AI readiness assessment: 25 questions on data, skills and infrastructure

More than 35 per cent of fashion executives already use generative AI in daily operations, and 92 per cent of fashion organisations say they will increase generative AI investment this year, yet only 1 per cent describe their rollout as mature. According to the State of Fashion 2026 report, up to 90 per cent of AI initiatives in fashion still fail to scale beyond the pilot phase, often because of weak underlying data and tech foundations.

MIT's 2025 study on the GenAI Divide found that just 5 per cent of integrated AI pilots are extracting millions in value. Gartner had already predicted that at least 30 per cent of generative AI projects would be abandoned after proof of concept by the end of 2025. The gap between ambition and execution is widening.

This diagnostic assessment provides 25 questions organised across four readiness dimensions. The questions are not a maturity model but a pre-investment checklist designed to expose gaps in data, skills, infrastructure and change capability that will determine whether AI initiatives scale or stall.

Why do most fashion AI pilots fail to scale?

McKinsey projects that generative AI could boost operating profits in fashion, apparel, and luxury by USD 275 billion by 2028. This year, 92 per cent of fashion executives intend to increase their AI investment, yet only 1 per cent report having mature deployments. On top of that, 90 per cent of pilots fail to scale.

KPMG findings showed that retailers are worried about AI skills, but industry panels challenged this thinking by arguing that poor data quality is the bigger issue. AI amplifies the quality of your data. If your product, inventory and customer data are inconsistent, AI will produce inconsistent outputs. Before investing heavily in AI tools, retailers should focus on data governance, product attribution and system consistency.

Challenges include high implementation costs, fragmented legacy systems, limited in-house expertise and uncertainty around governance and ethics. The technology is rarely the binding constraint. Organisational readiness is.

What data questions must you answer before starting?

Data readiness determines AI performance more than model selection. The following seven questions diagnose whether your data infrastructure can support AI at scale.

1. Can you access transactional data in real time? If product, inventory and sales data live in separate systems with batch updates, AI tools will work with stale information.
2. Are product attributes standardised and complete? Missing colour codes, inconsistent size labels and incomplete fabric descriptions will degrade recommendation engines and search.
3. Do you have a single source of truth for inventory? If stock data differ between your warehouse management system, e-commerce platform and point of sale, AI cannot allocate accurately.
4. Can you trace the lineage of your data? Knowing where each data point originated, when it was last updated and which transformations were applied is essential for debugging AI errors.
5. Is customer data unified across touchpoints? If a shopper's browsing history, purchase record and loyalty status are fragmented, personalisation engines cannot function.
6. Do you have documented data quality metrics? Without baseline measurements for completeness, accuracy and consistency, you cannot assess whether AI is improving or degrading data quality.
7. Can data be exported in formats AI tools require? If your systems cannot produce CSV, JSON or API feeds on demand, integration timelines extend significantly.

The use cases are fairly easy to get to something usable, but very complex to get to production level accuracy. Most of the customers coming to AI vendors have built something, but it is not working at a good enough level to roll out.

How do you assess technical skills and AI literacy?

The AI skills gap in retail is the number one barrier to AI maturity. 58 per cent of retailers name it, ahead of cultural resistance at 57 per cent and legal, compliance, and data privacy concerns at 54 per cent. A report from Voyado and Retail Economics, based on a survey of more than 300 retail leaders across the UK, DACH, and the Nordics, found that 58 per cent of retailers cite an internal skills gap as their primary barrier.

Six questions expose whether your organisation has the technical capability to deploy and maintain AI systems.

1. Does your IT team have API integration experience? Most AI tools require real-time data exchange via APIs. If your team has only managed batch ETL processes, the learning curve is steep.
2. Can staff distinguish between AI-generated and human-verified outputs? AI literacy at the business level is as important as technical expertise. Merchandisers and buyers must know when to trust and when to override AI recommendations.
3. Do you have personnel who can write and maintain prompts? Generative AI tools require prompt engineering. Without this capability, output quality degrades rapidly.
4. Is there internal capacity to evaluate vendor claims? Fashion AI vendors often cite accuracy figures from controlled benchmarks. Your team must be able to replicate tests on your own data.
5. Can your team read and interpret model performance metrics? Precision, recall, F1 scores and confidence intervals are standard AI metrics. If decision makers cannot interpret them, oversight becomes impossible.
6. Do you have a process for continuous model monitoring? AI models degrade over time as product catalogues and customer behaviour shift. You need staff assigned to track performance and trigger retraining.

Successful AI adoption starts with leadership understanding the business problem first, and the technology second. The panel recognised leadership understanding is a significant barrier for AI adoption in fashion businesses. IT leadership should be continually exposed to AI resources and tools to be able to develop and implement strong policies first, before empowering their business with AI.

Which platform and infrastructure questions expose gaps?

AI systems require technical infrastructure beyond standard retail IT. The following seven questions assess platform readiness.

1. Are your core systems cloud-native or on-premise? Cloud platforms like Azure, AWS and Google Cloud are the standard deployment environment for AI. If you run legacy on-premise systems, migration costs and timelines must be factored in.
2. Can your infrastructure handle increased API call volumes? AI-powered personalisation can generate thousands of API requests per minute. Load testing is essential.
3. Do you have a development and staging environment separate from production? Testing AI tools in production risks data corruption and system downtime. Proper environments are non-negotiable.
4. Is your network configured to allow third-party AI services? Firewalls and security policies often block the outbound connections AI tools require. IT security must review and approve access.
5. Can your platform support A/B testing at scale? AI improvements must be validated through controlled experiments. Your e-commerce platform must support variant testing and statistical analysis.
6. Do you have automated backup and rollback procedures? When an AI system fails, you need the ability to revert to the previous state immediately.
7. Is your infrastructure compliant with data residency requirements? If you operate in the EU, GDPR requires that certain data remain within specific geographic regions. Cloud configurations must reflect this.

Levi Strauss has migrated core IT workloads to Microsoft Azure, adopting Microsoft Fabric, OneLake, and over 1,000 AI agents to modernise reporting and analytics. The transformation underscores the importance of data readiness and agentic AI in driving real-time business insights. By consolidating its digital infrastructure on a single cloud platform, Levi's eliminated data silos that had long hampered real-time decision-making.

What change management questions reveal organisational readiness?

Skills and culture are one barrier, not two. Capability without permission to act stalls. Workforce skills without culture leaves capable people stuck in an organisation that won't let them experiment. Culture without skills creates enthusiasm with nowhere to go. The 58 per cent skills gap and the 57 per cent cultural resistance in the research aren't competing explanations.

Five questions diagnose whether your organisation can absorb the operational changes AI will impose.

1. Have you identified a named executive sponsor for AI initiatives? AI projects without C-level accountability drift. The sponsor must have budget authority and decision rights.
2. Are affected teams involved in vendor selection and pilot design? Merchandisers, buyers and planners who will use AI tools daily must have input. Top-down mandates without user buy-in fail.
3. Do you have a process for documenting and communicating AI decisions? When an AI recommendation is overridden, the reason must be logged. This feedback loop is essential for model improvement.
4. Can you allocate protected time for AI training and onboarding? If staff are expected to learn new tools on top of existing responsibilities, adoption stalls.
5. Is there a defined escalation path when AI produces unexpected results? Users need to know whom to contact when an AI tool recommends a stock level that contradicts their judgement. Without clear escalation, trust erodes quickly.

The most effective transformations balance quick wins with long-term ambition, targeting high-value tasks with relatively easier implementation in areas like marketing or back-office functions, while reinvesting gains to advance core product and customer-facing processes.

How did LVMH build AI readiness across 75 brands?

Luxury powerhouse LVMH has embraced AI, partnering with Google Cloud to deploy predictive, generative, and agentic tools across its 75 maison brands. AI supports everything from supply chains and pricing to personalised in-store service and creative inspiration, with LVMH's MaIA agent handling over two million employee requests monthly.

While many brands are piloting AI, LVMH has spent the last four years laying the pipes. In partnership with Google Cloud, it has built a single data platform that connects all 75 of its maisons. Over the past four years, LVMH has worked with Google Cloud to develop a central data platform that supports predictive AI, generative AI, and AI agents while allowing each maison to maintain its independence.

MaIA is LVMH's internal AI assistant, powered by Google's Gemini, Imagen, and GPT models. It currently handles over two million requests a month across 40,000 employees. MaIA is used for ideation, data requests, pricing models, creative prompts and more. Built using Google Gemini, Imagen, and OpenAI's GPT models, MaIA is now handling over two million internal queries monthly across 40,000 employees.

The LVMH approach demonstrates three readiness principles. First, infrastructure before tools. The group invested in a unified data platform across all brands before deploying AI applications. Second, internal adoption before customer-facing features. MaIA proved AI value to employees before the group scaled consumer applications. Third, governance that permits brand autonomy within a shared technical foundation.

What are the risks and limits of readiness frameworks?

Readiness assessments carry three structural risks. First, they can create a false sense of preparedness. Answering questions affirmatively does not guarantee execution capability. A company may have clean data but lack the organisational discipline to act on AI insights.

Second, readiness frameworks can delay action. Leading companies are using AI to automate manual tasks such as reporting, documentation, and contract audits. They reinvest those savings into AI enablers, higher-quality data and scalable AI infrastructure. The goal is not to spend more but smarter. Waiting for perfect readiness means ceding ground to competitors who are learning by doing.

Third, checklists privilege known variables over emergent ones. Retailers are woefully unprepared for the shift to autonomous agentic AI, especially at the retail store level. However, there is a lack of clarity and lack of mature standards concerning how agentic AI will interface with such systems, as many retail data sets are not yet modernised. New AI capabilities appear faster than governance frameworks can incorporate them.

Big Tech's competition for AI talent is driving up costs and limiting availability for fashion players. Smaller brands face a compounding disadvantage. They have fewer resources to invest in data infrastructure, less bargaining power with technology vendors and thinner margins to absorb experimentation costs. Building robust AI infrastructure can also be out of reach for smaller labels, potentially widening the gap between industry giants and independent creators.

A further constraint is regulatory uncertainty. Personalisation features and virtual try-ons may involve consumer data, including biometric identifiers. Fashion companies must implement data-minimisation practices, appropriate security safeguards, clear privacy notices, and consent mechanisms consistent with applicable laws. Vendor selection and contracting should address training-data provenance, permissible commercial uses, security requirements, and indemnities for intellectual property claims. Compliance requirements shift faster than IT roadmaps.

What should fashion companies do now?

The following seven steps provide a practical implementation sequence, prioritising actions that expose gaps early and build foundational capability before scaling investment.

1. Audit data infrastructure first. Before evaluating AI vendors, commission an internal audit of your data quality, completeness and accessibility. Identify the single largest data gap and allocate budget to fix it. Don't start with AI adoption. Start with clean, structured and accessible data.
2. Appoint a cross-functional AI lead. This role must have authority over IT, merchandising and operations. The lead's responsibility is to define use cases, manage vendor relationships and track ROI. Successful AI adoption starts with leadership understanding the business problem first, and the technology second.
3. Run a 90-day diagnostic pilot. Select one high-value, low-complexity use case such as automated product tagging or email subject line generation. The goal is not to prove AI works but to expose integration friction, data gaps and skill deficits. Document every obstacle encountered.
4. Establish governance before scaling. A significant portion of industry discussion focused on governance and privacy, stressing that governance should extend beyond employees to technology partners and vendors. Recommendations included creating company-wide AI policies, defining acceptable AI use cases, restricting what employees can upload, reviewing vendor contracts, and ensuring third-party technology providers align with AI policies.
5. Invest in foundational training. IDC projects that by 2028, half of major retailers will deploy advanced tools to close the digital and AI skills gap. Workforce transformation will become a defining factor in retail. The retail workforce will require autonomy and trust-building as a key factor in discovering and delivering advantages through AI. Allocate protected time for training, not optional lunch-and-learn sessions.
6. Build vendor accountability into contracts. Specify data ownership, model retraining frequency, performance guarantees and exit terms. Vendor selection and contracting should address training-data provenance, permissible commercial uses, security requirements, and indemnities for intellectual property claims.
7. Measure pilot success on operational, not demo, metrics. Fit recommendation received substantial funding but delivered the least value. Initial fit-tech promised a 30 to 40 per cent reduction in returns. Most deployments achieved 2 to 8 per cent. Insist on results measured in your own environment, on your own data, over a sustained period.

How do you know if you are ready to invest?

Readiness is not binary. No organisation will answer all 25 questions affirmatively. The assessment's value lies in prioritisation. If you cannot access real-time inventory data, no amount of technical skill or change management will make demand forecasting AI effective. If your teams lack prompt engineering capability, generative content tools will underperform regardless of data quality.

Three conditions indicate readiness to move from assessment to pilot. First, you have identified and fixed your single largest data quality issue. Second, you have secured a named executive sponsor with budget authority. Third, you have defined one specific business problem, measured its current cost and set a threshold for acceptable AI improvement.

Levi's launched an Outfitting AI tool in November 2025 that delivers personalised style recommendations through its app, and announced a super-agent platform built with Microsoft that integrates agentic AI into enterprise workflows via Azure and Teams. When a 173-year-old denim company deploys agentic systems to manage workflows, recommend outfits, and handle customer conversations and then publicly credits those systems for double-digit e-commerce growth, it validates the commercial case for deployment at scale.

The companies succeeding with AI in fashion share a common pattern. They prioritised infrastructure over features, internal adoption over customer-facing applications and operational discipline over technological novelty. Franck Le Moal, LVMH's Group IT director, stated that tech is mandatory to become super-efficient and at the same time keep the spirit and essence of luxury. This is the paradox that AI helps resolve.

The assessment questions in this article are diagnostic, not prescriptive. They expose gaps but do not specify solutions. The next step is to rank identified gaps by impact and addressability, then allocate resource accordingly. With turbulent conditions including volatile input costs, supply chain disruptions and slow growth straining fashion's economic model, AI is shifting from a competitive edge to a business necessity. Readiness is no longer optional.