When the buyer is a bot: how AI agents are entering fashion wholesale

In October 2025, at its IT Symposium in Orlando, Gartner made a forecast that should unsettle every wholesale director in fashion: by 2028, 90% of B2B buying will be "AI agent intermediated", pushing more than USD 15 trillion of business spend through what the research firm calls AI agent exchanges. If it holds, the counterpart reading a brand's line sheet in two years' time will often be software acting for a buyer, not the buyer in person.

The measured reality in fashion is far more modest. NuORDER's 2026 State of B2B eCommerce report finds that AI adoption in buying and planning rose from 11% to 17% over the past year. Gartner's own survey of 645 B2B buyers, fielded in August and September 2025, found that 45% used generative AI during a recent purchase, mainly to research vendors and products, while 69% still prefer to validate AI-generated insights with a sales rep.

So is the buyer already a bot? Not yet, and not in the way the headline suggests. AI is entering fashion purchasing through narrow, repeatable tasks: drafting replenishment orders, matching catalogues to trends, screening new suppliers, assembling trading reports. Fully autonomous agents placing seasonal orders remain rare. But the infrastructure for agent buying is being laid now, by Google, Walmart, Faire and the wholesale platforms, and the brands that will be found and reordered by machines are those whose product, price and availability data a machine can actually read.

What does Gartner actually predict about AI agents in B2B buying?

The 90% figure comes from Gartner's strategic predictions, presented by Daryl Plummer, its chief of AI research, at the Orlando event in October 2025. The wording matters. "Intermediated" means an agent sits somewhere between buyer and seller: searching, comparing, summarising, negotiating or executing. It does not mean nine in ten purchases will be decided without a human. It is a forecast about direction, presented to an audience of chief information officers, rather than a measured share of today's purchasing, and Gartner attached no fashion-specific estimate to it. Even so, it tells suppliers where large buyers and platforms intend to go, and how quickly they think the shift can happen.

Other research houses describe a buying process that is already changing, but more slowly. Forrester's State of Business Buying 2026, released in January, finds a typical purchase decision involves 13 internal stakeholders and nine external influencers, and that buying groups roughly double when a purchase includes generative AI features. In its 2025 survey, 36% of buyers said generative AI made them more confident they had made a well informed decision, while 20% felt less confident because they had met unreliable or inaccurate information.

McKinsey's 2026 B2B Pulse survey of nearly 4,000 decision-makers in 13 countries, published in May, adds that generative AI has risen into the top five channels buyers use for supplier discovery and evaluation. Only 22% of organisations say they have fully implemented generative AI, up from 19% a year earlier, with a further 31% actively adopting it. Buyers are comfortable spending more online: comfort with online orders above USD 50,000 rose from 59% in 2022 to 73%. Yet willingness to spend USD 500,000 or more online fell five percentage points from 2024, a reminder that large commitments still attract human scrutiny.

How are B2B buyers using AI in purchasing today?

Gartner's May 2026 buyer survey gives the clearest picture of current behaviour. Buyers drew on an average of seven information sources during a recent purchase. Of the 645 respondents, 45% used generative AI, 67% said they prefer a buying experience without a sales rep and 70% prefer a completely digital, self-service process. Trust remains split: 51% said they are more likely to encounter misleading information from generative AI, and 49% said the same of a sales rep.

Robert Blaisdell, who leads research in Gartner's sales practice, summarised the tension: buyers "are more comfortable using digital channels and GenAI to navigate the purchase process on their own", but "that does not eliminate the role of the seller." For fashion, where a wholesale account manager often doubles as a merchandising adviser, that is the likely near-term pattern: machines prepare and compare, people confirm and negotiate.

What are fashion retailers' buying teams already automating?

The most automatable part of fashion buying is the part that has grown fastest. JOOR's wholesale trends paper for 2026 shows the average time from order placement to shipping fell from 263 days in 2019 to 102 days in 2024. Evergreen styles rose from 37% of JOOR's gross merchandise value in 2019 to 49% in 2024, and buyers are shifting budget towards in-season purchasing. Shorter cycles and more continuity product mean more small, frequent, rules-based decisions: exactly the work software handles well.

Guess offers a documented example on the retail side. According to a case study published in June 2025 by its software provider Nextail, Guess keeps 50 to 60% of its stock for in-season needs and moved to fully automated replenishment based on hyper-local demand forecasts, going live across EMEA in six months. The reported results: full-price sell-through up 5 percentage points, store coverage down 7.5% and 13% fewer stockouts. These are vendor-published figures, but they show where automation pays first: deciding which units go where, every day, without a planner touching each line.

AllSaints is taking the same route further upstream. In April 2026 the British retailer said it was adding AI-native planning tools across trading, forecasting, allocation and replenishment, markdown pricing, financial planning and range planning. Its weekly trading packs, which teams started compiling on Sunday and finished after Monday, are to be available by 8am on Monday. "We have a great team, but they have to spend so much time manually pulling and analysing data," Alfie Meekings, chief transformation and technology officer, told Retail Gazette.

Neither case is an agent placing orders with outside brands. They automate internal decisions. But once a retailer's replenishment logic runs on forecasts rather than spreadsheets, the step to sending purchase orders for never-out-of-stock lines automatically to suppliers is short, and the supplier whose availability and prices can be queried by machine becomes the easiest to reorder from.

What are wholesale platforms and large retailers building for agent buying?

The marketplaces that sit between brands and stores are adding AI to the buyer's side of the screen. In March 2026 JOOR said it had built a proprietary tool using a fashion-specific vision-language model that searches the millions of products on its platform for styles matching runway trends, which it used for a shoppable Fall 2026 women's trend report aimed at its community of more than 675,000 retail buyers. Human trend analysts chose the six trends first; the model then found commercially available styles. JOOR also reports that 78% of new retailer clients acquired through its Discover matchmaking tool reorder, placing more than five orders on average in their first year.

Faire, which expects about USD 3 billion in gross merchandise volume and was valued at USD 5.2 billion in December 2025, already uses deep learning to rank products and has said it plans AI buying assistants to help retailers curate assortments and automate product uploads. Its chief executive, Max Rhodes, noted that only about 5% of the hundreds of billions of dollars US retailers spend sourcing inventory happens online. When Faire opened its platform to business-use buyers such as hotels and offices in June 2026, it kept firm guardrails: brands choose whether to sell to the segment, buyers must meet brand minimums and brands approve every order.

Large retailers and technology groups are building the rails. In January 2026 Google launched the Universal Commerce Protocol, developed with Shopify, Target, Walmart and Etsy, an open standard that lets agents handle the journey from product discovery to payment, designed to work with its Agent Payments Protocol and with the Model Context Protocol. OpenAI and Stripe have their own Agentic Commerce Protocol. These standards are aimed at consumer shopping first, but they define the rails that B2B agents are likely to reuse, and Google paired the launch with new data attributes in Merchant Center.

How ready are fashion brands' wholesale teams for AI buyers?

Brands are investing in wholesale again. In NuORDER's 2026 report, 78% of senior leaders rank B2B wholesale as their top investment priority, up from 72% a year earlier, and JOOR's market survey finds that 52% of brands call wholesale their most profitable channel. But the data foundations an agent would need are thin. NuORDER finds that only 9% of brands have fully integrated ERP systems. Inventory forecasting is the leading AI use case, cited by 44%, and AI adoption in buying and planning remains a minority practice.

The gap between Gartner's 90% forecast and a 17% adoption rate is not a contradiction. It is the distance between where large procurement organisations are heading and where seasonal fashion wholesale is today. It also shows where the work lies: before a brand can sell to an agent, its own systems must agree on what is available, at what price and when.

The selling calendar is making that harder. JOOR's market survey finds that the share of brands producing capsule collections rose from 19% in 2023 to 37% in 2025, while the share producing a full new collection slipped from 86% to 83%. More drops mean more product records, images and price lists to publish, more often, and more chances for the version a retailer's system reads to differ from the one in the brand's ERP. For a human buyer, a mismatch prompts a phone call. For an automated reorder, it produces a wrong order or no order at all.

What data does an AI buying agent need from a fashion brand?

A human buyer can work around a blurry image, a PDF line sheet or a price list sent by email. An agent cannot, or will silently skip the product. The requirements are mundane but strict: structured, consistent and current data that answers a buyer's questions without a phone call.

Two further requirements are less technical. Brands need explicit rules on what an agent may do on a buyer's behalf, such as which order sizes or changes require human approval, as Faire's approve-every-order model shows. And they need to decide which data they are willing to expose: real-time availability helps an agent reorder, but also tells competitors and large accounts a great deal about a brand's position.

What are the risks and limits of AI agents in wholesale buying?

The first limit is accuracy. In Forrester's 2025 survey, procurement professionals were more sceptical than other buyers: 28% felt less confident in a decision because of inaccurate AI output, against 19% of all buyers, and 22% said they had wasted time because of poor AI information. An agent that misreads a size curve or a delivery window can create an order that is technically valid and commercially wrong.

The second is the nature of fashion buying. Pre-season orders combine taste, brand strategy, exclusivity and relationships, and they are often negotiated around marketing support, margins and allocation of scarce styles. Gartner's own data suggest buyers want reps to validate AI findings, and Blaisdell warned sales leaders not to "interpret buyer preference for digital self-service as a signal that sellers matter less." McKinsey's finding that willingness to place very large orders online has fallen points the same way.

The third is concentration. If buying agents are built and controlled by a few platforms and large retailers, they will rank suppliers on criteria those owners set, from data completeness to margin and advertising. Faire's advertising business, launched two years before its December 2025 funding round, already accounts for more than 5% of its revenue. Brands should expect visibility to agent buyers to become something that can be bought, as search visibility became before it.

Finally, forecasts are not facts. Gartner's 90% figure is a strategic prediction with a broad definition, not an observed share, and none of the sources reviewed for this article report a fashion retailer letting an agent place seasonal orders with outside brands without human approval. The safest reading is directional: agent intermediation will grow, starting with research and replenishment, long before it reaches the showroom.

What should fashion wholesale teams do now?

1. Audit your line sheet as data. Check whether every style, colour and size exists as structured fields with consistent names across ERP, PLM and wholesale channels.
2. Publish availability by SKU and window. Make available-to-sell and delivery dates queryable for partners, starting with continuity and replenishment lines.
3. Encode your commercial rules. Turn minimums, prepacks, price tiers and cancellation terms into explicit rules rather than email agreements.
4. Define what a machine may approve. Set thresholds for order size, changes and substitutions that require a human sign-off on your side.
5. Start with reorders, not pre-season. Pilot automated or agent-assisted ordering with a few retail partners on never-out-of-stock lines, and measure sell-through and errors.
6. Track how platforms rank you. Monitor how marketplaces and retailer portals display and recommend your products, and what data completeness they reward.
7. Retrain account managers as advisers. Prepare sales teams to validate, explain and negotiate around what buyers' AI tools surface, since that is where buyers still want people.