Go back to Market Research
River Logic supplies software for modeling and optimizing manufacturing, logistics, sourcing, capacity, and financial decisions. Enterprise Optimizer generates mathematical models from a visual representation of a business. Value Chain Optimization packages this technology for enterprise planning, while River Logic’s Intelligent Assistant, RIA, helps users construct, run, and interpret scenarios. The product serves strategic, tactical, and selected operational planning tasks. 1 2 4
Overall supply chain score: 5.0/10
| Dimension | Score |
|---|---|
| Supply chain depth | 5.6/10 |
| Decision and optimization substance | 5.2/10 |
| Product and architecture integrity | 5.0/10 |
| Technical transparency | 5.2/10 |
| Vendor seriousness | 4.2/10 |
River Logic’s economic perspective is substantive. Its models connect operational choices with marginal costs, profitability, cash flow, and capital expenditure. Named customer evidence also shows useful changes to planning practice, including redesigning a model’s granularity and testing policy changes in the field. 2 10
The decisive limitation is the gap between optimizing a mathematical representation and making a sound commitment under uncertainty. River Logic has documented stochastic inputs and Monte Carlo workflows for years, but its detailed public example repeatedly optimizes after sampling each possible future. That is not the same calculation as selecting a decision before the future is known. Meanwhile, its current Decision Engine page promises universal feasibility and optimality, and its AI page describes fully accurate operational representations. Those assurances exceed what the disclosed modeling and validation mechanisms establish. 3 4 5
River Logic versus Lokad
The two vendors share a concern with economic decisions rather than forecast presentation alone. River Logic explicitly connects physical operations and financial consequences, including marginal profitability and objectives that can be converted into constraints. 2 15 This supports questions such as whether to expand a facility, change a sourcing network, or accept a different service policy.
Lokad’s inventory optimization organizes recurring operational choices around probabilistic forecasts and expected economic consequences. Its Supply Chain Scientists maintain the numerical logic that produces purchases, allocations, production choices, and other actions. The design perspective behind Lokad treats uncertainty and the cost of making and maintaining decisions as central concerns.
River Logic’s disclosed operating model gives more prominence to constructing an enterprise representation and comparing optimized scenarios. Its customer evidence includes consequential capital and network decisions; its operational allocation offer extends to inventory, suppliers, transfers, and contract terms. 20 Consider committing production to a plant before demand and transport delays are known. The choice must price spare capacity, late deliveries, and forgone sales across plausible futures; optimizing a different allocation after each future is revealed overstates the flexibility available today. The relevant distinction is how each system values that uncertainty when selecting the commitment.
Both approaches depend on skilled modeling. A business-specific representation can be wrong despite being coherent, and a maintained numerical policy can become obsolete. The comparison should include how quickly those errors are found and corrected, the work required, and returns after the full cost of operating the decision process.
What Enterprise Optimizer does
Enterprise Optimizer combines a visual business diagram, input records, and a repository of business rules to generate an optimization formulation. The current platform description includes financial roll-ups and marginal analysis; it says model results can be checked against actual financial statements. 2
The July 2020 Enterprise Optimizer 10.1 data sheet provides a more inspectable processing chain. Imported data resides in a model file, a translator applies the knowledge base and generates equations in memory, a solver processes them, and results are translated back into business outputs. The same document names Gurobi and River Logic’s own Mixed Integer Optimizer. 6
This is a substantial abstraction over writing every equation manually. It can preserve common accounting and process relationships and reduce repeated modeling work. Its quality nevertheless depends on what the library represents, what the customer supplies, and how exceptions are handled. A generated equation has the same exposure to an incorrect business assumption as a handwritten one.
The financial consistency check is useful but limited. Reproducing historical accounts can reveal missing costs, broken flows, or inconsistent quantities. It does not establish how demand, supplier performance, or operating conditions will evolve. Nor does it establish that every economically relevant alternative is available to the optimizer.
The current optimality claim confuses three different assurances
The EO Decision Engine page promises that all decisions are provably feasible and optimal, says complexity requires no approximation, and promotes automatic translation of existing Python models and spreadsheets. It compares its guarantee with lower success rates for LLM optimization modeling. 3
These statements collapse three separate tests:
| Test | What a successful result establishes |
|---|---|
| Formulation correctness | The equations represent the intended problem. |
| Mathematical solution quality | The returned solution satisfies and optimizes those equations to the stated standard. |
| Economic validity | The represented problem and its inputs produce worthwhile decisions in the real business. |
A solver can provide evidence about the second test. It cannot establish the first or third merely by solving. An omitted supplier failure, an incorrect interpretation of a cost, or a missing commercial alternative can survive a complete proof of optimality.
The benchmark comparison is therefore misleading. ORLM studies automated optimization modeling, while ORGEval evaluates whether generated LP/MILP formulations are structurally equivalent to reference models. 16 17 Conditional solver guarantees do not demonstrate that River Logic’s natural-language or code-to-model translation achieves perfect formulation accuracy on those tasks. The vendor publishes no corresponding evaluation on that page.
The universal guarantee also needs a computational scope. Feasibility and optimality depend on the encoded problem, solver termination, tolerances, and the existence of a solution. Supply chain scale and uncertainty do not disappear when constraints are generated automatically. An unconditional promise is particularly inappropriate for a system that lets users alter the problem.
LP/MIP is a serious modeling constraint
The current Decision Engine presentation explicitly invokes LP and MILP. 3 These formulations require a fixed representation for each solve. Real operations include discrete choices, changing commercial terms, uncertain durations, substitutions, and feedback between decisions and demand.
Some effects can be represented exactly within a suitable formulation; others require approximations or a substantial expansion. Adding uncertain futures increases the computational burden, while separating the problem can remove precisely the interactions that make enterprise-wide optimization valuable. River Logic’s public material does not establish its blanket claim that no relevant detail must be approximated away.
The engineering question is whether the chosen representation remains economically useful and revisable. More equations can produce a more detailed model without producing a better decision.
Uncertainty: a documented mechanism with an important limitation
River Logic does disclose stochastic modeling. Its 2020 data sheet identifies uncertain prices, costs, yields, volumes, and transit times. Current VCO and cloud pages continue to advertise Monte Carlo analysis. 6 1 8
The clearest technical example is a December 2015 tutorial using a fictional two-product manufacturer. It samples normally distributed selling prices, solves the model separately for each sample, and stores the results. The workflow runs 1,000 solves through configurable server jobs. 5
That procedure answers what the optimizer would choose if each sampled price were known when the decision was made. A purchase or capacity commitment usually occurs before that information exists. Selecting the best action separately in every future can make the attainable result look better than what one prior commitment could deliver.
A decision under uncertainty needs shared commitments across the possible futures, with later adaptation restricted to information available at that later time. It also needs the relevant dependencies between uncertain inputs. Sampling each scenario and solving it is a straightforward computational technique; it does not establish either requirement.
The tutorial demonstrates a historical workflow, not the full limits of today’s platform. The current public documentation still leaves the decisive step unexplained: how uncertain demand, supply timing, yields, and costs jointly determine the commitment to execute. Random inputs and a distribution of optimized outputs are insufficient evidence.
A digital twin needs predictive evaluation
The AI page says each scenario is a fully accurate representation of operations and that recommendations are valid and implementable. 4 A model can be internally consistent while giving an inaccurate picture of future operations.
For a manufacturing network, the useful tests concern future throughput, yields, lead times, outages, and costs at the horizons where decisions become irreversible. They also concern the consequences of changing policy. A representation that reproduces last year’s accounting does not automatically forecast the result of moving production to a different plant.
The larger the model, the more consequential these dependencies become. A plausible error propagated across purchasing, production, logistics, and finance can produce a very coherent bad plan. Public claims of accuracy need evaluated predictions and operating results, not just a complete-looking diagram.
Customer evidence: substantive progress, carefully separated from prospective savings
FedEx Office exposes the real modeling work
The 2021 announcement describes a joint FedEx Office, Accenture, and River Logic project covering a print-production and delivery network with approximately 2,200 stores and 18 commercial print facilities. 11
A 2022 account is more revealing. FedEx initially put so much detail into its models that they crashed, prompting a revision of the appropriate granularity. The resulting model used a rolling ten-day view. The team explored changes to delivery policies, tested them in selected markets, and expanded successful changes. The quoted potential reduction of up to 45% concerned plant-to-store transport spending; it was an opportunity, not a reported realized reduction. 10
This is strong evidence of an economically useful process: identify an assumption, revise the representation, change an available policy, and test the result outside the model. It also contradicts the idea that complexity can be absorbed without choosing what to represent and how.
The account does not quantify the combined work of FedEx, Accenture, and River Logic or publish a realized net-return calculation. Its value lies in the documented mechanism of improvement, rather than treating the largest percentage as achieved savings.
Philip Morris demonstrates planning reach, not realized savings of every identified opportunity
The vendor’s 2020 Philip Morris case describes a ten-year model in monthly periods, more than 40 manufacturing sites, and 70 input tables. It connects asset and sourcing choices with depreciation, taxes, and capital budgets. Reported process changes include a monthly rather than annual review and scenario analysis shortened from weeks to days. Hundreds of millions of dollars were identified as savings opportunities. 12
A September 2022 expansion announcement includes a PMI executive describing the platform as a valuable decision-support tool and explicitly referring to replication of part of its network. 13 That supports continued use, while preserving the deployment’s scope.
Identifying a large modeled opportunity can justify further work. Realizing it may require capital, commercial negotiation, implementation time, and exposure to different risks. The published case does not provide a cash-flow reconciliation showing which opportunities were executed and what remained after those costs.
Neither historical case measures the incremental benefit of the current RIA interface. The distinction matters when long-standing optimization results are presented alongside a new AI proposition.
Software bill of materials
The public technical record distinguishes several generations and responsibilities:
| Component | Role and evidence |
|---|---|
| Enterprise Optimizer | Business-model compilation, described above. |
| Gurobi and MIO | Named in EO 10.1; the 2020 Gurobi announcement describes a switch from a previous commercial solver. |
| Microsoft Azure | Current vendor-operated SaaS hosting. Connections to AWS or Google Cloud are integrations. |
| Power BI | Current VCO visualization; also part of historical scenario analysis. |
| EO Server | Historical job execution across Windows services, with APIs and external tasks. |
| Azure ML, R Server, SQL Server, SSAS | Explicitly named in the 2017 cloud announcement; their current status is not disclosed. |
These roles come from versioned product material, the solver provider, current product pages, and the dated cloud announcement. 6 7 8 1 18
The 2020 solver announcement credits Gurobi with expected solve-time improvements. River Logic’s contribution is the modeling and integration layer; the announcement does not document a River Logic improvement to Gurobi’s algorithms. 7
The current modeling-consultant vacancy assigns customer-requirement analysis, model construction, validation, integration specifications, and Power BI reporting to the consultant. SQL experience is required; Python and Alteryx are optional skills. 9
Infrastructure costs include model copies, scenario computations, data preparation, reporting, recovery, and the people operating them. The current cloud offer says customers need no separate Azure subscription or Microsoft licenses. 8 That shifts the purchasing arrangement; it does not remove the resources consumed or the cost of customer-specific models.
Configuration is a maintained program
The historical Monte Carlo example uses a control job, cloned execution jobs, variables, branching, queries, and result tables. 5 Dragging those tasks into a visual editor changes the authoring interface, but the result still has program state, dependencies, failure behavior, and testing requirements.
The maintenance question is especially important for generated models. When a library rule or customer assumption changes, someone must determine which decisions change, whether the change is intentional, and whether historical comparisons remain meaningful. Financial reconciliation helps, but it cannot detect every omitted operational effect.
The current Decision Engine advertises versioned models and controlled promotion from a personal library into a corporate library. 3 Those are useful governance features. Their practical value depends on regression testing, explanation of semantic changes, and the effort required to revise the model safely.
Security and extension boundaries
EO 10.1’s documentation explicitly lists Python scripts, SQL procedures, SSIS packages, and PowerShell among server tasks, alongside REST and SOAP access. 6 The older server tutorial also describes remote process execution subject to installation configuration and permissions. 5
These are powerful execution surfaces. A modeling error, unsafe generated code, or compromised dependency can become a production-data problem unless permissions and isolation are enforced outside the submitted logic. Relevant safeguards include restricted data access, network and secret boundaries, controlled dependencies, resource limits, and recoverable state.
The historical server capabilities and the current managed SaaS are different deployment contexts. The public cloud and AI descriptions do not specify which execution capabilities remain available to customer modelers, how external AI models are authorized, or how a compromised task is contained. 8 4
Current code-to-model import is another boundary requiring explanation. Translating a Python formulation, running that code, and accepting model-generated instructions have different risks. The product page describes import and translation, but does not publish an execution or isolation contract. 3 General claims of security, hosting quality, or traceability do not establish that contract.
Implementation time, labor productivity, and economic accountability
River Logic’s May 2026 explanation of deployment speed provides a concrete sequence: initial scoping, model construction with operations and finance, and validation against historical situations. It says most implementations make live planning decisions within six to eight weeks. 14
Early validation can reduce the risk of a long implementation that fails at the end. The same account acknowledges that translating costs, constraints, and revenue logic is substantive work. A managed cloud environment removes some infrastructure tasks; it does not remove this modeling burden.
The article’s financial illustration is weaker. It assumes an improvement equal to 2% of a $500 million annual revenue base and presents approximately $13 million as a yearly cost of delay. Those premises yield $10 million per year; roughly $13 million corresponds to about sixteen months. More fundamentally, the assumed revenue-relative benefit is described as conservative without a supporting measurement. 14
This is an accountability problem because a hypothetical benefit becomes an argument for accelerating the purchase. A testable proposition would specify the initial decisions, baseline, evaluation window, full cost, and responsibility for correcting inadequate results.
Current RIA examples include running validations, preparing scenarios, explaining utilization changes, and suggesting opportunities. 4 These functions can reduce model-operation effort. Public evidence does not yet quantify their effect on total recurring labor or distinguish saved analyst time from increased review and maintenance work.
For strategic decisions, eliminating every human review would be the wrong objective. The relevant productivity gain is faster, better-supported decisions with less total effort. For recurring operational allocation, the standard is more demanding: reliable execution at scale without creating a growing queue of exceptions for specialists.
Technology history and roadmap integrity
| Period | Documented development | What the evidence supports |
|---|---|---|
| 2015 | A detailed EO Server Monte Carlo tutorial. | Actual stochastic input generation and automated scenario solving. |
| 2016 | Launch of a cloud planning platform and announced $10 million funding. | A long-standing scenario-collaboration proposition. |
| 2017 | Azure and Cortana Intelligence integration, naming machine-learning and data components. | Integration of existing cloud technologies, without a published forecasting evaluation. |
| 2020 | EO 10.1 and preferred-solver agreement with Gurobi. | Continued investment in a visual model-to-solver system. |
| 2021–2022 | FedEx Office deployment and PMI expansion. | Named use and substantive planning changes. |
| 2026 | RIA and Decision Engine claims of automatic modeling and universal optimality. | A broader interface and assurance narrative whose strongest guarantees remain unvalidated. |
The tutorial, announcements, technical sheet, and customer accounts establish this chronology. 5 19 18 6 7 11 13
There is continuity in the underlying product. River Logic’s knowledge-based model generation and financial integration predate the LLM wave, and its customer work supports their practical use. The current claim that the company has always been AI rests on that expert-system history. 2 It does not establish that earlier versions had today’s language-model capabilities.
The historical machine-learning integration also needs its own evidence. Naming Azure ML and Cortana in 2017 did not demonstrate calibrated demand or supply forecasts, and the announcement does not supply such a test. 18 The current RIA interface should be assessed on delivered modeling accuracy, safe operation, and measured productivity rather than inheriting the entire credibility of the older optimizer.
The most revealing inconsistency is between the customer account of careful model simplification and field experimentation, and today’s categorical promise of fully accurate, always-optimal decisions. The former is credible engineering practice. The latter conceals the judgment and iteration on which the useful results depend.
Supply chain score
Supply chain depth
| Criterion | Score |
|---|---|
| Economic framing | 7/10 |
| Decision end-state | 6/10 |
| Conceptual sharpness on supply chain | 6/10 |
| Freedom from obsolete doctrinal centerpieces | 6/10 |
| Robustness against KPI theater | 3/10 |
Average: 5.6/10
Economic framing — 7/10. Marginal profitability, capital choices, and operationally derived financial consequences are central. Their treatment under joint uncertainty remains insufficiently explained. 2 12
Decision end-state — 6/10. The product addresses concrete sourcing, capacity, and allocation choices. Much of the demonstrated operating model still ends in analyst-led scenario selection and subsequent implementation. 20 13
Conceptual sharpness on supply chain — 6/10. Cross-functional trade-offs and policy changes receive substantive treatment. Internal model consistency is too often presented as assurance about reality. 15 4
Freedom from obsolete doctrinal centerpieces — 6/10. Explicit objectives and policy alternatives extend beyond consensus-forecast administration. Scenario-by-scenario planning still leaves the commitment under uncertainty unresolved. 1 5
Robustness against KPI theater — 3/10. Large savings opportunities and universal optimality claims are not equivalent to realized economic results. Their presentation weakens otherwise relevant case evidence. 12 3
Decision and optimization substance
| Criterion | Score |
|---|---|
| Probabilistic modeling depth | 3/10 |
| Distinctive optimization or ML substance | 5/10 |
| Real-world constraint handling | 7/10 |
| Decision automation and labor productivity | 5/10 |
| Resilience under real operational complexity | 6/10 |
Average: 5.2/10
Probabilistic modeling depth — 3/10. Stochastic inputs are documented. The public mechanism does not establish a joint policy that respects when information becomes available. 5 6
Distinctive optimization or ML substance — 5/10. Business-knowledge translation is a substantive modeling contribution. External-solver performance and fluent interaction do not establish a distinctive risk-adjusted decision method. 2 7
Real-world constraint handling — 7/10. Named deployments connect operational and financial restrictions across consequential networks. Scaling requires choices about representation, and uncertainty remains incompletely specified. 10 12
Decision automation and labor productivity — 5/10. Customer evidence includes shorter scenario turnaround and automated computational workflows. Total maintenance labor and the incremental contribution of current agents are unmeasured. 12 4
Resilience under real operational complexity — 6/10. FedEx provides a concrete example of revising a model and testing policy changes in operation. The public record does not generalize this into an inspectable product-wide validation process. 10
Product and architecture integrity
| Criterion | Score |
|---|---|
| Architectural coherence | 6/10 |
| System-boundary clarity | 6/10 |
| Security seriousness | 3/10 |
| Software parsimony and ownership cost | 4/10 |
| Compatibility with programmatic and agent-assisted operations | 6/10 |
Average: 5.0/10
Architectural coherence — 6/10. The model-generation core gives the product a coherent center. Current deployment details and component retirement are less visible. 2 18
System-boundary clarity — 6/10. The distinction between modeling, solution packages, hosting, and external data is understandable. Extension permissions and responsibility for downstream execution need clearer contracts. 1 8
Security seriousness — 3/10. Historical extension capabilities reveal substantial execution power. Current public material does not demonstrate containment or recovery appropriate to those risks. 6 8
Software parsimony and ownership cost — 4/10. Managed hosting simplifies procurement, while specialist model work persists. Full operating costs and the lifetime effort of changing assumptions are not quantified. 8 9
Compatibility with programmatic and agent-assisted operations — 6/10. Documented automation can support repeated computation. Current agent behavior lacks a reproducible public control contract. 5 3
Technical transparency
| Criterion | Score |
|---|---|
| Public technical documentation | 6/10 |
| Inspectability without vendor mediation | 5/10 |
| Portability and lock-in visibility | 4/10 |
| Implementation-method transparency | 6/10 |
| Evidence density behind technical claims | 5/10 |
Average: 5.2/10
Public technical documentation — 6/10. Older technical material permits meaningful scrutiny. The newest assurance and agent claims are less technically documented. 6 4
Inspectability without vendor mediation — 5/10. Readers can examine the historical computation sequence and compare current claims with it. Complete generated formulations and current numerical evaluations remain unavailable. 5 3
Portability and lock-in visibility — 4/10. Exporting results does not establish portability of the proprietary business rules and executable model semantics. 6
Implementation-method transparency — 6/10. The delivery sequence and consultant responsibilities identify substantial modeling work. Recurring maintenance effort and failure rates across projects remain unknown. 14 9
Evidence density behind technical claims — 5/10. Named cases and detailed older material support real capability. Universal accuracy and optimality assurances exceed that evidence. 10 3 4
Vendor seriousness
| Criterion | Score |
|---|---|
| Technical seriousness of public communication | 3/10 |
| Technology roadmap integrity | 5/10 |
| Conceptual sharpness | 5/10 |
| Economic accountability | 3/10 |
| Defensibility in an agentic-software world | 5/10 |
Average: 4.2/10
Technical seriousness of public communication — 3/10. The formal guarantees conflate solving a model with representing the business correctly. The comparison with LLM formulation benchmarks does not repair that gap. 3 16
Technology roadmap integrity — 5/10. The modeling concept has continuity and named deployment evidence. The current AI claims lack an equally clear record of evaluated delivery. 19 13 4
Conceptual sharpness — 5/10. The economic modeling perspective is valuable. Categorical twin and solver assurances obscure the distinction between assumptions, predictions, and proved mathematical results. 2 3
Economic accountability — 3/10. Early validation and operational pilots provide useful safeguards. Identified opportunities and hypothetical benefits still dominate the public return narrative, without a complete net-return test. 10 14
Defensibility in an agentic-software world — 5/10. Maintained business knowledge and model implementation experience can remain valuable. Conversational access alone adds little protection; measured improvement in model quality and upkeep is decisive. 2 9
Sources
Value Chain Optimization. Current VCO scope, objectives, and scenario tools. Accessed November 4, 2026.
[2] Enterprise Optimizer Digital Planning Twin Platform
Enterprise Optimizer Digital Planning Twin Platform. Current business-knowledge, financial, and model-generation description. Accessed November 4, 2026.
EO Decision Engine. Current Decision Engine claims, including import, model governance, benchmarks, and optimality. Accessed November 4, 2026.
[4] AI Powered: River Logic Intelligent Assistant
AI Powered: River Logic Intelligent Assistant. Current RIA functions and accuracy assurances. Accessed November 4, 2026.
[5] Monte Carlo Optimization for Beginners
Monte Carlo Optimization for Beginners. Technical tutorial published December 16, 2015 according to the page’s structured metadata; EO Server 4.1 example. Accessed November 4, 2026.
[6] Enterprise Optimizer 10.1 product data sheet
Enterprise Optimizer 10.1 product data sheet. July 2020, version 10.1. The original filename retains an older version number. Accessed November 4, 2026.
[7] Gurobi: River Logic selects the Gurobi Optimizer
Gurobi: River Logic selects the Gurobi Optimizer. Solver-provider announcement; the vendor’s news archive dates it August 18, 2020. Accessed November 4, 2026.
[8] Cloud Native and Built for Choice
Cloud Native and Built for Choice. Current vendor-operated SaaS and integration description. Accessed November 4, 2026.
[9] River Logic careers: Modeling Consultant
River Logic careers: Modeling Consultant. Employer’s public modeling-consultant responsibilities and qualifications. Accessed November 4, 2026.
[10] Improving Capacity Planning and Operations with a Digital Planning Twin
Improving Capacity Planning and Operations with a Digital Planning Twin. Vendor-hosted FedEx Office account, August 2022, including model granularity and policy pilots. Accessed November 4, 2026.
[11] FedEx Office, Accenture, and River Logic collaboration
FedEx Office, Accenture, and River Logic collaboration. Collaboration announcement, August 3, 2021. Accessed November 4, 2026.
[12] Philip Morris International case study
Philip Morris International case study. River Logic-authored 2020 case study, preserved by FeaturedCustomers. Accessed November 4, 2026.
[13] Philip Morris International expands use of River Logic
Philip Morris International expands use of River Logic. Expansion announcement with a named PMI executive, September 20, 2022. Accessed November 4, 2026.
[14] What Weeks to Value Actually Means
What Weeks to Value Actually Means. Nathan Goldstein article, May 29, 2026. Accessed November 4, 2026.
Who We Are. Current company description of model knowledge and objective handling. Accessed November 4, 2026.
[16] ORGEval: Graph-Theoretic Evaluation of LLMs in Optimization Modeling
ORGEval: Graph-Theoretic Evaluation of LLMs in Optimization Modeling. Version 1, October 31, 2025; defines the formulation-equivalence evaluation. Accessed November 4, 2026.
[17] ORLM: A Customizable Framework in Training Large Models for Automated Optimization Modeling
ORLM: A Customizable Framework in Training Large Models for Automated Optimization Modeling. Version 1, May 28, 2024; research on automatic optimization modeling. Accessed November 4, 2026.
[18] River Logic’s 2017 Azure and Cortana Intelligence announcement
River Logic’s 2017 Azure and Cortana Intelligence announcement. Vendor announcement, June 28, 2017. Accessed November 4, 2026.
[19] River Logic launches prescriptive analytics cloud platform
River Logic launches prescriptive analytics cloud platform. Vendor announcement, May 27, 2016. Accessed November 4, 2026.
Order Allocation. Current operational allocation, sourcing, and contract scope. Accessed November 4, 2026.