Go back to Market Research
SkyPlanner APS is cloud production-scheduling software from Finland’s Skycode Oy. Manufacturers use it to assign operations to workstations, sequence jobs, account for material availability and working shifts, and send work queues to the shop floor. Its Arcturus engine recalculates schedules from the factory’s orders, production model and priorities; ERP integration and time recording connect planning with execution. 1
Overall supply chain score: 5.5/10
| Dimension | Score |
|---|---|
| Supply chain depth | 5.6/10 |
| Decision and optimization substance | 5.0/10 |
| Product and architecture integrity | 6.0/10 |
| Technical transparency | 5.8/10 |
| Vendor seriousness | 5.0/10 |
SkyPlanner has a credible purpose: eliminate the manual arrangement of interdependent production jobs. Its public manuals explain actual scheduling choices, and named customers describe work removed from factory planning. The weakness is the criterion governing those choices. Users balance delivery, lead time, customer importance and setup preferences; the documentation does not establish an objective that prices the economic consequences of each alternative under uncertainty. A feasible sequence can therefore remain financially inferior even when it follows every configured preference.
The newer agent interface adds useful mechanisms for changing and testing the model. It also extends the reach of the same priority-driven engine. The important unresolved questions are how reliably those priorities represent money, how uncertain supply and processing times change the selected schedule, and who maintains that interpretation as the factory changes.
SkyPlanner versus Lokad
SkyPlanner starts with a factory model and produces a detailed sequence of work. Lokad starts with an economic decision problem: which purchases, allocations, production quantities or other actions offer the best expected return, given uncertain futures and shared constraints. The distinction matters when a late order, an extra setup and idle capacity have very different monetary consequences.
SkyPlanner’s public scheduling controls express preferences through relative strengths. A planner can favor particular customers or compress production lead times, but must translate the commercial situation into those preferences. Lokad’s quantitative supply chain approach makes that economic translation part of the numerical model, with probabilistic forecasts informing the decision calculation. A change in the cost of lateness should alter the decision through its economic effect, without relying on someone to discover a suitable slider position. 2
The maintenance obligations also differ. SkyPlanner supplies configuration, integration and custom-development services; its customer must define the operating model and decide when it needs revision. Lokad’s Supply Chain Scientists build and maintain the data interpretation and decision logic as part of its inventory optimization service. Neither approach escapes engineering work. The relevant comparison is how much recurring effort produces a tested improvement in decisions, as described in Holimization. 8
For a manufacturer replacing manual machine loading, SkyPlanner addresses a concrete and relatively bounded problem. A broader economic optimization project must also decide what to produce, purchase or defer when demand, deliveries and yields are uncertain. The published SkyPlanner mechanisms do not establish that broader capability.
What Arcturus actually decides
Priorities determine the compromise
The dynamic-priorities manual distinguishes two objectives that are easily confused. Sorting work by due date can make other jobs late; maximizing the number completed on time can instead delay one order to protect several others. The same control panel includes lead-time compression, customer and job importance, start eligibility and grouping similar jobs to reduce setups. 2
This is meaningful scheduling logic. It is also an incomplete economic model. Counting late jobs treats a minor inconvenience and a stopped customer production line as equivalent unless someone introduces a compensating priority. Reducing setup time may justify overtime, larger batches or earlier production in one situation and destroy value in another. Relative priorities conceal the exchange rate between these consequences.
The vendor does not publish Arcturus’s mathematical formulation, search method or reproducible performance benchmark. Its claims of rapid calculation therefore establish neither an exact optimum nor a particular solver family. The decisive missing result is a comparison showing that its selected schedules outperform plausible alternatives economically, with the same operating constraints.
Materials and capacities are modeled, uncertainty remains outside the calculation
The material-availability option incorporates warehouse stock and expected receipts into scheduling. Its documentation goes further and says this prevents material-shortage disruptions. An expected arrival date cannot provide that guarantee: the supplier may deliver late, short or with unusable material. 3
Workstation efficiencies are configured as percentages that modify relative processing speeds. The manual requires already-planned jobs to be unscheduled before these efficiencies are changed. These are explicit production parameters, rather than a published probability model of processing time. 4
The important stochastic question is joint: a late component, a failed operation and an urgent order can compete for the same remaining capacity. Replanning after each event helps recovery, but does not determine which precaution was worth taking beforehand. The reviewed documentation supplies no joint distributions for demand, supplier lead times, yields and processing durations, or a decision procedure valuing their combined consequences.
Constraint handling is more substantial. SkyPlanner models operation dependencies, alternative workstations, tools, shifts, material availability and setup-related properties. These mechanisms can prevent obvious scheduling impossibilities and coordinate linked stages. Their value depends on the accuracy of the production model and the priorities resolving conflicts. 1
Automation, agents and the work left behind
Automatic recalculation predates the current agent language. The April 2025 manual documents a selectable rescheduling interval. Repeated schedule generation therefore need not require repeated planner clicks. 5
The current MCP interface lets an external assistant inspect and modify production data and invoke Arcturus. It can also help construct a model from spreadsheets and develop an integration. Each cloud installation has a sandbox copied from production; selected scheduling settings and company-wide rules can be carried back after a trial. This creates a practical route from a proposed configuration change to a tested schedule. 6
Conversational access alone adds little intellectual substance to the scheduling engine. Asking an assistant to raise a priority reproduces the planner’s existing intervention. The stronger opportunity is to diagnose a recurring failure, correct the model once and remove the need for that intervention. SkyPlanner’s copied sandbox supports that work, but the public material does not demonstrate a complete correction cycle with historical replay, economic acceptance criteria, versioned release and rollback.
The distinction also changes the productivity benchmark. A faster answer to a planner’s question is useful; a durable rule correction can remove thousands of future questions. MCP should be judged by the latter opportunity and by the total work remaining in planning, integration and model maintenance.
Customer evidence identifies actual work removed
Eino Korhonen Oy describes a plant with 26 injection-molding machines and about 500 molds. Its former preparation of the next day’s schedule took three to four hours daily. The case reports roughly one person’s annual workload released through automatic loading, with worklists specifying the next mold and machine. Family molds, which produce different items in the same run, were among the requirements tested during selection. 11
KJH-Comp reports delivery reliability increasing from around 50% to as high as 98% in 2026, and scheduling work falling from two supervisors to one. Its implementation included purchased-part readiness checks, skill matching and a C9000 ERP connection. These are identifiable changes to production work, beyond a more convenient planning display. 12
Both accounts concern manufacturers moving away from substantial manual coordination. They support labor-saving potential in that setting. Neither publishes a net-return calculation separating software, integration, customer development effort and the other operating changes. The 98% figure measures delivery adherence, not forecast accuracy; its significance depends on the commitments and production conditions against which it was achieved.
Technology history and roadmap integrity
| Period | Public position | Delivery evidence and remaining question |
|---|---|---|
| 2022 | AI scheduling, rapid recomputation and prediction of problem scenarios | An information package described a production model fed by ERP data and governed by user priorities. The predictive language lacked a documented probabilistic mechanism. |
| 2023–2024 | Factory deployment | KJH-Comp ran the product alongside its existing process for several months and continued joint feature development after full adoption. |
| 2025 | Exceptional AI capability and rapid deployment | A brochure claimed market-leading AI strength without a comparative benchmark; manuals documented specific rules, material constraints and timed recalculation. |
| 2026 | Agentic scheduling and MCP | Tool access and sandbox-based configuration accompany a more precise published explanation of what ordinary APS automation does and does not establish. |
The 2022 package already used AI to describe sequencing production work and advertised anticipation of problem scenarios. It supplied no learning method, prediction evaluation or economic benchmark. This early AI language was broader than the technical evidence accompanying it. The 2025 brochure escalated the rhetoric to the strongest AI on the market, again without a published comparison. 16 17
There is also a substantive improvement in the communication. Jussi Mäntylä’s article, updated October 10, 2026, explicitly separates timed rescheduling from broader agent workflows, future capacity derived from known orders from forecasting, and reacting to a received update from predicting a disruption. It identifies illustrative scenarios as such and asks for before-and-after measurements over comparable workloads, including software and integration costs. 18
The current account is more technically useful. The remaining ambition should be assessed against delivered decision quality and engineering productivity. Connecting a language model does not resolve the missing economic objective or turn expected receipt dates into probabilistic supply forecasts.
Software bill of materials and integration integrity
The available materials identify a relatively focused application and its interfaces. They do not disclose the production database, application runtime, scheduling implementation or complete dependency inventory. Public recruitment material located for the Finnish company did not resolve those gaps.
| Scope | Documented materials | Operational consequence |
|---|---|---|
| SkyPlanner core | Proprietary Arcturus scheduler, browser application, configured production data | Customers can change the operating model but cannot inspect or replace the scheduling method through the documented controls. |
| Hosting described in the 2025 brochure | Amazon Web Services | Infrastructure operation remains part of the vendor’s delivery cost and responsibility. |
| Current residency policy | Standard EU/Germany hosting for customers requiring EU residency; other agreed deployments possible | Actual location and providers must be established for the contracted deployment. |
| ERP and shop-floor integration | Versioned REST API, tenant context, order/process-step/time-log resources | Identity mapping, retries, state ownership and reconciliation are engineering obligations. |
| Optional agent access | MCP tools used by an external assistant | Permissions and data disclosures cross another application boundary. |
The API V4 guide, updated October 28, 2026, exposes the integration’s important failure modes. External references do not guarantee idempotent creation; timed-out writes need reconciliation before retrying. Paginated reads are not snapshots of a changing dataset, and a timestamp on an order cannot detect every change in its children or time logs. 7
These details matter economically. A duplicate order can consume apparent capacity twice. A missed correction can leave the ERP and scheduler disagreeing about completed work. A connector that transports JSON successfully may still produce the wrong factory plan. The guide’s explicit treatment of these problems is a strength, while the reconciliation code remains part of the installed solution’s cost.
Arcturus’s computational footprint remains opaque. Published speed claims lack the workload, constraint density, hardware and decision-quality measurements needed to establish how calculation costs grow. The architectural uncertainty is concentrated in the proprietary core and the quality of its surrounding integrations.
Configuration maintenance and ownership cost
The current subscription starts at €199 monthly for five workstations, with €20 for each additional workstation and unlimited users. A workstation is a scheduled resource, which can include a person or subcontractor. The commercial meter is thus broader than a count of physical machines. Integration and professional services are additional items. 9
SkyPlanner advertises over 1,000 configurable settings and offers separately scoped custom work where configuration is insufficient. Its service process calls for written assumptions, an estimate and validation against real production examples. Future requirements are scoped again. This is a concrete engineering service, with a recurring cost when the factory’s model changes. 8
The onboarding packages also define different boundaries. The eight-hour option covers sample data, configuration and training. The five-day package includes understanding the other system’s API and estimating the integration; a finished production integration is not established by the package’s duration. 10
KJH-Comp provides a useful view of the elapsed work: parallel operation ran from September 2023 until the following April, with active customer prioritization of features and continued development afterward. 12 That history is more informative for a complex installation than a generic promise of quick setup.
The continuing expense is not limited to purchasing expert hours. Someone must detect when a duration estimate has drifted, determine whether a recurring delay reflects bad data or the wrong objective, and test the correction against other orders. A thousand settings create expressive capacity and a large space of interactions. Sandboxed experiments help; repeatable tests and inspectable change history determine whether those experiments accumulate into maintainable logic.
Security boundaries and economic responsibility
The user-role manual separates shop-floor time recording from administrative functions and reserves scheduling-rule changes for the highest role. API V4 requires an API key associated with the specified tenant. MCP adds per-user read or write permissions separately for production and sandbox, with logged calls. These are concrete access boundaries. 15 7 6
The unresolved issue is damage after an authorized agent or credential is compromised. The public material does not explain independently enforced limits on the scale of production mutations, approval of sensitive changes or recovery from an incorrect batch of writes. A copied sandbox protects an experiment only while the dangerous operation remains outside production. Logging helps investigation; it does not itself stop a destructive change.
The privacy documents also need tighter alignment. The October article states that customer data is never used to train AI models. The June policy allows anonymized or aggregated operational information for product improvement and rules out training for other customers unless separately agreed. The May terms permit anonymized and aggregated data to improve AI models. Those are different promises; permission in a contract does not establish actual training, but the public absolute does not describe the permission retained. 18 14 13
Economic responsibility is clearer. Section 9 of the terms treats AI outputs as advisory, requires the customer’s independent judgment and assigns the resulting decisions and business outcomes to the customer. Liability is capped by subscription fees rather than production value at risk. 13 This places the burden of a bad economic model on the buyer even as the product automates the resulting schedule.
A credible implementation should therefore define failure economically, identify which model defects SkyPlanner must remedy, and measure the customer’s continuing workload. The current article’s measurement advice is a useful starting point. It does not supply a contractual commitment to maintain profitable decision logic.
Supply chain score
Supply chain depth
| Criterion | Score |
|---|---|
| Economic framing | 3/10 |
| Decision end-state | 8/10 |
| Conceptual sharpness on supply chain | 6/10 |
| Freedom from obsolete doctrinal centerpieces | 5/10 |
| Robustness against KPI theater | 6/10 |
Average: 5.6/10.
Economic framing — 3/10. Scheduling preferences influence real trade-offs, but the documented objective does not price lateness, setup, inventory and lost sales together. The monetary meaning of the chosen priorities must be supplied and maintained outside the published calculation. 2
Decision end-state — 8/10. The product produces assignments and work sequences that operators can execute. EKOY’s mold-and-machine worklists provide a concrete account of that connection. The broader procurement and production-quantity decisions remain outside the demonstrated end-state. 11
Conceptual sharpness on supply chain — 6/10. Material readiness and production dependencies are treated as genuine restrictions. Receipt dates nevertheless remain exposed to supplier uncertainty, so a feasible plan is conditional on inputs the factory cannot fully control. 3
Freedom from obsolete doctrinal centerpieces — 5/10. Automatic scheduling removes manual loading, but customer-selected importance and timing priorities still govern the compromise. There is no demonstrated replacement of this preference tuning with a coherent economic decision criterion. 2
Robustness against KPI theater — 6/10. Customer accounts identify operational measures and workloads rather than presenting delivery adherence as forecast accuracy. Missing cost attribution and workload comparability prevent the reported improvements from establishing general net returns. 11 12
Decision and optimization substance
| Criterion | Score |
|---|---|
| Probabilistic modeling depth | 2/10 |
| Distinctive optimization or ML substance | 4/10 |
| Real-world constraint handling | 7/10 |
| Decision automation and labor productivity | 7/10 |
| Resilience under real operational complexity | 5/10 |
Average: 5.0/10.
Probabilistic modeling depth — 2/10. Expected material availability and configured processing efficiencies do not provide a joint forecast of uncertain supply, production and demand. Without that projection, the cost of precautionary capacity or inventory cannot be evaluated consistently across possible futures. 3 4
Distinctive optimization or ML substance — 4/10. Arcturus has a defined scheduling role, but its search method and comparative performance remain undisclosed. The exceptional AI claim in the brochure has no accompanying benchmark establishing algorithmic superiority. 17
Real-world constraint handling — 7/10. Multiple production resources and material dependencies extend the model beyond a simple due-date sort. The published examples support factory scheduling depth, while yield variability and joint procurement choices remain unestablished. 1 11
Decision automation and labor productivity — 7/10. Automatic rescheduling is documented, and customer accounts describe planning labor removed. The public evidence does not measure the full maintenance workforce or isolate additional savings from the newer agent interface. 5 12
Resilience under real operational complexity — 5/10. A sandbox and defined customization process support correction of the production model. A demonstrated sequence from diagnosing a bad economic decision through revised data or objectives, replay, acceptance and safe release is still missing. 6 8
Product and architecture integrity
| Criterion | Score |
|---|---|
| Architectural coherence | 6/10 |
| System-boundary clarity | 7/10 |
| Security seriousness | 5/10 |
| Software parsimony and ownership cost | 5/10 |
| Compatibility with programmatic and agent-assisted operations | 7/10 |
Average: 6.0/10.
Architectural coherence — 6/10. Scheduling and shop-floor progress share a documented application model. The undisclosed database and scheduling internals limit assessment of computational duplication and failure propagation within the service. 1 7
System-boundary clarity — 7/10. The integration guide explicitly allocates field ownership and explains reconciliation requirements. This exposes the work needed to keep the ERP and scheduler consistent, although successful implementation still depends on customer-specific mappings. 7
Security seriousness — 5/10. Distinct roles and environment-specific permissions constrain access. Public descriptions do not establish sufficiently fine limits on compromised production write access or a tested recovery procedure for consequential model changes. 15 6
Software parsimony and ownership cost — 5/10. Subscription pricing is understandable and custom work is scoped. The number of settings and continuing integration responsibilities create maintenance costs that the public cases do not quantify. 9 8
Compatibility with programmatic and agent-assisted operations — 7/10. APIs and agent tools reach operational state and scheduling actions. They enable substantial automation, but safe unattended execution still needs reliable reconciliation and bounded authority. 7 6
Technical transparency
| Criterion | Score |
|---|---|
| Public technical documentation | 7/10 |
| Inspectability without vendor mediation | 5/10 |
| Portability and lock-in visibility | 5/10 |
| Implementation-method transparency | 7/10 |
| Evidence density behind technical claims | 5/10 |
Average: 5.8/10.
Public technical documentation — 7/10. Public manuals explain scheduling behavior, configuration dependencies and integration failure modes. They stop short of the engine’s numerical formulation and a full production dependency inventory. 2 4 7
Inspectability without vendor mediation — 5/10. Users can inspect parameters and compare schedules, but cannot independently inspect the core method. The privacy policy identifies Arcturus as a proprietary scheduling engine rather than a general-purpose language model. 14
Portability and lock-in visibility — 5/10. The terms provide ordinary data export and a post-termination retrieval period. Exporting production records does not establish portability of the configured behavior or a replacement for Arcturus. 13
Implementation-method transparency — 7/10. Onboarding and customization pages specify investigation, scope, estimates and validation. Those steps make the service boundary visible; their cost and repeatability over years of model changes remain customer-specific. 10 8
Evidence density behind technical claims — 5/10. The manuals and named deployments support practical scheduling. Claims of market-leading AI and very fast calculation lack controlled comparisons of decision value at a specified workload. 17
Vendor seriousness
| Criterion | Score |
|---|---|
| Technical seriousness of public communication | 5/10 |
| Technology roadmap integrity | 6/10 |
| Conceptual sharpness | 6/10 |
| Economic accountability | 3/10 |
| Defensibility in an agentic-software world | 5/10 |
Average: 5.0/10.
Technical seriousness of public communication — 5/10. Concrete documentation coexists with exceptional AI claims. The current agentic article is more precise about available functionality, but commercial superiority remains asserted without comparative evidence. 17 18
Technology roadmap integrity — 6/10. The historical scheduling proposition has delivered named factory deployments and increasingly explicit controls. The original predictive claims remain weakly substantiated; the dated clarification of agentic language is a useful correction. 16 18
Conceptual sharpness — 6/10. The vendor now distinguishes known production load, reactive recalculation and forecasting. Its commercial decision model still leaves the exchange between conflicting objectives to planner judgment. 18 2
Economic accountability — 3/10. The contract places business outcomes with the customer and limits the provider’s exposure. Published labor savings do not establish an obligation to diagnose and repair an economically defective model. 13
Defensibility in an agentic-software world — 5/10. Factory-specific constraints, integration knowledge and operating adoption have value. An assistant interface is readily reproducible; durable differentiation depends on better decisions and cheaper model improvement, neither of which has a published comparative demonstration. 8 11
Sources
The numbered sources below are the public documents used for this review, accessed November 4, 2026.
Read the source. Current production scope, resources and scheduling controls.
[2] SkyPlanner — Dynamic priorities
Read the source. June 11, 2025 manual; distinctions between due-date ordering, on-time job count and relative priorities.
[3] SkyPlanner — Consider the materials
Read the source. June 11, 2025 manual; material-stock and expected-receipt constraint.
[4] SkyPlanner — Workstation efficiencies
Read the source. September 9, 2025 manual; relative processing efficiency and rescheduling prerequisite.
[5] SkyPlanner — Automated rescheduling
Read the source. April 8, 2025 manual; interval-based recalculation.
[6] SkyPlanner — MCP server for production scheduling
Read the source. Current agent tools, copied sandbox, selective promotion and access permissions.
[7] SkyPlanner — Build your first integration with API V4
Read the source. October 28, 2026 integration tutorial; API V4 ownership, authentication, reconciliation and update behavior.
[8] SkyPlanner — Customization services
Read the source. Configuration and separately scoped development, including assumptions and validation.
Read the source. Current subscription meter and service boundaries.
[10] SkyPlanner — Assisted onboarding services
Read the source. Eight-hour, five-day and tailored deployment packages.
[11] SkyPlanner — Eino Korhonen customer account
Read the source. Named EKOY account describing machine loading, molds and released planning work.
[12] SkyPlanner — KJH-Comp customer account
Read the source. Named KJH-Comp account describing rollout dates, subsequent development and 2026 operating measures.
[13] SkyPlanner — Terms of Service, version 2.1
Read the source. Version 2.1, May 7, 2026; data rights, liability, service responsibility and exit arrangements.
[14] SkyPlanner — Privacy Policy
Read the source. June 5, 2026; hosting, scheduling engine, data use and optional AI dependencies.
[15] SkyPlanner — Type of User Roles
Read the source. June 13, 2025; role-specific access to scheduling settings.
[16] SkyPlanner — SkyPlanner APS information package
Read the source. Historical PDF distributed under the September 2022 upload path; scheduling and prediction claims.
[17] SkyPlanner — SkyPlanner brochure
Read the source. Historical 2025 brochure; AI claims, hosting, services and named customer testimony.
[18] SkyPlanner — Agentic Production Scheduling: Automation and Control
Read the source. Article updated October 10, 2026; current clarification of APS automation, agent workflows and outcome measurement.