Promise and Package Predictor. 

Kyle Knoblock is Staff Product Manager, Retailers at Rithum.

The post From guesswork to precision: How AI improves delivery promise accuracy appeared first on Rithum.

Shoppers rarely start their online search with a perfect query. It’s usually an idea of something, like “I need a couch good for a small apartment,” or “storage that can fit underneath a bed.” And Generative AI is making it easier for shoppers to get trustworthy recommendations, from those questions. 

In the Gartner report, Use GenAI to enhance commerce search and discovery experiences, they write: “Generative AI (GenAI) offers customers the ability to search by asking natural language questions or queries of what they intend to buy.” GenAI turns a shopper’s question into a useful set of products and context. Then there’s agentic AI, which goes one step further and takes actions, like narrowing choices and moving a shopper toward checkout. That flow still starts with discovery, and this report focuses on the GenAI patterns that support it. 

With GenAI, results are accompanied with context alongside products with recommendations connected to the shopper’s question. With guided selling, a shopper’s questions narrow down options, so it is important to have the most accurate and up-to-date product catalog information to ensure your products are included in those responses. 

When selling across multiple channels, consistency gets harder to maintain. If a shopper is directed to a webpage only to find that the product doesn’t match or is out of stock, they move on to another option and someone else gets the sale. GenAI can help shoppers decide faster. 

Download the Gartner report to see the full examples, diagrams, and recommendations.  

Gartner, Use GenAI to Enhance Digital Commerce Search and Discovery Experiences, By Aditya Vasudevan, Mike Lowndes, 27 November 2024 

GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved. 

The post When shoppers phrase their search as sentences, your product catalog has to be the answer  appeared first on Rithum.

Last week OpenAI said it will start testing ads in ChatGPT. Within the hour, we had clients reaching out for our POV. And we love that: our clients are invested in optimizing every AI advancement, and we get excited about this stuff. But in the AI commerce world, there’s often a chasm between “announced” and “available.” In that in-between space, agencies pitch one thing, vendors promise another, and a whole lot of smoke-and-mirrors compete for your attention (and money). Meanwhile, there’s a black box of information from the actual AI platform.

In the near term, that uncertainty creates a very practical question for brands and retailers: what will it take to protect and grow AI share of voice in AI-driven shopping experiences as the rules change?

The entire ecommerce industry sees the same news about ads in OpenAI and gets excited. But no one truly knows what the buying model will look like (bidding? context targeting? something new?), what reporting will exist, or how the platforms will enforce relevance and transparency. 

What I, and others watching this announcement closely, do know is that ads inside an AI experience carry a unique risk: trust. In traditional search, users expect sponsored results. In an AI answer—where people are relying on the system to do the thinking—the separation between recommended and paid has to be unmistakable, or the whole experience will feel compromised. If ads are going to work in AI without breaking the product, they need to be tightly contextual—shown because they fit the answer, not instead of the answer. OpenAI has become a trusted source of information, and anything working against that, especially in commerce, erodes confidence in the platform itself.

Build AI readable product information now. 

Ads are the most recent example of how the mechanics and narrative of agentic commerce are being written week-by-week. Every part of agentic commerce is evolving in real-time, which creates constant pressure for brands and retailers to react just as quickly. When “someday” and “today” send mixed messages in bombastic headlines, how can you keep up?

The best way is to first recognize some of the uncomfortable truths, and what you can do to actually be ready for every stage of agentic commerce, whether it’s here or coming soon.  

Ads are the most recent example of how the mechanics and narrative of agentic commerce are being written week-by-week. Every part of agentic commerce is evolving in real-time, which creates constant pressure for brands and retailers to react just as quickly. When “someday” and “today” send mixed messages in bombastic headlines, how can you keep up?

You don’t need to predict the future to prepare for it. You just need to build the foundation that makes you recommendable, no matter what comes next. 

Another uncomfortable truth: right now, measuring AI share of voice is not a clean science.

The easiest version of AI visibility measurement is what many teams are doing today: you “put yourself into AI and see what comes back.” Run prompts, record results, compare competitors, and repeat.

That’s not wrong, but it’s also not perfect. Different people will see different answers. They’re typing different prompts, they’re in different locations, they phrase their prompts slightly differently, and the model may personalize based on prior behavior. When the platform itself is constantly changing, any number of variables can impact the true answer.

And if you buy a monitoring tool, there’s always a chance those variables just get wrapped in a prettier dashboard.  

It’s also difficult to scale. If you want product-level visibility, you’d need to test countless variations across categories, attributes, and intents. That becomes unwieldy fast, especially for large catalogs. And even if you spend the money to get those metrics, you still have to normalize the long-tail reality of prompts: three people ask the same thing in three different ways.

There’s no single “AI share of voice number” that behaves like a paid search impression share metric. And most brands aren’t directly feeding product data into large language models at scale yet. The ecosystem isn’t “plug in your catalog and win” (even if the pitch decks make it sound that way).

So, again, the path to visibility today is about teaching the AI to understand your product the way a real shopper evaluates it. Which leads to a practical directive: Make your product information crawlable, consistent, and verifiable.

The hardest truth: You don’t have an AI visibility problem. You have a PDP and trust problem.

If AI can’t confidently show you to your prospective customers, you don’t have an AI visibility problem. You have a truth and consistency problem across your PDPs, your marketplace listings, your imagery, your specs, your UGC footprint, and more.

If your data doesn’t match from one channel to the next, the AI sniffs out the inconsistencies, and it leaves you out of the answer. 

As the internet fills with content that AI recognizes as AI, then AI itself looks harder for authenticity.

If you’ve googled recently, you’ve seen Reddit going up higher in traditional search and even community opinions outranking polished brand websites. This on its own is a reflection of how AI is evolving: as the internet fills with content that AI recognizes as AI, then AI itself looks harder for authenticity. A forum thread, a review, a creator’s demo… these can’t be generated at scale by AI. And agentic commerce-based AI prioritizes authenticity over all else. Public conversations are now a material input into how AI recommendation systems form confidence. This is a huge hit to your brand voice.

A social post that says: “Buy the Jansport backpack for back-to-school” is brand-forward, but AI-invisible. A social post that says: “Get your water-resistant fabric, stain-resistant bottom, five compartment backpack that fits a 16” laptop and is overhead-bin friendly” is feature-forward and matches how people actually ask AI to shop.

In an agentic world, your brand name doesn’t bring the same level of trust. But strong features and PDPs mapped to the agentic commerce way of shopping create the new trust layer that AI looks for.

Trust is the new KPI. 

The agentic commerce era is uniquely exhausting. By the time you’ve finished reading the latest big AI announcement, the internet has already claimed that announcement is a new playbook. Even if that playbook is built on pure speculation.

Partial rollouts, unclear business models, lofty promises, and black-box systems have dropped the ecommerce industry into an uncomfortable, hazy ground of AI myths and truths. There are a lot of unknowns. But you can absolutely prepare for those unknowns in ways that will compound your readiness for wherever the industry advances.

Every layer of agentic advancement returns to a single concept: trust. 

Trust is why agentic shopping feels magical to consumers. 
Trust is why ads will look different than ever before. 
Trust is why UGC matters. 
Trust is why clean, consistent product truth is vital. 

In the agentic web, you have to earn inclusion. And inclusion is granted by systems that reward clarity, consistency, credibility, and proof.  

The headlines will keep shifting week by week. The black boxes will stay dark. But the flickers of guiding lights to follow shine from your clean data, consistent product truths, and a partner to help make sure you are staying at the cutting-edge. That is the best way to compound trust, and trust is what we know for a fact will always help you earn inclusion, even as the rules keep changing.

The post The uncomfortable truths of the agentic commerce era appeared first on Rithum.

• Snapshot: eTail Insights surveyed marketplace and ecommerce leaders for Mastering the marketplace in retail and ecommerce to benchmark marketplace trends including what’s working, what’s breaking, and where budgets are going next. 
  Why it matters: These benchmarks show how marketplace teams describe their day-to-day reality right now. 

• Listing optimization no longer separates anyone. In the survey, every team rates its listing optimization capability as good or better, which tells you where the “easy wins” have already been harvested. 
  Why it matters: Listings may look “done,” but they stop being the lever that moves growth. 

• Retailer and brand teams lose time and margin on coordination. Inventory coordination across platforms ranks as the top challenge at 40%, followed by marketplace data integration at 33%. 
  Why it matters: When systems disagree, the fallout lands in fulfillment, customer service, and returns. 

• Leaders are budgeting for trust optimization above all else. Over the next 12 months, the top investment is customer service and review management systems at 51%, ahead of marketplace advertising and content expansion at 46%. This makes sense, as credibility optimization is key to meeting both customers’ and agentic shopping tools’ high standards. 
  Why it matters: Teams are funding the places where customers notice problems first and where reputations change fastest. 

• Profit pressure is shaping the stack. Competitive and dynamic pricing tools sit close behind at 41%, a signal that fees and price competition are forcing sharper math. 
  Why it matters: If teams can’t see the true cost per order, pricing tools can push them into underpricing. 

• Marketplaces are moving from “channel” to “center of gravity.” 81% expect marketplace sales to become more important to overall strategy, and 73% expect that shift within the next year. 
  Why it matters: Marketplace execution is becoming core operations, not an add-on project. 

A customer adds a refrigerator water filter to their cart. They’ve bought the brand before, and everything looks familiar. One click from paying, they pause. The model name is close, but not identical.  The customer taps the chat bubble in the corner of the screen to make sure they’re purchasing the right part.  

That bubble used to connect them to a live agent. Now it’s often an AI assistant, trained to answer product questions quickly, so the shopper doesn’t abandon checkout. 

“Is this the filter for model X?” 

The assistant replies instantly: “yes, it’s compatible.”  

But it’s not.  

The filter fits a similar model with a near-identical name. The product listing doesn’t clearly distinguish the two, so the AI fills in the gap with the most plausible-sounding answer. The customer buys anyway. The filter arrives, doesn’t fit, and gets returned. 

The return reason isn’t “wrong model.” It’s “defective.” The frustration isn’t just with the product. It’s with the guidance the shopper trusted. 

This is the kind of mistake customers try to avoid. When it happens anyway, they don’t blame the model, they blame the seller. Marketplaces penalize getting it “almost right” with returns, bad reviews, and support costs. The warning appears in performance long before you see it on a dashboard. 

The new marketplace problem: the storefront that moves faster than operations 

Most organizations aren’t dabbling in marketplaces anymore. Nearly everyone sells on Amazon. Most also sell on Walmart.com, according to eTail’s Mastering the marketplace in retail and ecommerce report. 

But while most respondents say their marketplaces strategy is at least somewhat effective for revenue, there is a deeper problem. Each marketplace comes with its own rules and requirements. Retailers and brands don’t describe their biggest marketplace challenges as “getting listed” (the industry has evolved past that point). They describe them as keeping platforms aligned. 

Inventory coordination across platforms ranks as the top challenge at 40%. Marketplace data integration follows at 33%. 

Those aren’t minor headaches. They sit underneath the moments customers remember: the delivery promise that never was, the unwelcomed out-of-stock surprise, the placement part that “should have worked,” or the one-star review that deters others from buying your product. 

Meanwhile, marketplace importance keeps rising. 81% say marketplaces will become more important to overall strategy, and 73% expect that shift to happen within twelve months. So, as opportunity grows, the friction grows along with it. 

Where marketplace execution starts to separate teams

Listing optimization is not the weak point: every respondent rates their listings as good or better. But when everyone feels strong in the same area, it is no longer a competitive advantage. Instead, the advantage shifts to what’s more difficult to keep consistent at scale, such as inventory accuracy, data consistency, review health, margin control, and service quality. 

It also moves to the places where organizations admit weakness. In this case, one-third rate “review and reputation management” as fair or poor. 

That gap matters because reviews aren’t just social proof. They shape ranking, conversion, and what shoppers learn before they buy. They also feed the summaries shoppers increasingly rely on, whether that’s a marketplace recap, a review highlight, or an AI-style shopping assistant answer. A shopper who doubts your listing doesn’t read your FAQ. They read your reviews. 

Marketplaces reward speed, but trust decides who keeps the sale 

Marketplaces raise expectations without lowering complexity. Shoppers expect quick answers, fast shipping, and painless resolution even when the catalog and fulfillment reality are messy.  

Trust decides whether a retailer or brand keeps the sale. On marketplaces, customers judge trust through what happens when something goes wrong and what other buyers say happened to them. That’s why the report’s investment priorities matter. Leaders are investing in customer service and review management systems at 51% over the next year, ahead of advertising and content expansion at 46%. 

Competitive and dynamic pricing tools follow close behind at 41%, underscoring how quickly the fees and shifting costs can erode margins. Teams don’t lead with service systems unless service continues to break. 

The catalog is the experience now 

The story of a mismatched water filter isn’t really about AI. It’s about what happens when product information fails under pressure and gets repeated across channels. 

Marketplaces have always punished unclear product information, but the old failure mode looked different. A customer read the listing, guessed, and learned the truth when the package arrived. 

Now the failure happens earlier. Customers ask questions mid-journey and even when the catalog is incomplete or inconsistent, the assistant still answers. When two SKUs look nearly identical, the assistant still picks one. When variation structure is messy, the assistant still tries to resolve it. If your product data leaves room for interpretation, the assistant will fill it, and you will own the consequences. 

What breaks first when teams scale marketplaces 

Most marketplace failures don’t start with a dramatic crash. Then one day, someone notices reviews shifting in tone. Support tickets spiking. Returns rising. A marketplace rep flags an issue. A finance team sees margin slip and can’t tie it to one decision. 

Coordination can be the biggest bottleneck. Inventory coordination and marketplace data integration top the list. If you can’t keep systems aligned, you can’t keep promises aligned. 

Three shifts that make marketplace growth easier to scale 

Most teams don’t need a radical reinvention. They need to tighten the parts that create expensive downstream effects. 

These three easy shifts  map directly to what leaders say is slowing them down. 

1) Treat inventory coordination like a revenue problem 

Inventory coordination across platforms ranks as the top marketplace challenge. This sits at the center of marketplace performance. 

When inventory is wrong, you don’t just lose a sale. You lose ranking and trust. You waste customer support time. You pay for returns. You create reviews that never should have existed. 

Start by making your inventory view boring and reliable. Establish one source of truth for what’s sellable. Push that truth to every platform on a cadence you can defend. 

That shift prevents the “invisible losses” teams learn to tolerate, including cancellations, refunds, angry reviews, and time spent explaining mistakes. 

2) Treat review management as a system, not a reaction 

One-third of survey respondents rate “review and reputation management” as fair or poor. That’s one-third of teams saying they still treat reviews reactively—and know that it’s hurting them. 

Reviews deserve a better approach because they aren’t random. They’re patterns. 

A shopper complains that the filter “didn’t work.” Another says the product “felt defective.” A third says the listing “misled them.” The issue is expectation. 

Route review themes back to the source. If customers keep confusing models, fix the variation structure. If customers misunderstand compatibility, tighten attributes. If customers misread product claims, update bullet copy. 

When review management becomes a loop instead of a fire drill, you stop treating bad reviews as a cost of doing business. 

3) Align marketplace data before you scale ads 

Marketplace data integration ranks as the second-biggest challenge at 33%, and it’s easy to underestimate how costly those integration speed bumps become once you push harder on growth levers. 

Incomplete integration creates quiet inconsistencies: price mismatches, delayed updates, missing attributes, inventory errors that propagate across systems. 

Then advertising scales and performance doesn’t follow. You spend more to acquire customers you can’t keep, and you blame the campaign when the problem lived upstream. 

Tighten the data path before you scale spend. Your future self will thank you in fewer support tickets and cleaner margin. 

Where AI fits without hijacking your strategy 

AI can help marketplace teams move faster. It can spot inconsistencies, flag missing fields, draft content variants, and speed up service workflows. 

It cannot rescue messy foundations. Give it incomplete data and it produces confident answers built on gaps. That’s why product truth has become a performance lever: AI increases the speed of both accuracy and error. 

What mastering the marketplace looks like in 2026 

Most teams feel good about listings. The report shows the friction elsewhere: inventory coordination, marketplace data integration, and review management. That’s why next year’s spend tilts toward customer service and review systems ahead of content and advertising. 

Marketplace growth requires consistency. Keep one version of the product true across platforms: what it is, whether it’s available, what it costs after fees, and what the customer should expect. The report shows why this is hard at scale. Inventory coordination ranks as the top challenge at 40%, followed by marketplace data integration at 33%. When those basics slip, the cost surfaces as returns, heavier support volume, and reviews that live on the listing. 

Micah McGuire is Senior Product Marketer, Brands, at Rithum. 

The post Mastering the marketplace: Build credibility with accurate inventory, data, and reviews appeared first on Rithum.

During peak season, more than $1.5B in Cyber 5 sales moved through Rithum across channels, regions, and categories. That’s the first wave. The second arrives after the shipping notifications stop: returns season. Calendars reset, staffing normalizes, and inventory shifts back toward warehouses. For the shopper, the decision often happened earlier—at checkout, when they relied on whatever the product page made clear and whatever it didn’t. 

And as shopping shifts from product pages to AI agents that summarize and recommend, returns get harder to manage. When an agent gives the highlights, shoppers may never scan the full product detail page (PDP), compare specs side by side, or catch details that prevent a mismatch. If the summary is wrong, the purchase can still go through, and the retailer still absorbs the return. The harder part is accountability and diagnosis. If shoppers never see the original product page, you lose the trail of what they were told and you can’t easily trace a return back to a missing detail, a bad summary, or a mismatched listing. 

How can you prepare for returns season? Use January to audit the promises you made during peak: which products were misunderstood, which channels dropped key details, and which fixes would have prevented repeat returns. 

What returns show 

In the U.S., online returns will top $363B, driven by a 24.5% return rate, according to Rithum’s 2025 global returns & profit impact report.  

Rithum’s consumer research shows the same drivers repeating: 61% cite poor fit and over a third say the item didn’t match the description or photos. 

Fit plays out differently by category. The “didn’t match” driver applies across categories. 

Electronics returns can trace back to compatibility or what was included. Home returns follow scale and finish. Parts and industrial returns surface around fitment, specifications, or installation assumptions. Across categories, the same thing happens: unclear listings leave customers to guess. 

In January, returns cluster. The same products come back through the same channels for the same missing details. In a Rithum returns fireside chat featuring Fabian Ortmann, Head of Returns, ZEOS; Kevin Brown, Director, Sales and Strategic Partnerships, Essendant Fulfillment Services; and Louis Camassa, Director of Product, Rithum, Ortmann argued that returns should be treated as a data source, not just a cost center, one of the clearest ways to pinpoint where expectations broke by product, channel, and market. 

That framing changes how brands and retailers should use January. For returns season, the work should shift from cleanup to finding out what went wrong. 

Where channel listings diverge 

Channel differences start costing money in January. During peak, products are listed, reformatted, and rewritten to fit each marketplace’s format. That’s expected. The risk is what gets lost. 

One channel may carry a compatibility field. Another may not. One listing spells out what’s included, while another assumes it’s obvious. One channel enforces short titles that drop (important) qualifiers where another keeps them in tact.  

Louis described this in the fireside chat: when listings leave room for interpretation, customers answer the question themselves. Sometimes they guess right. Sometimes they don’t and that can lead to a higher rate of returns. 

Why timing changes the math 

Returns also don’t happen on your schedule. In the same conversation, Essendant’s Kevin Brown described the gap between decision and action. Customers often decide quickly whether they’re keeping something. The return itself may take weeks to arrive. 

That delay narrows options. Inventory comes back later. Resale windows shrink. Seasonal goods lose flexibility. The cost isn’t only the refund. It’s what the business can no longer do with the product. 

January is also when slow fixes compound. If product information is inconsistent across channels and updates move slowly, the same mismatch keeps shipping while teams debate which version of the truth is “correct.” 

Rithum’s 2026 commerce readiness index points to why this is common. 91% of retailers and 78% of brands report poor data quality. Nearly 75% say inaccurate data leads to bad decisions. 

When product records aren’t consistent or trustworthy, January becomes more triage than correction. 

What policy can and can’t do 

Return policy shapes demand, but it doesn’t explain repeat returns. Policy still matters because shoppers read it before they buy, especially when they’re unsure. 

Rithum’s research shows return policies influence 41% of purchase decisions, 88% expect free returns, and 47% won’t click “buy” without them. 

Most preventable returns aren’t caused by policy. They’re caused by uncertainty that existed before policy mattered. 

What to check first 

Start with the returns that teach you something. Look at SKUs that return quickly after delivery. Fast returns tend to signal expectation breaks, not end-of-season cleanouts. 

Compare the same SKU across your top channels. Don’t assume it matches your internal record. Check titles, key attributes, images, what’s-included language, and variants. 

Track “didn’t match description/photos” returns by channel. Rithum found that 33% of shoppers cite mismatch with the description or photos as a reason for returns. 

Then move from insight to change quickly. If updates take weeks to reach every listing, the same mismatch keeps shipping. 

Where AI fits 

AI won’t fix messy product data. The eTail report’s preparedness finding reinforces the same issue returns already expose: only 2% of organizations describe themselves as fully prepared with structured product feeds and governance. 

Rithum’s 2026 commerce readiness index shows how widespread the gap still is: 91% of retailers and 78% of brands report poor data quality, and nearly 75% say inaccurate data leads to bad decisions.  

Whether the buyer is human, a marketplace algorithm, or an emerging shopping interface, the requirement doesn’t change. Product truth has to be consistent, category-appropriate, and fast to update. AI will move decisions faster, but it will also scale whatever your data gets wrong. 

What to do next 

The goal is to reduce the avoidable ones and recover value faster on the rest. 

The brands and retailers who do this well don’t wait for the next peak. They use January to tighten product truth, align it across channels, and push corrections while the signal is still fresh. Learn how Rithum can help.

Jordan Christensen is Director, Client Experience, Retailers at Rithum. 

The post Returns season is here. And AI raises the stakes  appeared first on Rithum.