Retail Demand Forecasting in 2026: Blue Yonder, Relex, o9, ToolsGroup, and Why Walmart Built Its Own
The demand forecasting vendor landscape sorted honestly — what Relex does that Blue Yonder doesn't, where o9 sells the platform story, and why Walmart and Amazon run their own forecasting stacks.
Retail demand forecasting is the oldest enterprise AI use case still actively being re-platformed. Statistical forecasting at SKU-store level has been in production since the early 2000s; machine-learning augmentation became standard by the late 2010s; and the 2024-2026 wave is the transformer-based, exogenous-feature-rich, foundation-model-flavoured rebuild. The vendors competing for that rebuild are well-funded, technically credible, and selling overlapping stories in ways that confuse buyers.
This is the honest landscape — what each vendor actually does, who wins which deals, and where the largest retailers gave up and built their own.
The category in 2026#
Demand forecasting at a real retailer is not one model. It is a portfolio of forecasts at different aggregations (chain, region, store, SKU, day, hour), feeding different downstream decisions (replenishment, allocation, promotional planning, labour scheduling, pricing). The vendor that solves all of those well is the platform sale; the vendor that solves one of them well is the point-tool sale that gets acquired or commoditised over a five-year cycle.
The major buyers in 2025-2026 are mid-tier grocery, specialty apparel, hardlines retail, and the regional convenience chains that watched Walmart and Amazon build forecasting in-house and concluded they could not. Walmart’s Element platform and Amazon’s internal forecasting stack are the reference architectures everyone is implicitly benchmarking against.
Blue Yonder#
Blue Yonder (the renamed JDA Software, owned by Panasonic Connect since 2021) is the incumbent. The product breadth is the widest in the category — demand, replenishment, allocation, workforce, transportation. The codebase is older, the cloud rebuild is genuinely real but uneven across modules, and the customer base skews to large established retailers that bought JDA twenty years ago and have not been ready to switch.
Blue Yonder’s 2024-2025 message has been the “Cognitive Solutions” rebrand — generative AI on top of the planning suite, Microsoft Fabric and Snowflake integration, and the Luminate Platform as the unifying data layer. The honest read: the AI is real where the underlying data quality is real; the legacy modules drag against the rebuild story; and customers report mixed results on time-to-value for the newer features. The DHL acquisition rumour in early 2025 did not materialise, and the company continues as a Panasonic subsidiary.
Relex Solutions#
Relex (Finnish, founded 2005, secured a meaningful 2024 secondary round at a billion-dollar-plus valuation) is the vendor most often winning competitive grocery deals against Blue Yonder. The wedge is honestly built — fresh-food forecasting, store-level granularity, and a planning engine that handles the categorical reality of grocery (perishables, weather sensitivity, promotional cannibalisation, multi-store allocation under constraint).
Tesco, Ahold Delhaize, Coop Denmark, Loblaw, Morrisons, and a long tail of European and North American grocers run Relex at scale. The 2025 product push added a generative-AI planner interface — a natural-language layer over the underlying optimisation — and tighter integration with Microsoft Power Platform for the long tail of planner workflows.
Where Relex loses: very large general-merchandise retailers that want one platform for grocery and non-food. Where it wins: any retailer where fresh food is a meaningful share of revenue.
o9 Solutions#
o9 (Dallas-headquartered, founded by former i2 Technologies leadership, IPO filed and pulled in 2024, refile expected in 2026) is the platform-story vendor. The pitch is “Enterprise Knowledge Graph plus AI plus planning” — a single semantic model spanning demand, supply, finance, and commercial planning. The customer base is heavy on consumer goods (Unilever, P&G, Anheuser-Busch InBev), durables (Caterpillar, Deere), and increasingly retail (Walmart Mexico, Aldi Süd, several specialty apparel brands).
The o9 advantage in retail is the cross-functional planning story — demand plan, supply plan, financial plan, and S&OP all sharing one data model. The risk is implementation cost and timeline; the typical o9 program is multi-year and multi-system-integrator, and the deals that go wrong tend to go wrong on integration sprawl.
ToolsGroup#
ToolsGroup (acquired by Stellex Capital in 2023, made a bolt-on acquisition of Evo Pricing in 2024) is the probabilistic-forecasting specialist. The technical credibility is real — stochastic demand modelling, multi-echelon inventory optimisation, and meaningful production deployments at retailers including Costa Coffee, BP, Speedo, and a long list of mid-tier industrials. The product position is “we do the math others don’t,” and the typical buyer is a planning leader frustrated with the over-simplified forecasts coming out of the larger platforms.
ToolsGroup will not win the enterprise platform sale against o9 or Blue Yonder. It will win the specific-problem sale where forecast accuracy is the binding constraint.
SAP IBP and Joule#
SAP Integrated Business Planning is the default for SAP-ERP retailers — Walmart Mexico (partially), Costco’s international operations, several large European retail groups. The 2024-2025 story has been Joule, SAP’s generative AI layer, layered across IBP, S/4HANA, and the broader Business Technology Platform. The honest assessment: Joule is real, the integration with IBP is genuine, and the customers extracting value are the ones that had a working IBP deployment before Joule arrived. For retailers without an existing SAP planning footprint, the IBP-plus-Joule sale faces a long sales cycle against more focused competitors.
Walmart’s internal stack#
Walmart’s Element platform — the internal AI/ML platform underneath replenishment, pricing, labour scheduling, and increasingly logistics — is the reference architecture the vendors compete against. The honest version: Walmart built it because no commercial platform could handle the scale of Walmart at the granularity Walmart wanted, and the team behind it included alumni of Amazon, Google, and the major platform vendors. The forecasting models are a mix of gradient-boosted trees, deep learning for specific high-volume SKUs, and an ensemble layer that picks the right model per category and store.
For any retailer not at Walmart’s scale, building this internally is a multi-hundred-engineer multi-year commitment. The economics work for Walmart, Amazon, Target, Kroger, and Tesco. They do not work for almost anyone else.
The holiday-season failure patterns#
The patterns we see fail in production, especially during Q4 demand peaks:
- Promotional cannibalisation modelled too simply — the forecast assumes the promotion drives incremental volume; in reality the promotion shifts volume from other SKUs, other stores, or other time periods. Sales come in on plan; margin comes in well below plan.
- Weather sensitivity hard-coded — the model knows it rained on day T but doesn’t know the new long-range forecast for day T+7 has shifted, and replenishment is set days in advance.
- New-item forecasting that defaults to a category average — works for the average new item, misses badly on the items that go viral on TikTok or fail completely.
- Multi-channel demand stitching — online demand and store demand modelled separately, leading to over-replenishment in stores when online demand spikes and the fulfilment shifts to BOPIS.
- Forecast horizon mismatch — the model is optimised for the four-week ahead horizon; the inventory commitment is twelve weeks ahead; nobody notices until the orders land.
The 2024 holiday at several major US specialty retailers exposed all five patterns. The 2025 holiday was better at retailers that did the post-mortem work; meaningfully worse at retailers that didn’t.
What we recommend in 2026#
For a mid-tier retailer asking which platform to evaluate:
- Grocery and convenience with meaningful fresh — start with Relex. The fresh modelling is honestly differentiated.
- General merchandise or apparel with strong existing data engineering — evaluate Blue Yonder Cognitive Solutions and o9 in parallel; the decision usually comes down to existing SAP or Oracle footprint and the size of the cross-functional planning ambition.
- Forecast-accuracy as the binding constraint — ToolsGroup is the focused buy.
- Already on SAP with working IBP — Joule on top of IBP is the path of least resistance.
The internal-build path is reserved for retailers above roughly 50 billion USD in revenue with existing platform engineering capability. Below that, the build economics rarely work.
Related reading#
- The retail personalization playbook
- Supply chain visibility platforms
- Grocery AI and Instacart Connected Stores
Where pdpspectra fits#
We help retailers stand up the data layer underneath whichever planning vendor they choose — ERP, POS, e-commerce, weather, and external-signal pipelines, plus the feature store and orchestration layer that production forecasting needs. Our ML and MLOps practice handles the model lifecycle.
Forecasting accuracy is mostly a data engineering problem dressed in ML clothing. If you are scoping a platform replacement or a forecasting accuracy program, tell us about the catalogue.