Assortment planning RFP: 25 questions for AI platforms

Retailers commissioning AI-enabled assortment planning systems face a paradox: the technology promises significant productivity gains, yet most implementations fail to deliver. In a recent survey by McKinsey & Company, 71 per cent of merchants said AI merchandising tools have had limited to no effect on their business. The gap between potential and performance typically stems from poor data foundations, fragmented systems and unclear integration requirements. A rigorous request for proposal should address these operational realities before evaluating algorithms or user interfaces.

What data infrastructure must the platform support?

AI merchandising systems depend entirely on data quality and accessibility. Most AI tools require reliable data to work effectively, such as customer, products, stock, transactions, and clickstream analytics. Other data valuable to the AI may include returns and competitor pricing. Yet nearly half of retail decision-makers are battling with data accessibility issues, and only two in five (42 per cent) have successfully bridged gaps between data silos. Before asking vendors about their algorithms, confirm they can specify minimum data requirements, handle missing or inconsistent data, and integrate with existing data warehouses or lakes.

• Which transaction, inventory, product attribute and hierarchy data are required versus optional?
• How does the system handle incomplete product histories, seasonal items or new SKUs with limited sales data?
• What data quality checks are built into ingestion pipelines?
• Can the platform operate on batch updates or does it require real-time streams?
• How are product hierarchies, store clusters and size curves defined and maintained?

Only 23 per cent of omnichannel retailers had a unified data layer capable of feeding real-time AI models in 2025. The remaining 77 per cent run AI on partial data, producing recommendations that contradict across channels and forecasts that miss cross-channel demand shifts. Specify whether the vendor expects you to deliver a unified customer view or if their system can reconcile fragmented sources.

How transparent and explainable are AI forecasts?

Merchandisers will not trust recommendations they cannot interrogate. The most common reason AI tools fail to gain traction is that buyers and planners cannot understand why the system made a particular suggestion. Request vendors demonstrate how users can see which factors drove a forecast, compare model outputs to historical performance and override recommendations when business context demands it.

• Can users see which variables (seasonality, promotions, weather, trend signals) influenced a specific forecast?
• How are forecast confidence intervals displayed and updated?
• What audit trail exists when users accept, modify or reject AI recommendations?
• How does the system learn from user overrides?
• Are models retrained automatically or on a defined schedule?

In merchandising, AI can automate data consolidation, forecasting, and promotional planning to unlock capacity gains of about 40 per cent. However, systems remain fragmented, data is too messy to use to deliver useful recommendations, and adoption is uneven: 61 per cent of respondents say that their organisation isn't at all or is only slightly prepared to scale AI across merchandising. Transparent models reduce resistance and accelerate adoption.

What integration depth is required with existing systems?

Leaders consistently pointed to the same challenges: reconciling data across multiple systems, managing exceptions, a lack of vision to interoperable process flows and relying heavily on spreadsheets to maintain operational flow. The platform must connect to ERP, POS, merchandise financial planning, allocation, pricing and PLM systems without creating new bottlenecks.

• Which ERP, POS and warehouse management systems have certified integrations?
• Does the platform offer pre-built connectors or require custom API development?
• How are product, location and hierarchy master data synchronized?
• Can assortment plans flow directly into purchase orders and allocation runs?
• What happens when upstream systems (PLM, financial planning) change hierarchies or timelines mid-season?

Data formatting inconsistencies and legacy system limitations are common hurdles during integration. Ask vendors to provide timelines and resource requirements for integration, including whether they expect retailer IT teams or third-party consultants to lead the work.

Does the workflow align with merchandising roles?

The best algorithms are useless if the platform does not match how buyers, planners and allocators actually work. Confirm that Open-to-Buy (OTB) and Weekly Sales, Stock and Intake (WSSI) capabilities are directly connected to assortment decisions, and ask how the platform manages carryover and new product planning within a unified assortment workflow. Separate tools for financial planning, assortment breadth, depth and allocation create version control problems and slow decision cycles.

• Can users plan width (option count) and depth (buy quantity) in the same interface?
• How does the system handle pre-season planning, in-season adjustments and markdown optimization in one workflow?
• Are store cluster assignments static or can they be recalculated based on performance?
• What approval workflows and collaboration features exist for cross-functional teams?
• Can the platform support both category-led and vendor-led buying processes?

Request a demonstration using your actual product hierarchies, buying calendar and user roles rather than generic retail examples.

What vendor support and service levels are guaranteed?

Implementation and ongoing support determine whether the platform delivers value or becomes shelfware. Specify service level agreements, escalation procedures and response times before contract signature. Retail software runs 24/7. Confirm the team's incident response SLA, monitoring setup and how breaking changes to third-party APIs (payment processors, marketplace feeds) are handled.

Data portability is the floor, and most buyers still under-negotiate it. Enterprise SaaS contract benchmarking of hundreds of mid-market and enterprise deals signed in 2025 to 2026 found only about 23 per cent of contracts carried an explicit data portability clause, only about 18 per cent specified a machine-readable format, and only about 11 per cent included vendor-assisted migration support. Include exit terms, data export formats and transition assistance in the RFP.

How are pricing models structured and what is the total cost?

AI merchandising platforms use varied pricing models: per-user subscriptions, transaction volume tiers, percentage of GMV or custom enterprise licensing. Ask vendors to provide transparent pricing that includes all components: software licences, implementation services, ongoing support, data storage, API calls and training. Request scenarios at different scale levels (current state, 50 per cent growth, acquisition or new channel launch) to understand how costs evolve.

• Is pricing based on users, SKUs, transactions, revenue or another metric?
• What is included in the base licence versus optional modules (allocation, markdown optimization, space planning)?
• Are there additional fees for integrations, data storage above certain volumes or API usage?
• What are the payment terms and contract length? Are there early termination or data portability costs?
• What professional services are required and what are their rates?

Finally, ask vendors to provide at least two reference customers in similar channels, geographies and scale who completed implementations in the last 18 months. Speak to both merchandising users and IT teams to understand the full experience.