How do you build an AI business case your CFO will sign? A template for fashion
Most AI projects stall between pilot and P&L. A practical business case template for fashion companies: baseline, value drivers, full costs, risks and stage gates, with examples from planning, content and service.
KEY TAKEAWAYS Summary by the editors
- A credible AI business case starts from a measured baseline of one process, not from a technology, and states the value driver in units a CFO already tracks.
- In McKinsey's State of AI 2025 survey, 88 percent of respondents reported regular AI use in at least one function, but only 39 percent attributed any EBIT impact to AI.
- McKinsey found that AI high performers, about 6 percent of respondents, were nearly three times more likely to have fundamentally redesigned workflows.
- MIT's Project NANDA reported in 2025 that only 5 percent of custom enterprise generative AI tools reached production.
- Total cost should include data preparation, integration with ERP and other systems, change management, model usage and ongoing monitoring, not only licences.
An AI business case a CFO will sign links one specific process to a measured baseline, a clear value driver expressed in existing financial metrics, a full cost estimate including data and integration work, named risks and stage gates that release budget only when results are proven. The technology is the smallest part of the document. What convinces finance is evidence that the team knows what will change, how it will be measured and when to stop.
Why do so many AI projects fail to show ROI?
Adoption has outpaced financial impact. In McKinsey's State of AI 2025 survey of 1,993 participants, conducted from June to July 2025, 88 percent reported regular AI use in at least one business function, but only 39 percent attributed any level of EBIT impact to AI, and most of those reported less than 5 percent of EBIT. MIT's Project NANDA reported in 2025 that only 5 percent of custom enterprise generative AI tools evaluated by organisations reached production.
Fashion is no exception. In The State of Fashion 2026, published in November 2025, McKinsey and The Business of Fashion reported that more than 35 percent of executives already use generative AI in areas such as online customer service, image creation, copywriting and product discovery. Use is common; documented returns are not.
Common reasons for weak cases include:
- The project starts with a tool, not with a process and its cost.
- Benefits are counted as hours saved without a plan for what happens to those hours.
- Data preparation and system integration are left out of the cost.
- There is no baseline, so improvements cannot be proven.
- Pilots have no exit criteria and drift for months.
What should an AI business case template contain?
| Section | Content | Example (product content generation) |
|---|---|---|
| 1. Process and owner | One process, one accountable business owner | Product descriptions for wholesale and e-commerce, owned by head of e-commerce |
| 2. Baseline | Current volume, cost, time, quality | Articles per season, minutes per description, error rate, time to listing |
| 3. Value driver | Financial metric the change affects | Earlier listing (sell-through), lower agency cost, fewer retailer rejections |
| 4. Target and evidence | Expected change and why it is plausible | Results from a controlled pilot on one category |
| 5. Full cost | One-off and running costs | Data cleanup, PIM and ERP integration, licences, usage, review time, training |
| 6. Risks | Quality, legal, data, adoption | Incorrect claims, brand tone, copyright, staff acceptance |
| 7. Stage gates | Criteria to continue, scale or stop | Scale only if error rate below baseline and time saving confirmed |
| 8. Measurement | How and when results are reported | Monthly dashboard against baseline, reviewed by finance |
How do you translate AI benefits into financial terms?
Finance teams trust benefits that map to lines they already manage. Typical fashion examples:
- Demand planning and replenishment: lower forecast error leads to less markdown and fewer stockouts; express as gross margin and inventory turns.
- Product content: faster time to listing and fewer data errors; express as earlier revenue and lower external content cost.
- Customer service: share of requests resolved without an agent; express as cost per contact, while tracking satisfaction.
- Wholesale order processing: faster order entry and fewer errors in ERP; express as cost per order and fewer credit notes.
- AI agents in back-office tasks: automated handling of routine steps such as invoice matching; express as cost per transaction and cycle time.
Hours saved only count when they are redeployed or remove cost. State explicitly whether the case assumes lower spend, more output with the same team or higher quality, and do not count the same hours twice.
What costs do fashion AI projects usually underestimate?
Licences and model usage are visible; the rest often is not. Data cleanup in product, customer or transaction records, integration with ERP, PIM or order management, review and approval time for outputs, security and legal checks, training and change management, and ongoing monitoring of model quality all belong in the total cost. For agent-based projects that write into ERP, add controls for permissions, audit logs and fallback procedures.
How should stage gates and pilots be set up?
- Discovery (2 to 4 weeks): measure the baseline and confirm data availability; stop if data is missing.
- Pilot (6 to 12 weeks): run on one category, market or team with a control group; agree success thresholds in advance.
- Decision: finance reviews results against the baseline; scale, adjust or stop.
- Scale: roll out with integration into core systems and named process changes.
- Run: report benefits quarterly and re-check model quality and costs.
What risks should the business case name?
Keep the first case small. A single, well-measured process with a modest budget creates an internal reference that later cases can cite. Finance teams are more willing to fund a second and third project when the first one reported honestly, including what did not work, than when several large programmes start at once without comparable baselines.
Name the risks a CFO will ask about before they ask: output quality and the cost of errors, especially for product claims and prices; dependency on a vendor or a model that may change; data protection and intellectual property; regulatory obligations such as transparency for AI-generated content; and adoption risk if teams do not trust or use the system. For each risk, state the control and who owns it.
A business case that admits uncertainty and limits exposure through stage gates is more likely to be approved than one that promises a large return on day one. The goal is not a perfect forecast, but a decision process that turns a pilot into measured value or ends it early.
Frequently asked questions
How do you calculate ROI for an AI project?
Measure the baseline cost and performance of one process, estimate the change AI will bring in financial terms, and compare that benefit with the full cost including data preparation, integration, licences, usage and change management. Confirm the estimate with a controlled pilot before scaling.
Why do most AI pilots fail to deliver ROI?
Common reasons are starting with a tool instead of a process, missing baselines, ignoring integration and data costs, and not redesigning workflows. McKinsey's 2025 survey found only 39 percent of organisations attributed any EBIT impact to AI.
What costs should an AI business case include?
One-off costs such as data cleanup, system integration, security review and training, and running costs such as licences, model usage, human review and monitoring. Agent projects that act in ERP also need permission controls and audit logs.
How long should an AI pilot run in fashion?
Long enough to measure against a baseline, often a few weeks to a few months on one category or market. The key is to agree success thresholds and a stop criterion before the pilot starts.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
