Go back to Market Research
OPTANO develops optimization applications for network design, production, transport, inventory, and workforce planning. Its platform combines data integration, scenario management, user interfaces, and customer-specific mathematical models. OPTANO Modeling provides a separate .NET programming interface through which developers construct models and call optimization solvers. The company has belonged to Kearney since 2022. (1, 2, 18, 24)
Overall supply chain score: 5.3/10
| Dimension | Score |
|---|---|
| Supply chain depth | 5.2/10 |
| Decision and optimization substance | 5.0/10 |
| Product and architecture integrity | 5.0/10 |
| Technical transparency | 6.0/10 |
| Vendor seriousness | 5.2/10 |
OPTANO offers inspectable modeling tools and substantial application engineering. Its central weakness is the dependence on mathematical programs whose tractability and economic adequacy must be established separately for each operation. A solver can optimize an encoded objective while missing the uncertainty, nonlinear consequences, or available decisions that determine actual profit. The vendor’s blanket speed claims obscure how demanding its own published industrial examples become. (4, 10, 25)
Its more careful publications recognize those limits: model corrections, realistic comparisons, and even a feasibility study finding little incremental benefit. That record is more useful than the broad AI positioning. The unresolved commercial question is whether OPTANO can sustain economically sound decisions after all modeling, scenario review, integration, and maintenance work are counted. (8, 9, 15)
OPTANO versus Lokad
OPTANO builds an application around a mathematical representation of the customer’s planning problem. Planners compare scenarios and use the resulting plans to coordinate operations. Lokad’s inventory optimization service organizes the work around recurring decisions, their financial consequences, and uncertainty about future demand and supply.
The difference becomes important when a target conflicts with economic value. Preserving a minimum stock level may force expensive production; maximizing service may consume inventory needed for more valuable demand. OPTANO can represent costs and competing objectives, but the chosen hierarchy or weights still determine which trade-offs the application is allowed to make. A mathematical optimum does not establish that those priorities reflect the business’s economics.
Lokad’s Supply Chain Scientists are responsible for building and maintaining the decision logic, including revising a formulation that produces costly recommendations. Introduction to Supply Chain develops that perspective. The relevant comparison includes the people needed to operate each approach: a technically sophisticated model can still impose a large recurring workload if planners and consultants must continually inspect and repair its decisions.
What the optimization core establishes
Public modeling tools, bounded mathematical capabilities
OPTANO Modeling exposes mathematical programs through a .NET API, with public documentation, examples, and an MIT license. Supported interfaces include linear, mixed-integer, and quadratic optimization capabilities across several solvers. The NuGet package reached version 3.31.0.685 in September 2026; its release note fixes incorrect CPLEX gap reporting for multiple objectives. This is maintained software with a visible technical surface. (2, 3, 4)
The limitation is the model class. Supply chains contain discrete choices, nonlinear costs, substitution, uncertain yields and receipts, and changing opportunities. Linear programming forces strong simplifications. Mixed-integer formulations can express more choices, but the additional detail expands the search and can make useful solutions difficult to obtain. Decomposing the business into easier local problems can discard precisely the dependencies that create economic value.
Adding future scenarios does not remove that problem. It expands a formulation that may already be computationally strained, while requiring a probabilistic account of the future that the solver itself does not supply. The decisive test is whether the resulting model preserves the important economic choices at a practical scale. A small optimality gap only concerns the problem that was encoded.
The vendor’s research contradicts its unrestricted speed claims
The current solver page promises millions of decision variables and thousands of what-if scenarios in seconds. It does not specify the workload, formulation, hardware, or approximation behind those statements. (25)
OPTANO’s 2024 algorithm-tuning paper describes a much harder industrial reality. Its network-planning instances had more than 470,000 variables, including over 50,000 integers, and were impractical to solve to optimality even over days. With default Gurobi 8.1 parameters, two of thirty instance-and-seed runs took nearly eight hours to find a first feasible solution. Tuning reduced the reported maximum to four minutes, or nine minutes under leave-one-out validation. The study used three customer instances and smaller training copies with some decisions fixed in advance. (10)
This is real engineering beyond simply connecting a third-party solver. It also shows why variable count alone is a poor scale claim. A useful configuration for one formulation does not establish robust performance after the network, uncertainty model, or decision space changes. Nor does faster feasibility establish better financial decisions: the formulation may still omit an expensive failure mode.
The broader OPTANO Algorithm Tuner exposes its code and methods. It searches parameter settings for target algorithms, including genetic and other search approaches. That work improves how a chosen algorithm runs; it does not decide whether the business problem has been formulated correctly. (11)
Hierarchical objectives can protect the wrong priorities
The public Modeling documentation explains hierarchical optimization precisely. An objective is optimized, then a constraint prevents lower-priority objectives from worsening its result. Objectives at the same level can be combined using weights. (5)
At Daimler Truck, the published application orders its goals around satisfying demand, preserving network robustness and minimum stocks, and then minimizing costs. It connects planning with stock and forecast information, SAP interfaces, a data warehouse, and tracking of implementation by plants. (6)
That hierarchy has an economic consequence: a higher-priority target can remain protected even when a small concession would save substantial money. Some commitments are genuine obligations and belong in hard constraints. Others are management preferences whose value should be compared with their cost. The case does not disclose how those categories are distinguished or how priorities are revised when the resulting plan is economically poor.
Uncertainty and inventory
OPTANO’s current inventory description explicitly uses distribution models for both demand and delivery times. It also retains safety stocks, ordering cycles, minimum-stock rules, and comparisons between warehouse strategies. This is substantive recognition of supply uncertainty, but the page does not disclose calibration, dependence between uncertainties, or the joint economic policy behind the stocking decisions. (13)
The October 2026 capacity-planning interview is more careful about scenario decisions than generic what-if marketing. It keeps today’s commitments common across possible futures, allows only adjustments that remain feasible later, and distinguishes the mathematical bound from savings against an existing plan. It also calls for comparisons using information available at the time and for correcting models when recommendations expose missing rules or outdated assumptions. (9)
These are necessary disciplines. They do not establish a scalable implementation across a large stochastic supply chain. The public explanation lacks a calibrated joint demand-and-supply forecast, decision-level validation, and the cost of maintaining the expanded model. Scenario consistency prevents one modeling error; it cannot make an inadequate set of futures complete.
Reinforcement-learning research exposes the unresolved design problem
A research chapter published in October 2026, co-authored by OPTANO staff, reports the ML4SafetyStock project. It derives demand and lead-time distributions from historical network data and builds a Python/NumPy simulator with backlogs, costs, and service requirements. Agents at individual warehouses share a reward function. In the reported multi-echelon comparisons, the approach tended to hold more stock than a decomposition heuristic while maintaining operation in difficult configurations. The authors explicitly identify the results as simulations and retain assumptions about demand, lead times, and network structure. (14)
This work investigates a real limitation of classical approaches. Its result is not a deployed probabilistic supply chain service. The simulator must predict the consequences of actions accurately enough to train a useful policy, and its reward must value those consequences correctly. A network that avoids collapse by holding more stock may still destroy value through capital consumption or obsolescence. The research leaves further refinement of observations and rewards as continuing work; it does not publish a production financial comparison.
Customer evidence and economic accountability
Holcim: an implemented network model with incomplete return accounting
The Holcim case covers production and delivery of cement across a network, including sales prices, production and logistics costs, capacities, and make-or-buy alternatives. It describes participation from sales, production, logistics, and finance. The software supported strategic and tactical scenarios, with a requirement that optimization runs take no more than five minutes. (7)
The case establishes cross-functional modeling and operational delivery. Its reported benefits are qualitative: cost and stock reductions, supply-demand alignment, and improved choices. It provides no normalized financial baseline, maintenance cost, or quantified continuing planning workload. The time requirement is an application requirement, not a universal result for arbitrarily enlarged models.
Lavazza: an unusually useful negative business case
Lavazza Professional commissioned a study of technician routing using three years of historical orders. The study found only marginal improvement from optimized assignment because the existing territories, experienced staff, and team coordination already performed well. The published response emphasized knowledge transfer and organization rather than a large software rollout. The customer also highlighted the risk of over-engineering the model. (8)
Publishing this result gives OPTANO’s economic claims more credibility. It demonstrates a willingness to distinguish technical feasibility from a worthwhile investment. The remaining limitation is the baseline’s scope: proximity to an optimum applies to the modeled routing task. It cannot establish that the entire service operation, including downtime and spare-parts availability, is economically optimal.
A recurring production pattern trades flexibility for lower coordination cost
OPTANO’s Rhythm Wheel approach fixes a repeating product sequence, typically over eight to twenty weeks, while adjusting quantities using current stocks and forecasts. Planners can insert additional products. An anonymous pilot covering five lines in three countries reports 10–15% lower inventories, one or two fewer shifts per line, and 5–10% lower operating costs. (20)
A stable sequence can reduce changeover and coordination costs. It also restricts the available actions before future demand is known. Its economic merits depend on whether that restriction costs less than the stability it buys. The reported pilot does not disclose the observation period, accounting baseline, or full planning and engineering effort. Its percentages cannot establish that the same rule remains appropriate for long tails, introductions, or large demand shifts.
Productivity, model maintenance, and ownership cost
OPTANO’s application workflow is centered on scenario creation, comparison, and collaboration. Its public platform includes web clients, dashboards, Excel exchange, and Power BI integration. These features can reduce data handling, but their existence does not establish that routine decisions execute successfully without planner review. (1)
The important distinction is between faster planning activity and less planning work. Producing many alternatives quickly can still leave a team comparing them, reconciling priorities, and explaining changes. The reviewed cases do not measure decisions per total human hour, including customer planners, OPTANO specialists, and consulting partners. Kearney ownership makes that combined operating cost particularly relevant.
The custom-development interview describes a division of labor: consultants identify objectives and constraints, operations-research experts build the model, and developers turn it into an application. Reusable platform functions reduce repeated development, while the decision model remains customer-specific. (15)
That organization can handle difficult requirements, but the responsibilities continue after installation. A new supplier, changed product mix, or revised service commitment can require changes to data semantics, constraints, and objective values. Solver tuning adds another maintenance task when the structure of the problem changes. Model code and application code need to remain consistent with the operation and with each other.
OPTANO Modeling supports ordinary software-engineering practices, including versioning and unit and integration tests. Those mechanisms make changes more reviewable. They cannot establish that a cost omitted from the formulation has been discovered. (22)
The October interview’s correction process addresses the holimization problem directly: a planner’s objection may reveal a missing rule, an outdated assumption, or a previously overlooked opportunity. The remaining evidence gap is a production example tracing economic failure through reformulation, testing, and improved subsequent decisions. Without that record, the cost and effectiveness of the continuing improvement process remain uncertain. (9)
Bill of software materials and security
| Layer and evidence | Documented materials | Ownership consequence |
|---|---|---|
| Modeling package, September 2026 | .NET Standard 2.0; FParsec, FSharp.Core, and MathNet dependencies | A relatively legible modeling library, distinct from the full deployed application. |
| Solver interfaces | Alternative commercial and open-source solvers, including Gurobi and CPLEX | Solver licensing, compatibility, and performance remain operating dependencies. |
| Current application platform | Web clients, configurable imports, Excel exchange, Power BI, OIDC | Shared application functions reduce repeated work but still require integration and permission management. |
| Algorithm Tuner | .NET, Akka components, logging, mathematical libraries, solver adapters | A separate tuning tool with its own runtime and dependencies. |
| April 2021 IT Assistant vacancy | Windows server/client administration, virtualization, cloud infrastructure | Historical evidence of internal infrastructure work; it does not identify the current customer deployment stack. |
| ML4SafetyStock research | Python/NumPy simulator | An experimental environment whose transfer into commercial operation remains unspecified. |
Solver selection is project-specific. The full cost depends on the application delivered: its data warehouse, connections, model runtime, reporting, and support arrangements. Public materials do not quantify that footprint or the resources needed for representative production workloads.
The platform documents role-based access and OpenID Connect for reusing existing identities. The certificate page foregrounds ISO, TISAX, and SOC 2; those attestations do not explain how a solver worker is isolated, which data an integration credential can read, or who can deploy altered model code. (1, 17)
For a custom optimization application, the critical boundary separates ordinary planning permissions from the authority to change executable logic and production data. The public architecture does not describe that boundary in enough detail to assess containment.
Portability is better at the modeling layer than at the application layer. An MIT-licensed API and explicit solver interfaces make technical dependencies visible. Moving a deployed application still requires its data mappings, customer-specific logic, integrations, and operating knowledge. The public offering does not specify how those assets transfer when a customer changes provider. (2, 15)
Historical promises and follow-through
| Period | Public development | Assessment |
|---|---|---|
| 2009–2022 | Optimization business followed by Kearney acquisition | The 2022 announcement describes more than thirty joint projects, but also makes sweeping AI and universal-optimization claims. |
| 2021 | Algorithm Tuner 2.1.0 | Public methods and code give the older algorithmic positioning substance. |
| 2024 | Published tuning research | A concrete contribution improves a third-party solver’s behavior; the study also exposes difficult scalability limits. |
| 2024 technology positioning | Continuously learning AI and faster optimization | The marketing is broader than the disclosed application-level learning and performance evidence. |
| 2025 | Digital-twin and Rhythm Wheel explanations | The vendor discusses abstraction and actual planning rules; the predictive validity of the resulting models remains unmeasured. |
| 2026 | Modeling maintenance, inventory research, and a careful capacity-planning discussion | These are identifiable advances, with simulation findings and conceptual guidance still requiring production validation. |
Sources: 3, 9, 10, 12, 14, 18, 19, 20, 21.
The 2022 acquisition announcement promoted AI solutions dating back to 2009 and claimed an optimal solution to any problem. The disclosed technical work supports a much narrower history of mathematical modeling and algorithm configuration. That early AI positioning overstated the breadth of the capability; it should not be read as evidence of today’s general-purpose reasoning or autonomous execution. (18)
Current AI material adds claims of continuous learning and improved forecasting without a defined benchmark or disclosed update process. By contrast, the inventory research names its assumptions and remaining work. The vendor’s strongest technical communication is therefore more restrained than its commercial overview. (19, 14)
The digital-twin discussion also recognizes that a model represents a selected part of a business, with abstraction chosen to control complexity. That is a useful starting point. Once the model predicts how a delay or countermeasure changes costs and service, its conditional predictions require validation. The article does not provide that validation. A coherent virtual network can still produce systematically wrong forecasts of economic consequences. (21)
The engineering record supports narrower claims than universal speed, continuous learning, and comprehensive optimization. The commercial promise should match the workloads and outcomes established by the technical publications.
Supply chain score
Each dimension averages five criteria on a 0–10 scale. The overall score averages all 25; 5.28 rounds to 5.3.
Supply chain depth
| Criterion | Score |
|---|---|
| Economic framing | 6/10 |
| Decision end-state | 6/10 |
| Conceptual sharpness on supply chain | 6/10 |
| Freedom from obsolete doctrinal centerpieces | 3/10 |
| Robustness against KPI theater | 5/10 |
Average: 5.2/10.
Economic framing — 6/10. Models include meaningful costs and commercial choices. Hierarchical targets can override economic trade-offs without a disclosed justification for their priority. (5, 6, 7)
Decision end-state — 6/10. Applications generate operational and strategic plans and can track implementation. The public offering remains clearer about planner-supported choices than unattended decision production. (6, 24)
Conceptual sharpness on supply chain — 6/10. Published examples address shared resources, timing, and cross-functional consequences. The joint economic treatment of uncertainty is less established than the coordination of known constraints. (7, 9)
Freedom from obsolete doctrinal centerpieces — 3/10. Safety-stock rules, service targets, and fixed planning patterns remain central. Optimizing those parameters does not establish that the chosen policy captures the most valuable actions. (13, 20)
Robustness against KPI theater — 5/10. Some publications distinguish a worthwhile investment from an attractive mathematical result. Anonymous percentage improvements and broad performance claims remain insufficiently specified. (8, 20, 25)
Decision and optimization substance
| Criterion | Score |
|---|---|
| Probabilistic modeling depth | 4/10 |
| Distinctive optimization or ML substance | 5/10 |
| Real-world constraint handling | 7/10 |
| Decision automation and labor productivity | 4/10 |
| Resilience under real operational complexity | 5/10 |
Average: 5.0/10.
Probabilistic modeling depth — 4/10. The research treats demand and supply uncertainty explicitly. Public evidence does not establish a calibrated, deployed joint model with economically validated decisions at commercial scale. (13, 14)
Distinctive optimization or ML substance — 5/10. Algorithm configuration provides an identifiable engineering contribution. It improves computation within a chosen formulation without resolving the formulation’s supply chain limitations. (10, 11)
Real-world constraint handling — 7/10. Named applications represent consequential production, inventory, and network restrictions. Computational difficulty limits how much detail can be retained and explored jointly. (6, 7, 10)
Decision automation and labor productivity — 4/10. Data handling and calculation can be automated. Scenario review and custom-model support remain substantial, without a total-workload measurement demonstrating sustained productivity. (1, 15)
Resilience under real operational complexity — 5/10. OPTANO explicitly describes correcting incomplete models. A documented economic failure followed by successful reformulation would be stronger evidence than the present methodological explanation. (9, 22)
Product and architecture integrity
| Criterion | Score |
|---|---|
| Architectural coherence | 6/10 |
| System-boundary clarity | 6/10 |
| Security seriousness | 4/10 |
| Software parsimony and ownership cost | 4/10 |
| Compatibility with programmatic and agent-assisted operations | 5/10 |
Average: 5.0/10.
Architectural coherence — 6/10. A modeling layer, solver interfaces, and reusable application functions form a legible architecture. The integrity of each customer’s integrated application still depends on continuing engineering. (1, 2, 15)
System-boundary clarity — 6/10. The roles of model builder, solver, and planning application are distinguishable. Production authority and failure handling are less fully documented. (6, 9)
Security seriousness — 4/10. Identity federation and access roles are documented mechanisms. They leave model deployment, worker containment, and integration privileges insufficiently explained. (1, 17)
Software parsimony and ownership cost — 4/10. The modeling library is inspectable, but the full application and its upkeep are not quantified. Custom logic and solver tuning can consume continuing specialist effort. (3, 10, 15)
Compatibility with programmatic and agent-assisted operations — 5/10. Developer interfaces and testable models support automated engineering work. They do not establish unattended operation of the customer-facing decision process. (2, 22)
Technical transparency
| Criterion | Score |
|---|---|
| Public technical documentation | 7/10 |
| Inspectability without vendor mediation | 6/10 |
| Portability and lock-in visibility | 6/10 |
| Implementation-method transparency | 6/10 |
| Evidence density behind technical claims | 5/10 |
Average: 6.0/10.
Public technical documentation — 7/10. APIs, examples, release notes, and research expose actual mechanisms. The public technical reference is more complete for the tools than for delivered supply chain applications. (2, 5, 23)
Inspectability without vendor mediation — 6/10. Developers can examine the modeling interface and tuning code directly. Reproducing customer behavior still requires private models, data, and application configuration. (2, 10, 11)
Portability and lock-in visibility — 6/10. Licensing and solver boundaries are explicit. Transfer of the surrounding application and operational knowledge is less transparent than reuse of the modeling layer. (2, 4, 15)
Implementation-method transparency — 6/10. The division of engineering responsibilities is visible. Public accounts leave ongoing staffing, change effort, and commercial acceptance conditions largely unspecified. (6, 15)
Evidence density behind technical claims — 5/10. Research and named applications support real technical work. Broad solver and AI claims substantially exceed the scope of the published demonstrations. (10, 14, 19, 25)
Vendor seriousness
| Criterion | Score |
|---|---|
| Technical seriousness of public communication | 4/10 |
| Technology roadmap integrity | 6/10 |
| Conceptual sharpness | 5/10 |
| Economic accountability | 6/10 |
| Defensibility in an agentic-software world | 5/10 |
Average: 5.2/10.
Technical seriousness of public communication — 4/10. Universal performance language conflicts with the limitations evident in the technical record. Buyers need the bounded explanation supplied by the research, not the blanket claim. (10, 18, 25)
Technology roadmap integrity — 6/10. Software maintenance and published methods show continuity. New research remains distinguishable from delivered application capabilities, though the commercial AI narrative blurs that distinction. (3, 12, 14, 19)
Conceptual sharpness — 5/10. The careful publications distinguish computational results from business outcomes. The broader proposition still underplays how difficult it is to preserve economically relevant behavior in tractable models. (9, 21, 25)
Economic accountability — 6/10. Publishing a marginal-benefit feasibility result is meaningful evidence of restraint. Contractual responsibility, total-cost reconciliation, and remediation after disappointing production returns remain unspecified. (8, 15)
Defensibility in an agentic-software world — 5/10. Domain modeling and maintained integrations remain useful assets. Accessible code-generation tools put pressure on generic application-building work; reliable economics and lower total operating labor must justify the specialist platform. (2, 15, 22)
Sources
Sources were accessed on November 4, 2026.
[1] OPTANO platform architecture
OPTANO platform architecture. Published 2025-09-18 (technology index); accessed November 4, 2026.
[2] OPTANO Modeling documentation
OPTANO Modeling documentation. Published undated; accessed November 4, 2026.
[3] OPTANO Modeling package 3.31.0.685
OPTANO Modeling package 3.31.0.685. Published 2026-09-03; accessed November 4, 2026.
[4] Supported solver interfaces
Supported solver interfaces. Published undated; accessed November 4, 2026.
[5] Hierarchical and weighted optimization
Hierarchical and weighted optimization. Published undated; accessed November 4, 2026.
[6] Daimler Truck global program planning
Daimler Truck global program planning. Published undated; accessed November 4, 2026.
[7] Holcim network optimization
Holcim network optimization. Published undated; accessed November 4, 2026.
[8] Lavazza Professional feasibility study
Lavazza Professional feasibility study. Published undated; accessed November 4, 2026.
[9] Rolling capacity planning: OPTANO and Gurobi interview
Rolling capacity planning: OPTANO and Gurobi interview. Published 2026-10-07; accessed November 4, 2026.
[10] Configuring Mixed-Integer Programming Solvers for Large-Scale Instances
Configuring Mixed-Integer Programming Solvers for Large-Scale Instances. Published 2024-05-30; accessed November 4, 2026.
[11] OPTANO Algorithm Tuner documentation
OPTANO Algorithm Tuner documentation. Published undated; accessed November 4, 2026.
[12] Algorithm Tuner change log
Algorithm Tuner change log. Published 2021-11-12 (latest listed release); accessed November 4, 2026.
[13] Supply chain, inventory, and production functions
Supply chain, inventory, and production functions. Published undated; accessed November 4, 2026.
[14] ML4SafetyStock: machine learning for safety stocks in production and distribution networks
ML4SafetyStock: machine learning for safety stocks in production and distribution networks. Published 2026-10-23; accessed November 4, 2026.
[15] When complex planning needs custom software
When complex planning needs custom software. Published 2026-02-04; accessed November 4, 2026.
[16] OPTANO IT Assistant recruitment notice
OPTANO IT Assistant recruitment notice. Published 2021-04-13; accessed November 4, 2026.
OPTANO certificates. Published undated; accessed November 4, 2026.
[18] Kearney acquisition announcement
Kearney acquisition announcement. Published 2022-11-09; accessed November 4, 2026.
[19] AI and machine learning at OPTANO
AI and machine learning at OPTANO. Published 2024-04-05 (technology index); accessed November 4, 2026.
[20] Rhythm Wheel production planning
Rhythm Wheel production planning. Published 2025-08-22; accessed November 4, 2026.
[21] From transit to triumph: digital model boundaries
From transit to triumph: digital model boundaries. Published 2025-02-17; accessed November 4, 2026.
[22] OPTANO Modeling development practices
OPTANO Modeling development practices. Published 2026-03-12 (technology index); accessed November 4, 2026.
[23] OPTANO Modeling change log
OPTANO Modeling change log. Published 2025-11-21 (latest listed release); accessed November 4, 2026.
[24] Supply chain optimization offering
Supply chain optimization offering. Published undated; accessed November 4, 2026.
[25] OPTANO solver technology claims
OPTANO solver technology claims. Published 2024-04-05 (technology index); accessed November 4, 2026.