Artificial intelligence has generated plenty of attention in fashion, but the more question is: Where is it actually delivering value?
Increasingly, the answer stretches across the supply chain.
AI is being deployed to select fabrics, predict product performance, calculate costs, automate compliance and identify inefficiencies on sewing lines. At retail, it is helping companies make assortment, inventory and promotional decisions. Yet the technology itself may be the easier part. Making it work requires clean data, clearly defined business problems and employees willing to change established ways of working.
That was a recurring theme during the AI-led Transformation Across the Fashion Value Chain panel at Bharat Tex 2026 in July. The discussion brought together technology providers and industry executives to examine AI applications throughout the apparel value chain.
Rather than treating AI as a single solution, the speakers described a collection of increasingly specialized tools aimed at specific pain points. The common thread was using data to make decisions earlier, faster and with greater precision.
For executives deciding where to invest, it’s important to understand that not every problem needs AI, and not every AI project is worth pursuing.
“Focus on the value, not on hype,” said Seema Agarwal, VP and Client Partner – Retail for AI firm Fractal. “Figure out what is the use cases which can make the biggest impact.”
Taking Guesswork Out of Product Development
Some of the earliest opportunities are emerging before a garment reaches production.
Fabric represents a significant portion of garment cost, but developing it traditionally requires communication between brands and mills followed by multiple physical iterations.
Aseem Rajendra Chitkara, co-founder of B2B sourcing platform STCH, described how his company is building a system that instead draws on a database of fabric inputs and outputs. A user can specify desired characteristics, such as sheen or twill definition, and the technology identifies the construction and other inputs that could produce the requested result.
The goal is to reduce physical development rounds and give brands and manufacturers a quicker route from concept to a workable fabric.
The company is applying a similar approach to design. Its system analyzes product attributes like fabric, wash, fit and trims that are performing in the market and matches those signals against a brand’s individual positioning.
The business case isn’t simply producing designs faster. Chitkara said the objective is to improve sell-through and reduce inventory by giving companies better information before they commit to a product.
The technology has already moved beyond testing in at least one instance. Chitkara described using the system after brands in the U.K. asked whether garments they were sourcing from Turkey could instead be produced in India.
“We successfully did it,” he said. “I will not say it was complete, but it was fairly successful.”
The resulting program has already produced 50,000 garments, according to Chitkara. Stitch is now expanding the datasets behind its tools and testing its design system with additional brands.
Getting to Cost Earlier
Another opportunity that sits at the intersection of product development and sourcing is determining whether a design can hit its target cost before suppliers spend time costing it.
Shruti Grover, CEO and Co-Founder of AI planning company MannyAI, described technology that reads a tech pack and automatically generates a bill of materials, including trims, order quantities and size ratios.
For one Indian activewear company, the technology is being used to generate roughly 1,500 purchase orders per month covering approximately 4 million garments. Grover estimated that it eliminates between 2,000 and 3,000 emails each month.
The company is also piloting a “flash costing” tool designed to tackle another longstanding source of inefficiency: products that reach suppliers for costing only to turn out to be commercially unworkable.
Grover estimated that 15 percent to 30 percent of designs sent to manufacturers for costing fail to meet their targets. By that stage, however, both time and resources have already been spent evaluating the product.
With the AI tool, designers can upload a tech pack or sketch and receive an estimated bill of materials, labor requirements and unit cost. They can then model changes—removing a feature, changing a material or adjusting another specification—to understand the effect on cost before moving ahead.
That creates an opportunity to bring commercial considerations closer to design rather than discovering problems after a product has entered the sourcing process.
From Hours to Seconds
At H&M, AI is already being applied to decisions that traditionally require specialized technical expertise.
Rakesh Ranjan, Business Area Manager of H&M, described the retailer’s work automating standard minute value, or SMV, recommendations. SMV measures how long a qualified operator should take to produce a garment and feeds into decisions involving costing, capacity and production planning.
Traditionally, an industrial engineer might spend one to three hours calculating the figure. That becomes difficult to scale for a global retailer handling large numbers of products.
H&M’s system extracts a product sketch from a tech pack, uses image recognition to find similar historical styles and adjusts the result for variables including target group and fabric.
“Initially it used to take over one to three hours, now in seconds,” Ranjan said.
H&M is also using automation for quality compliance. Testing across a global supplier base generates large quantities of test results and certificates that must be checked against the requirements of individual products and selling markets.
The retailer developed a dynamic rules engine that identifies the necessary tests. AI agents then review results and flag potential compliance or authentication problems to H&M and its suppliers in real time.
The result is a process capable of handling large product volumes without requiring the same level of manual review.
But Ranjan also cautioned against assuming every business problem requires artificial intelligence.
“If processes are set and rules are known, then a simple algo can solve problems,” he said. AI becomes more useful where companies need prediction or must analyze large quantities of complex or unstructured information.
In this way companies can sort worthwhile investments from projects that simply carry an AI label.
Unlocking the Factory’s Data
AI’s possibilities look different on the factory floor, where one of the biggest obstacles is not analyzing data but collecting it in the first place.
Bodo Voet, Founder and CEO of sewing intelligence firm OpenSeam, approached apparel manufacturing after working on AI and industrial Internet of Things applications at Microsoft. He quickly encountered a familiar problem: many industrial environments don’t generate standardized, usable data.
“The tech was not the difficult part,” he said. “The part was the operational implementation within industrial environments that were not ready.”
Garment manufacturing presents the problem on a massive scale. Vahldiek estimated that roughly 50 million sewing machines are operating globally, with most providing little or no digital production data.
OpenSeam developed sensors that can be attached to existing sewing machines to capture information about what is happening on the production floor.
For factories, this reveals information about lost production time.
The technology can help identify constraints, changes in throughput over a shift and potential relationships between performance, skills, training and line planning. In one Thailand pilot, the factory used the information to reconsider how workers were grouped and how production targets were allocated.
The larger opportunity is to make previously invisible factory activity available to manufacturing execution systems, ERPs and other platforms.
That data gap also highlights a competitive imbalance between brands and manufacturers.
Retailers have spent years collecting digital information about customers, products and transactions. Factory operations remain much less visible.
Vahldiek argued that manufacturers shouldn’t necessarily give away the information they eventually capture.
“It should remain a competitive advantage of a factory to have the data to actually do better,” he said.
What should be shared, he argued, are standards governing how that information is collected and structured. Common standards could allow manufacturers and technology providers to build interoperable systems without requiring factories to surrender proprietary information.
Connecting AI Across the Supply Chain
For retailers, meanwhile, the technology is increasingly being applied to predicting what customers will buy.
Agarwal described combining information from websites, apps, email, CRM systems, loyalty programs and other channels to create a unified customer record. AI models can then determine potential next actions and refine their predictions as consumers continue interacting with a company.
On the merchandising side, AI can analyze attributes including color, sleeve length, neckline, pattern and price to estimate demand for new products that don’t yet have their own sales histories.
Once a season begins, models can combine current performance with historical and market information to identify underperforming products and recommend actions such as transfers, promotions or markdowns.
The potential next step is moving from recommendation to execution. Specialized agents could update forecasts, launch marketing campaigns or allocate stock, with humans remaining involved in the decision process.
That could create a much tighter connection between what the consumer is doing and decisions made throughout the supply chain.
Chitkara argued that brands and manufacturers ultimately need to think about those opportunities together. Retailers may be focused on inventory and sell-through while factories concentrate on cost, productivity and production. But optimizing one piece of the system doesn’t necessarily optimize the whole.
“There is an inefficiency in the complete flow, inefficiency in the complete system,” he said.
Data Before AI
The panel repeatedly returned to one prerequisite: data.
Without sufficient usable information, even sophisticated AI models have little to work with.
Ranjan compared data to the raw material entering a factory, AI to the factory itself and the business result to the finished product.
“Unless you have good data, you can’t have AI,” he said. “Even if you have good data but you don’t have a business case, the outcome is not there.”
H&M has invested heavily in that infrastructure. Ranjan said the company has more than 2,000 people in its technology organization and more than 300 people across data science, engineering, machine learning and related data roles. But he stressed that those resources become useful only after the underlying ecosystem has been digitized.
Manufacturers don’t need H&M-sized teams to begin.
Ranjan advised suppliers to identify use cases tied to strategic outcomes, invest in their data and business processes and recognize that AI requires continuous iteration rather than a one-time implementation.
The final piece is change management.
Apparel remains an industry where many processes have long relied on spreadsheets, email, manual inspection and individual expertise. Installing a new system doesn’t automatically change those habits.
For apparel executives, that may be the most useful reality check as AI investment accelerates.
The competitive advantage will go to the businesses that identify where technology can solve a real problem, build the data infrastructure to support it and integrate the resulting intelligence into everyday decisions.
Or, as Agarwal summarized the path forward: “Think big, start small and scale fast.”
