How should fashion companies budget for AI in 2027?
A 2027 AI budget works best as a portfolio: a small base for foundations and compliance, a core for proven use cases, and a capped experimental pot. Usage-based costs and data work are the lines most often missed.
KEY TAKEAWAYS Summary by the editors
- An AI budget for 2027 should be structured as three layers: foundations (data, security, compliance, training), proven use cases with owners and baselines, and a capped pot for experiments.
- McKinsey's 2026 survey found that 60% of respondents expect to increase AI investment over the next year, 28% spend more than 10% of their ICT budget on AI, and about 20% say operating costs including tokens have constrained use.
- Switching and data egress charges for cloud services end in the EU on 12 January 2027 under the Data Act, which changes the economics of exit and multi-supplier set-ups.
- Gartner commentary reported by CIO Dive warns that ROI is harder to calculate when many people use a tool and workflows are not clearly defined, so budget lines should be tied to specific workflows.
- Training is both a budget line and a legal consideration, since Article 4 of the EU AI Act on AI literacy has applied since 2 February 2025.
Fashion companies should budget for AI in 2027 as a portfolio with three layers: a foundation layer for data, security, compliance and training; a core layer for use cases that already have an owner, a baseline and a target; and a small, capped pot for experiments. The most common budgeting errors are to fund pilots without funding the data work behind them, and to ignore usage-based fees that grow with adoption. This guide does not give benchmark spend figures, because no reliable fashion-specific ones are published in the sources reviewed.
What is the market context for 2027 AI budgets?
McKinsey's 2026 survey of 1,719 respondents found that 60% expect to increase AI investment over the next year and that 28% spend more than 10% of their enterprise ICT budget on AI. At the same time, the share attributing at least some EBIT impact to AI stayed at 37%, and about 20% said operating costs, including tokens, have constrained their AI use. The report's own summary is that conviction is growing faster than the financial returns organisations can attribute to it.
Earlier Gartner commentary, reported by CIO Dive in 2024, anticipated that CFOs would first be asked to articulate an AI strategy and, a couple of earnings calls later, to say what the return was. That scrutiny has not gone away, and a budget that cannot show its basis will be cut first.
What belongs in each budget layer?
| Layer | Typical contents | Funding rule |
|---|---|---|
| Foundations | Data cleaning and product attribute standards, integration work, security review, governance, AI literacy training | Fund as infrastructure; justify by what it enables, not by direct savings |
| Core use cases | Forecasting, content production, customer service assistants, planning and reporting support | Fund only with a business owner, a measured baseline and review dates |
| Experiments | Small trials of new tools or ideas | Capped amount, time-boxed, with a stop rule and a lessons review |
A sensible habit is to release core-layer funds in tranches tied to milestones: first the data and pilot, then the wider rollout once the first value check has been passed.
Which costs are most often missed?
- Usage-based fees: pricing per token, request or seat can rise quickly as adoption grows, so set alerts and monthly reporting.
- Data preparation: cleaning, matching and enriching product, customer and supply data often costs more than the AI tool.
- Integration and maintenance: connecting to ERP, PLM, e-commerce and wholesale systems, and keeping connections working when the vendor changes.
- Change management and training: time away from the job counts as cost.
- Governance and legal review: contract reviews, risk assessment and documentation.
- Exit and switching costs: the price of moving away from a supplier if it proves necessary.
How do the 2027 EU rules affect budgeting?
Two European developments matter. First, under the Data Act, switching and data egress charges for data processing services end on 12 January 2027, according to the European Commission, with a transitional period before that in which providers may charge costs directly tied to the switch. Buyers can therefore assume lower exit costs for covered services after that date, although migration still takes staff time. Second, the EU AI Act continues to phase in: the Commission's high-level summary lists obligations for Annex III high-risk systems from 2 December 2027, so budgets for classification, documentation and oversight should be planned in 2027 if any systems may fall in scope. Many typical fashion use cases, such as forecasting or content generation, are unlikely to be high-risk, but systems used in employment decisions can be.
Article 4 on AI literacy has applied since 2 February 2025, so a training budget is not optional for companies using AI systems, though the Commission says no specific level or certificate is mandated.
How should you decide between build and buy?
McKinsey's survey found that 32% of respondents had decided against buying at least one software product or feature because they could build it in-house with agentic coding tools. Building lowers licence cost but adds development, security and maintenance costs and key-person risk. Compare total cost over three years, including the staff time to maintain the tool, the cost of security review and the effect if the builder leaves, and treat the comparison as an estimate rather than a certainty. A bought tool carries its own risks, notably dependence and price changes, so the comparison should list those too.
How should you handle uncertainty in the numbers?
Present each core use case with a low, expected and high case, and say what assumption drives the difference: adoption rate, hours saved that are actually redeployed, or usage volume. Usage-based pricing makes the high case a cost risk as well as a benefit, so a spending cap or alert per use case is a sensible control. Where a supplier cannot explain how its price will behave if usage doubles, treat that as a finding in itself.
A mid-year review should be written into the plan. If a use case has not reached its first value check by then, the default decision is to stop or reduce funding, with the released money returned to the experiment pot.
How do you present the budget to the board or CFO?
- State the three layers and the amount in each.
- For each core use case, give owner, baseline, target, review date and net cost including running costs.
- Show what will be stopped if the first value check is missed.
- Separate one-off from recurring costs, and show usage-based costs with a range.
- Report what was spent and achieved in 2026, including failures.
Frequently asked questions
How much should a fashion company spend on AI in 2027?
No reliable fashion-specific benchmark is published. McKinsey's 2026 survey found 28% of respondents spend more than 10% of their ICT budget on AI, but that is a global, cross-industry figure and not a target. Build the budget from specific owned use cases.
What are the hidden costs of AI projects?
Usage-based fees, data preparation, integration, training, governance and legal review, and the cost of switching suppliers. These often exceed the licence cost of the tool.
Do EU rules change AI costs in 2027?
Yes in two ways. Switching and data egress charges for covered data processing services end on 12 January 2027, and the AI Act's high-risk obligations for Annex III systems apply from 2 December 2027 according to the AI Act high-level summary.
Should AI budgets be fixed for the whole year?
A tranche-based approach is more prudent: release funds as milestones such as data readiness, pilot results and first value checks are met, and keep a capped pot for experiments.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
