Retail Data Consulting Firms, Ranked by Public Evidence

Depending on the project, also compare the firms on the data science, data analytics and AI consulting pages.

The M5 accuracy competition asked teams to forecast 42,840 hierarchical time series of Walmart unit sales; 5,507 teams took part (Makridakis, Spiliotis and Assimakopoulos, International Journal of Forecasting, 2022). Findings a buyer can use:

• Most entries lost to a simple benchmark. Only 415 teams, 7.5%, beat the best benchmark, a bottom-up exponential smoothing model, with the forecasts they submitted.
• The winners were machine learning, used in specific ways. All top-performing methods were machine learning methods, most built on LightGBM; they combined several models and learned across many series at once instead of fitting one model per product.
• External data mattered. All winning submissions used information beyond sales history, and price-related features were among the most important. Among the benchmarks, adding promotions and special events improved exponential smoothing by about 6% and ARIMA by 13%.
• Gains shrink where orders are placed. The top 50 teams improved on the benchmark by an average of 40% at the top of the hierarchy, about 23% in the middle levels and about 3% at the product, product-state and product-store levels.
• Accuracy gains were not costed. The organizers could not translate accuracy gains into inventory savings, because cost data such as holding costs and lost sales were not available.

The practical lessons: ask any vendor at which level its accuracy figures are measured, insist on a backtest on your own history against a simple benchmark, and judge the result by stock, waste and service levels as well as forecast error.

• PCI DSS applies to entities that store, process or transmit cardholder data, or could affect the security of the cardholder data environment (PCI Security Standards Council). Version 4.0.1 has been the only active version since December 31, 2024. Keep full card numbers out of analytics platforms unless the platform is meant to be in scope.
• The CCPA applies to for-profit businesses that do business in California and meet one of its thresholds on revenue, the volume of residents' personal information they handle, or revenue from selling it. Consumers have the rights to know, delete, correct, opt out of sale or sharing, and limit the use of sensitive personal information, and must not be discriminated against for using them (California Attorney General).
• Other jurisdictions have their own rules. If you sell to customers elsewhere, give bidders the list of laws your legal team applies.

For a customer data project, ask how opt-outs and deletion requests reach every copy of the data, including feature stores, model training sets and audiences shared with partners. Governance-focused firms are listed on the data governance consulting page.

1. Backtest. Will you forecast a past period of our data, at item-and-store or item-and-warehouse level, and compare the result with a simple benchmark such as seasonal naive or exponential smoothing?
2. Demand signals. How do you handle stockouts, promotions, price changes, new products and holidays?
3. Business measures. How will we measure the effect on stock, waste, availability and margin, not only forecast error? Is there a control group of stores or categories?
4. Customer data. How do you match customers across channels, and how do consent choices and deletion requests flow through to models and audiences?
5. Systems. Which point-of-sale, ERP, warehouse and e-commerce platforms have you integrated, and on which data platform?
6. Ownership. Who owns the code, features and trained models when the contract ends, and what documentation and retraining runbooks will you hand over?
7. Evidence. Which retail case studies on your website did this team deliver, and can we speak to that client?

Send every shortlisted firm the same scope and questions with our RFP template.

1. Assessment. Inventory sources, data quality and latency; pick the first decisions to improve and the measures that will judge them.
2. Foundation. Product, store and customer masters; daily sales, inventory and price feeds; a promotions and events calendar.
3. Pilot. One use case on a subset of categories or stores, with a backtest first and a live comparison against a control group.
4. Rollout. Wider deployment, integration into ordering, pricing or marketing systems, and monitoring for forecast drift.
5. Run. Retraining, support and handover to your team.

Rates and worked budget examples are on the pricing page; platform costs are covered in our data platform TCO analysis.