Go back to Market Research
OMP develops Unison Planning, a supply chain planning suite for manufacturers in chemicals, life sciences, consumer goods, metals, and other process industries. The software combines demand forecasting, inventory policies, sourcing and production planning, detailed scheduling, and order promising. Customers configure planning models and solvers, connect them to their operational systems, and use the results to prepare and execute supply plans. (1, 5)
Overall supply chain score: 4.8/10
| Dimension | Score |
|---|---|
| Supply chain depth | 5.6/10 |
| Decision and optimization substance | 5.0/10 |
| Product and architecture integrity | 4.6/10 |
| Technical transparency | 4.0/10 |
| Vendor seriousness | 4.8/10 |
OMP’s main limitation is the distance between a coordinated planning suite and a continuously improved economic decision system. Its documented operating model combines mathematical programming, specialized heuristics, inventory policies, and planner judgment. Costs enter the calculations, but so do service targets, safety-stock penalties, and manually chosen priorities. Passing decisions between planning layers can enforce consistency while preserving economically damaging assumptions about what each layer is allowed to change.
The current inventory offering explicitly claims stochastic treatment of demand and supply, simultaneous optimization of safety stock and lot sizes, and constraints shared with production. Those claims deserve assessment at their actual scope: inventory policy selection. Public material does not establish how joint uncertainties, conditional receipt dates, production sequencing, and later corrective actions enter a common monetary objective. Running more scenarios or explaining an optimizer’s answer does not resolve that modeling problem. (2)
The November 2026 UnisonIQ release adds agent configuration and natural-language orchestration around the existing engines. Its Bekaert example describes planners comparing S&OP scenarios. That is a narrower achievement than eliminating the work required to operate and improve a supply chain decision process. OMP’s implementation and support material documents substantial work around customer models, solvers, integrations, and upgrades; that work belongs in the economic assessment. (4, 23, 24)
OMP versus Lokad
OMP organizes planning around a configurable industrial model, several decision horizons, and specialist solvers. Lokad organizes its offering around recurring economic decisions: purchases, production quantities, allocations, and other actions ranked by expected financial consequences. Its inventory optimization documentation describes probabilistic demand and lead times, customer-specific monetary trade-offs, and Supply Chain Scientists who maintain the executable logic.
Consider a chemical producer deciding whether to extend a production campaign. A longer run saves a setup but consumes shared capacity, creates additional co-products, and leaves less room for another product. Some material may arrive late; some output may expire before it sells. OMP documents campaign planning, shelf-life constraints, and multiple ways of combining solvers. Its current inventory page also asserts stochastic supply modeling. The unresolved issue is how these elements interact when choosing the campaign, rather than when calculating an inventory target or inspecting a separate scenario. (1, 2)
Lokad’s proposition is to place these competing uses of resources and uncertain consequences inside the economic calculation, then revise that calculation when operations reveal a mistake. The perspective is developed in Introduction to Supply Chain and Holimization. A comparison should examine the resulting decisions, the interactions actually retained, and the combined labor of customer and supplier. Lokad’s specialist work and cloud consumption must be counted alongside OMP’s configuration and support costs.
For OMP, the practical question is whether a poor recommendation leads to a reusable correction of the underlying economics or to another planner adjustment. Its published services provide a path for consultant-led changes. They do not establish how quickly a customer can revise an objective, replay past decisions, test a replacement, and deploy it safely without accumulating more interdependent rules. (23, 24)
What Unison Planning computes
Coordinated solvers do not establish a common economic optimum
OMP’s expert FAQ describes several arrangements: an MRP-like calculation followed by production optimization and allocation, or an all-in-one solver combined with local repair. S&OP choices can become constraints for operational solvers. The same FAQ says implementations may use actual costs, standard penalties, or a mixture. (1)
This design has an economic consequence. Suppose a tactical calculation fixes sourcing quantities before detailed scheduling exposes an expensive sequence of setups. The downstream solver can optimize within those quantities while missing the cheaper decision to change sourcing. Coordination requires a way for the newly discovered cost to revise the upstream choice. Shared data and feasibility checks alone cannot accomplish that.
OMP’s 2021 e-book identifies LP and MIP in its value-optimization solver and presents tailored, mixed objectives. These methods make the formulation decisive. Linear models suppress discrete economics unless additional structure is introduced; MIP can represent batches, setups, and other discontinuities, but at a potentially large computational price. OMP does not publish a representative formulation showing which interactions survive at industrial scale, or a comparison of realized financial outcomes against simpler decision policies. (5)
The issue extends beyond obtaining an answer quickly. A planner may discover that producing a co-product changes demand for another item, or that a contractual penalty depends on several late orders together. If retaining that interaction makes the model or its decomposition impractical, the system will optimize a knowingly incomplete account of the business. Solver speed and optimality gaps cannot validate the economic assumptions that define the calculation.
Stochastic inventory claims need a decision-level demonstration
The inventory page explicitly recognizes uncertain lead times and describes demand and supply variability as stochastic inputs. It claims joint safety-stock and economic-order-quantity optimization, inventory-value and capacity constraints, and a shared model with production. Its policy machinery also includes ABC segmentation, service levels, and decoupling points. (2)
These are more specific claims than a generic reference to uncertainty. They still leave the decisive mechanics undisclosed: the distributions, dependencies between demand and receipts, treatment of rare disruptions, information available at each decision, and how the optimizer values shortages against cash and waste. No accessible formulation or reproducible example shows how the inventory engine’s uncertainty representation reaches the campaign and scheduling decisions described elsewhere.
A shared raw material illustrates the gap. Two products may each appear adequately buffered when examined separately, yet both fail during the same supplier delay. The decision depends on the common receipt event, the competing margins, and the capacity available after arrival. Increasing two safety stocks is only one possible response. Changing sourcing, production order, or customer commitments may create more value.
OMP’s life-sciences FAQ permits probabilities for tenders in simulation planning. That establishes a specific probabilistic input, but not a calibrated joint distribution for demand, lead times, yields, and the actions taken after new information arrives. (11) Adding scenarios to LP/MIP is also costly: decisions and constraints multiply across futures, while discrete campaign and setup choices remain. The relevant evidence would show that materially richer uncertainty changes decisions profitably at a sustainable computational and maintenance cost.
Manufacturing detail is visible in deployments
The VDM Metals case identifies melting-shop scheduling, subcontractor planning, campaign planning, scrap returns, and capable-to-promise calculations based on material and capacity checks. It connects lower work-in-process and shorter lead times to better visibility of production imbalances. The Alpro case identifies shelf-life-sensitive production and movement of quarantined products while retaining traceability. These are operational mechanisms, rather than customer logos standing in for an explanation. (28, 29)
Their evidential limit is financial attribution. Neither account exposes the complete decision model, an unchanged baseline, or the operating cost of maintaining it. They establish that OMP addresses industrial constraints; they do not establish the economic superiority of its optimizer or the generality of its uncertainty treatment.
Technology history: integration, twins, AI, and agents
OMP’s older technical material reveals considerable continuity. An integrated-planning presentation hosted among 2013 conference materials already described in-memory integration, a common data model, and mathematical solvers balancing weighted costs. The central architecture therefore predates the current language of digital twins and decision intelligence. (6)
| Period | Published proposition | Mechanism and subsequent evidence |
|---|---|---|
| 2013 conference materials | Integrated planning with in-memory technology and cost optimization | Shared model and weighted solver objectives; the foundation of later platform claims. |
| November 2020 | Telescopic digital twins, explainable AI, resilience, and Data Genie | OMP described linked models for different planning scopes and machine learning to update production rates. |
| October 2021 | Autonomous planning, cognitive features, deep learning, and value optimization | The e-book paired broad AI language with specialist planning features, in-memory execution, and LP/MIP. |
| July–November 2022 | Machine learning forecasting and a more accurate digital twin | Forecasting commentary and Data Genie provided identifiable, narrower mechanisms. |
| 2025–March 2026 | Decision-centric planning and always-on agents | Evonik and Bekaert examples emphasized scenario analysis; the March release combined agents with human validation. |
| November 2026 | Agent Library, Agent Studio, and Unison Companion orchestration | Named early adoption at Bekaert involved natural-language S&OP scenarios; the numerical engines remained Unison Planning’s existing engines. |
Sources: 6, 7, 5, 8, 9, 12, 13, 14, 4.
The 2020 explanation usefully acknowledged that its fashionable terms did not identify particular algorithms. It described the telescopic twin as several interlinked planning models and Data Genie as learning production parameters from execution data. By 2021, however, the sales narrative bundled cognitive features and autonomous planning without comparable evidence of broad intelligence or labor eliminated. The documented mechanisms were narrower than the umbrella language. (7, 5)
The 2022 forecasting article cited M4 and M5 as evidence for machine learning, while acknowledging limits in available company data. Competition results support investigating specific forecasting methods; they do not validate OMP’s implementation or establish economic returns from its forecasts. The historical test is whether later releases show more reliable decisions and less maintenance, not whether each new vocabulary has a corresponding announcement. (8)
There is also continuity in customer evidence. Bekaert’s September 2025 case already described advanced S&OP and scenario planning for its energy-transition business. The November 2026 release adds a natural-language way to run that work. Its incremental value must therefore be measured against the scenario capability already in use, including the remaining validation and model-maintenance effort. (13, 4)
Solvay’s March 2026 case is a useful check on broad AI positioning. It reports selection in September 2024 and a first business-unit go-live in November 2025; statistical forecasting matched or exceeded the previous process for more than half the tested product combinations. The account places further AI adoption ahead. That is a bounded implementation milestone, not evidence that autonomous planning already governs the network. (15)
OMP’s roadmap contains identifiable engineering work, particularly industrial solvers and master-data learning. The recurring weakness is that increasingly expansive autonomy language is not accompanied by a measured account of the work eliminated or the economic decisions improved. This gap lowers the roadmap-integrity assessment.
Digital-twin accuracy and the limits of self-correction
Data Genie is relevant to a problem often neglected by planning vendors: production rates, routings, and other master data can be wrong. OMP describes predictive models for planning parameters, ongoing accuracy monitoring, and updates from execution history. When fit deteriorates, the workflow may request additional business information and involve OMP’s data scientists. (10)
This is a mechanism for maintaining part of the model, not merely a promise to store cleaner data. It also does not validate every intervention simulated by the twin. A production rate estimated under ordinary utilization may fail during congestion; historical observations may mix different product combinations or operating regimes. A scenario that changes those conditions needs evidence about the resulting forecast error and its effect on purchases, schedules, and delivery promises.
OMP’s November 2022 podcast explicitly discussed twin accuracy. The missing demonstration is predictive validation across consequential interventions, including uncertain supply and dependencies between plants. Improving an estimated rate does not establish that a recommendation based on the entire network model will improve financial results. (9)
The same boundary applies to reformulation. Data Genie can suggest better parameters and routings. A newly discovered shortage cost, substitution effect, or available procurement action changes the decision problem itself. OMP’s public account does not show an inexpensive, repeatable path from that discovery to revised logic, historical replay, and controlled deployment.
AI agents and the productivity claim
UnisonIQ separates generative language handling from numerical calculations. Its current page says the generative component does not calculate or decide alone; optimization engines check proposed recommendations. Organizations define execution thresholds and escalation rules. These controls can prevent some inappropriate actions, but economic adequacy still depends on the model the engines evaluate. A feasible recommendation can destroy value. (3)
The November release describes agents preparing changes for approval and customers configuring agents in Agent Studio. Approval can be a chosen governance rule. The deeper limitation is the demonstrated workload: Bekaert’s example remains planners asking for scenarios, examining results, and discussing choices. There is no reported denominator for staff hours, portfolio size, decision frequency, or effort supplied by OMP and partners. (4)
Evonik’s earlier case likewise described multiple supply scenarios per day; the March 2026 announcement promoted hundreds of scenarios. More evaluations may improve a difficult capital or sourcing decision. They may also create more comparisons for planners to maintain and interpret. Scenario count does not measure productivity. The useful measure is total effort per economically sound, executed decision at comparable scope and quality. (12, 14)
Agent Studio also creates another layer of maintained customer logic: instructions, tool permissions, escalation conditions, and tests. Its value depends partly on whether agents can help improve reusable models and decision rules. Reproducing a planner’s sequence of queries and approvals faster leaves much of the original operating model intact.
Software bill of materials
OMP’s public engineering material identifies a C++ core and a growing set of services around it. The table identifies documented components and the scope of their use.
| Component | Public evidence and scope | Ownership-cost implication |
|---|---|---|
| C++, Boost, Qt | OMP’s February 2020-hosted BeCPP presentation, slide 7; current C++ engineering recruitment supports continuity of the core language. | Native application logic and UI/framework dependencies require coordinated builds and upgrades. |
| React, C#, ASP.NET Core | Same historical stack slide, assigned to front-end and full-stack development. Current use of every historical component is unconfirmed. | Web services and interfaces add deployment and compatibility work around the numerical core. |
| Docker, Kubernetes, CI/CD tooling | Historical stack slide and current engineering roles. | Container scheduling, image maintenance, deployment control, and observability remain operating responsibilities. |
| Microsoft Azure | Current cloud offering. | Hosting transfers infrastructure operation to OMP; compute, memory, storage, and recovery still have costs. |
| SQL Server, Oracle, Azure SQL | Employer database-administrator postings discuss internal systems and customer-hosted database experience; an older posting refers to more than 150 customer instances. | OMP supports a heterogeneous database estate. The postings do not establish that every customer deployment runs both database engines. |
| Python SDKs, event streams, data projections | Current Product-team recruitment describes an extension layer under development, with services exposing platform data to extensions. | Additional copies, contracts, and event handling must preserve data semantics and access boundaries. |
Sources: 17, 18, 19, 20, 21, 22.
The evidence supports an established native core with cloud, web, database, and extension infrastructure around it. Planning data has to remain meaningful across the core, external systems, projected data, and event-driven extensions, while customer-specific logic survives product upgrades.
OMP’s cloud page advertises low TCO and no customer infrastructure footprint. That changes who operates the infrastructure, not how much total work it requires. Its support offering explicitly covers database and integration health, operating-system and database compatibility, patches, releases, and upgrade assistance. Those obligations have to be financed by subscription and services revenue. (22, 24)
Security claims do not reveal the containment architecture
The cloud page leads with service-auditor validation, ISO alignment, and failover options. These do not explain tenant isolation, privilege boundaries, exposure of production data, recovery after malicious changes, or how an extension is prevented from reaching unrelated systems. Failover addresses availability; it does not establish recovery from corrupted or deliberately altered state. (22)
The Python role describes an extension interface with data access, event publishing, and SDK contracts. This development interface creates a security-critical boundary: a compromised dependency or overprivileged service can affect both data and decisions. Public material does not expose enforced network, secret, resource, or data-access restrictions for it. (19)
OMP’s advertised agent guardrails similarly need an enforcement mechanism outside the agent’s generated instructions. Rules about what an agent should do are not sufficient against prompt injection, compromised tools, or erroneous configuration. The public product page explains decision rights but not the architecture that preserves them after one component is compromised. (3) The security score therefore remains low on inspectable engineering evidence.
Configuration maintenance, ownership cost, and accountability
OMP describes implementation as readiness assessment, design, configuration, tests, a dry run, cutover, and hypercare. Its continuing service includes solver tuning and periodic adaptation to business changes. This is maintained decision software, even where the customer interacts through planning screens rather than source code. (23)
The recurring bill includes customer specialists who understand the model, vendor or partner consultants who revise it, regression tests after changes, and planners who operate exceptions. VDM Metals describes an internal competence center and data-management program. Bayer’s 2026 migration announcement describes rebuilding processes, master data, and working practices, not simply replacing a software version. These responsibilities make the availability of knowledgeable people part of the production architecture. (28, 16)
Unison Express addresses some initial burden through predefined planning cycles, test scripts, training, and an explicitly fixed scope. That is a concrete attempt to reduce design work and bound delivery. The economic trade-off appears when a customer’s valuable decision requires something outside the template. A quick implementation is beneficial only if the resulting model captures enough of the business and can be revised at an acceptable lifetime cost. (25)
Arxada shows why scope and attribution matter
EY’s Arxada account reports volume planning and detailed scheduling live within four months, value-stream visibility at six months, and a first-year benefit of two to three times that year’s implementation investment, attributed to Arxada’s supply-chain executive. It also describes continued EY monitoring and issue resolution. The account presents reductions of 80% in service failures, 90% in backorders, 60% in obsolescence risk, and 40% in inventory as outcomes the program is on track to deliver. Those percentages are not all reported as completed measurements. (26)
OMP’s summary says the rollout spanned 24 plants. EY identifies 24 plants as Arxada’s corporate footprint and describes global deployment as continuing. The underlying account does not substantiate the broader completed-rollout reading. This matters when a four-month milestone is presented as evidence of implementation speed at global scale. (27, 26)
The case does not publish the baseline, absolute benefits, treatment of working-capital release versus recurring earnings, or complete customer and partner costs. Benefits from process redesign and implementation services also cannot be attributed solely to an optimization algorithm.
OMP commits publicly to implementation assistance and continued support. Its public economic claims do not define a general failure threshold, bounded customer obligations, or responsibility for recovering an underperforming investment. The resulting accountability gap is concrete: a buyer cannot infer from the published success stories who will diagnose a bad recommendation, change its underlying economics, and pay for that work when returns disappoint.
Detailed supply chain score
The five dimensions contain five equally weighted criteria each. The scores are editorial assessments of the documented offering, not measured performance results.
Supply chain depth
| Criterion | Score |
|---|---|
| Economic framing | 6/10 |
| Decision end-state | 7/10 |
| Conceptual sharpness on supply chain | 6/10 |
| Freedom from obsolete doctrinal centerpieces | 4/10 |
| Robustness against KPI theater | 5/10 |
Average: 5.6/10.
Economic framing — 6/10. Monetary costs enter OMP’s solvers, and its inventory offering discusses working capital and obsolescence. Penalties and targets can nevertheless substitute for the financial consequences of actions. The disclosed methods do not establish a common expected-value calculation across procurement, inventory, and production. (1, 2)
Decision end-state — 7/10. Scheduling and capable-to-promise deployments reach operational choices about resources and commitments. OMP earns credit for those outputs. Scenario comparison and proposed changes still occupy a substantial part of the published agent workflow, limiting the demonstrated extent of unattended execution. (28, 4)
Conceptual sharpness on supply chain — 6/10. Campaigns, co-products, shelf life, subcontracting, and quarantine are materially relevant constraints. Their representation is clearer than the treatment of coupled uncertainty and later corrective decisions. Industrial detail cannot compensate for an economically incomplete model of the future. (1, 11, 29)
Freedom from obsolete doctrinal centerpieces — 4/10. Planning horizons, inventory classifications, service targets, and fixed policy parameters retain prominent roles. They can organize work while restricting the actions the optimizer considers. The evidence does not show these conventions being systematically displaced by direct evaluation of marginal economic returns. (2, 5)
Robustness against KPI theater — 5/10. The Arxada case includes a named financial-return claim, while Solvay reports a bounded comparison with its previous forecasts. Neither establishes full net returns across the offering. Scenario counts and projected percentage improvements remain weak substitutes for completed economic measurements. (15, 26)
Decision and optimization substance
| Criterion | Score |
|---|---|
| Probabilistic modeling depth | 4/10 |
| Distinctive optimization or ML substance | 6/10 |
| Real-world constraint handling | 7/10 |
| Decision automation and labor productivity | 4/10 |
| Resilience under real operational complexity | 4/10 |
Average: 5.0/10.
Probabilistic modeling depth — 4/10. OMP explicitly claims stochastic demand and supply within MEIO and supports tender probabilities. Public methods do not reveal joint distributions, calibration, dependencies, or how uncertainty affects production choices beyond inventory targets. This prevents treating suite-wide stochastic economic optimization as established. (2, 11)
Distinctive optimization or ML substance — 6/10. Specialized industrial solvers and Data Genie’s parameter learning identify computational work beyond generic AI branding. Their economic advantage is difficult to inspect: public material lacks executable formulations, reproducible comparisons, and evidence that richer objectives remain tractable as models evolve. (1, 10)
Real-world constraint handling — 7/10. The VDM and Alpro accounts connect planning to specific production and distribution restrictions. This supports substantial industrial modeling depth. It does not establish that every economically material interaction is preserved through aggregation, solver handoffs, and decomposition. (28, 29)
Decision automation and labor productivity — 4/10. Agents invoke numerical engines and can prepare executable changes; they are not just document search. The published deployments provide no complete labor comparison covering customer, OMP, and partners. Faster scenario preparation cannot establish elimination of recurring planning and maintenance work. (3, 4, 26)
Resilience under real operational complexity — 4/10. Data Genie monitors parts of the model and services support subsequent changes. The missing evidence concerns reformulation: retaining a newly discovered economic interaction, testing its consequences, and deploying revised logic without prohibitive solver or human costs. Parameter updates alone do not resolve this limitation. (10, 23)
Product and architecture integrity
| Criterion | Score |
|---|---|
| Architectural coherence | 6/10 |
| System-boundary clarity | 6/10 |
| Security seriousness | 2/10 |
| Software parsimony and ownership cost | 4/10 |
| Compatibility with programmatic and agent-assisted operations | 5/10 |
Average: 4.6/10.
Architectural coherence — 6/10. A shared planning foundation and identifiable native core give OMP a coherent product structure. Different planning scopes still use distinct calculations, and the emerging extension layer adds data projections and event contracts. Coherence of data must not be mistaken for joint economic optimization. (7, 17, 19)
System-boundary clarity — 6/10. OMP describes bidirectional ERP integration, planning responsibilities, application management, and support. The remaining opacity concerns semantic ownership across transformations and how the customer’s planning logic transfers when interfaces or product versions change. Those boundaries determine who diagnoses a wrong decision. (10, 22, 24)
Security seriousness — 2/10. Certifications, failover, and declared agent rules do not expose containment after a service, user, or dependency is compromised. Public evidence does not establish enforced extension privileges, tamper-resistant state, or tested recovery from malicious changes. (3, 19, 22)
Software parsimony and ownership cost — 4/10. OMP’s materials reveal a substantial supportable estate and customer-specific configuration. Express limits initial design effort, but specialist maintenance and upgrades remain part of the operating model. Public cases do not disclose enough continuing labor and infrastructure cost to substantiate the low-TCO proposition. (17, 23, 24, 25)
Compatibility with programmatic and agent-assisted operations — 5/10. Integration machinery, Python extension work, and Agent Studio provide routes beyond manual screens. Public interfaces do not yet establish customer control over the numerical formulation, independent replay, or safe automated promotion of revised decision logic. (10, 19, 4)
Technical transparency
| Criterion | Score |
|---|---|
| Public technical documentation | 4/10 |
| Inspectability without vendor mediation | 3/10 |
| Portability and lock-in visibility | 3/10 |
| Implementation-method transparency | 6/10 |
| Evidence density behind technical claims | 4/10 |
Average: 4.0/10.
Public technical documentation — 4/10. The expert answers, engineering presentation, and integration descriptions expose useful mechanisms. They stop well before a developer manual for objectives, solvers, uncertainty models, and extension security. A prospective customer must rely heavily on vendor-mediated clarification. (1, 10, 17)
Inspectability without vendor mediation — 3/10. No reproducible public example connects raw inputs to an economically evaluated production or replenishment decision. Explanations generated inside UnisonIQ may help users interpret results, but they do not make the underlying engine independently inspectable. (3, 5)
Portability and lock-in visibility — 3/10. Data exchange is documented more clearly than portability of customer rules, solver configurations, and agent behavior. OMP-specific model knowledge and upgrade services therefore represent a material dependency whose replacement cost is not publicly characterized. (10, 24)
Implementation-method transparency — 6/10. OMP publishes the stages of configuration, testing, dry runs, support, and continuing adaptation. Express makes scope boundaries more explicit. Public descriptions still omit representative maintenance effort, change lead times, and costs after the initial deployment. (23, 25)
Evidence density behind technical claims — 4/10. Named cases identify operational functions and some outcomes. Much broader claims about stochastic optimization, always-executable recommendations, and autonomy lack disclosed tests. The difference between OMP’s Arxada summary and the underlying rollout account further limits confidence in headline scope. (2, 26, 27)
Vendor seriousness
| Criterion | Score |
|---|---|
| Technical seriousness of public communication | 5/10 |
| Technology roadmap integrity | 5/10 |
| Conceptual sharpness | 5/10 |
| Economic accountability | 4/10 |
| Defensibility in an agentic-software world | 5/10 |
Average: 4.8/10.
Technical seriousness of public communication — 5/10. OMP sometimes explains actual mechanisms, including the distinction between language handling and numerical engines. Absolute claims about feasibility and expansive AI terminology go substantially further than the disclosed validation. The technical explanation is uneven across the portfolio. (3, 5, 7)
Technology roadmap integrity — 5/10. The chronology shows a continuing planning architecture and identifiable additions such as Data Genie and agent orchestration. It does not show a measured completion of the autonomy proposition promoted in 2021. Recent customer examples largely report narrower forecasting, integration, and scenario milestones. (5, 9, 13, 15)
Conceptual sharpness — 5/10. OMP recognizes stochastic supply, industrial constraints, and model accuracy. Its public language nevertheless blurs an executable plan with a financially sound decision and coordination with comprehensive optimization. Those distinctions determine what its current evidence can support. (2, 3, 9)
Economic accountability — 4/10. Arxada supplies a specific customer-attributed return claim, and implementation services assign some practical responsibilities. The missing baseline, complete cost scope, and defined remedy for disappointing returns prevent the public proposition from being economically testable across customers. (23, 26)
Defensibility in an agentic-software world — 5/10. Industrial planning logic and execution knowledge are more defensible than a conversational interface. Agent Studio’s value depends on reducing the full cost of operating and revising that logic. Published evidence currently supports easier access to existing planning capabilities more strongly than a transformed productivity level. (1, 4, 19)
Source dossier
The sources below were checked on November 4, 2026. Publication dates are stated where available; current product pages and recruitment material describe the versions accessible on that date.
[1] Ask our innovation experts
OMP source. undated. Explains solver composition, costs and penalties, and planning-level handoffs.
[2] Inventory optimization software
OMP source. undated; current November 2026. Current claims about stochastic inputs, inventory policies, lot sizes, and production constraints.
[3] UnisonIQ: Supply Chain AI Agents and Companion
OMP source. undated; current November 2026. Distinguishes generative AI, numerical engines, decision rights, and agent configuration.
[4] OMP releases the UnisonIQ Agent Library and Agent Studio
OMP source. 2026-11-03. Current live release dated November 3; names Bekaert and describes the new agent workflow.
OMP source. 2021-10. Pages 12–13 and the value section connect autonomous-planning language to in-memory execution and LP/MIP.
[6] An integrated approach to supply chain planning
OMP source. 2013 event materials; presentation undated. Older OMP-authored conference presentation describing common-model integration and weighted-cost optimization.
[7] Supply chain planning language for robots (and humans)
OMP source. 2020-11-09. Defines linked digital twins and describes Data Genie updating production parameters.
[8] 5 reasons why machine learning has taken the lead in forecasting (with a caveat)
OMP source. 2022-07-06. OMP forecasting argument based on M4/M5, including its qualification about business data.
[9] Maximizing the accuracy of your digital twin
OMP source. 2022-11-30. Documents OMP’s explicit concern with digital-twin master-data accuracy.
[10] Data Management and Integration
OMP source. undated. Explains ERP interfaces, Data Genie monitoring, and model-maintenance responsibilities.
[11] Ask our life sciences experts
OMP source. undated. Details manufacturing constraints and probabilities for tender simulations.
[12] How Evonik Oxeno boosts supply performance with decision-centric planning
OMP source. 2025-01-29. Documents daily supply-scenario evaluation at Evonik Oxeno.
[13] Bekaert drives agility with advanced scenario planning
OMP source. 2025-09-25. Establishes Bekaert’s scenario-planning use before the latest agent release.
[14] OMP unveils decision-centric planning
OMP source. 2026-03-03. Promotes decision-centric planning through agents, scenarios, and human validation.
[15] How Solvay accelerates chemical supply chain planning with OMP
OMP source. 2026-03-24. Dates selection and first-unit deployment; reports the scope of the statistical-forecast comparison.
[16] Bayer to present how it rebuilt global planning with OMP
OMP source. 2026-09-17. Describes the process and master-data work behind Bayer’s planning rebuild.
[17] OMP: Making your day, BeCPP presentation
OMP source. 2020-02 hosting; presentation undated. Slide 7 is the OMP technology-stack diagram; the hosting path dates it to February 2020.
[18] Lead C++ software engineer, OMP
OMP source. undated; accessed 2026-11-04. Employer-authored role identifies C++ engineering, cloud work, testing, and observability.
[19] Software Engineer (Python and Data Platform), OMP
OMP source. undated; accessed 2026-11-04. Employer-authored role describes Python SDKs, data projections, event streams, and an evolving extension layer.
[20] SQL Database Administrator, OMP employer posting
OMP source. 2026; exact publication date unspecified. Employer-authored database role; distinguishes internal database duties from customer-hosted experience.
[21] Database Administrator, OMP employer posting
OMP source. circa 2023; listing says three years ago. Earlier employer-authored role mentions more than 150 customer instances and database operations.
[22] Cloud: access your planning solution in OMP Cloud
OMP source. undated. Azure hosting, low-TCO proposition, and the publicly stated security and availability assurances.
OMP source. undated. Configuration, tests, dry runs, continuing solver tuning, and business-change services.
OMP source. undated. Database health, patches, compatibility, upgrades, and customer-specific application knowledge.
OMP source. undated; current November 2026. Preconfigured planning, test assets, and fixed implementation scope.
[26] How Arxada’s chemical supply chain planning transformation paid off
EY source. undated. Implementer’s Arxada account: delivery timing, attributed financial return, continuing support, projected outcomes, and rollout scope.
[27] Fast-tracking value: Arxada
OMP source. undated. OMP’s shorter Arxada account, including the stronger statement about 24 plants.
[28] Supply chain transparency at VDM Metals
OMP source. undated; current November 2026. Specific scheduling, capable-to-promise, and organizational capabilities at VDM Metals.
[29] Alpro’s journey toward supply chain excellence
OMP source. 2020-05-24. Alpro’s production, shelf-life, quarantine, and tactical-to-operational planning account.