Go back to Market Research
Pigment is a cloud platform for building financial and operational planning applications. Users define multidimensional data, formulas, forecasts, scenarios, reports, and workflows. Its supply chain offering covers demand, inventory, capacity, and sales and operations planning, with links to financial plans. AI agents assist analysis and model construction. (1, 9, 25)
Overall supply chain score: 6.1/10.
| Dimension | Score |
|---|---|
| Supply chain depth | 5.2/10 |
| Decision and optimization substance | 5.4/10 |
| Product and architecture integrity | 7.0/10 |
| Technical transparency | 6.8/10 |
| Vendor seriousness | 6.2/10 |
Pigment’s strongest contribution is a programmable planning environment with increasingly capable tools for maintaining the model itself. Its documentation exposes forecasting methods, production safeguards, failure modes, and substantial engineering work around open-source components. The central supply chain limitation remains: a flexible model and faster scenario preparation do not establish an economic decision engine operating over joint demand and supply uncertainty. The Planner Agent remains on the public roadmap. (3, 9, 15, 17, 25)
Pigment explains how it contains generated Python and enforces read-only impersonation. It also documents less reassuring application-level behavior: formula changes can alter access controls, model edits can be permanent, and some newer AI features sit outside the standard deployment process. (9, 10, 14, 15, 16)
Pigment versus Lokad
Pigment gives a company tools to build its own planning applications. The customer decides which quantities to forecast, which formulas connect them, which scenarios to compare, and which plan to approve. Lokad’s supply chain optimization centers on computing economically ranked operational decisions under uncertainty. Its Supply Chain Scientists maintain the numerical model and the data preparation needed to produce those decisions. The broad design perspective appears in Introduction to Supply Chain. (1, 9)
For a component purchase, Pigment can connect a demand forecast, bill of materials, current inventory, lead-time assumptions, and a purchasing calculation. Cheerz describes such an application. The harder question is how much to buy when supplier arrival, demand, yield, and alternative uses for cash are uncertain. Lokad’s approach assigns economic consequences to the possible outcomes and evaluates the available actions together. Pigment provides a modeling environment in which much of that logic would still need to be designed and validated. (7, 29)
Pigment’s Modeler Agent can reduce implementation costs by automating formula and model construction. But producing a formula quickly does not establish that it expresses the right business objective. The productivity test is whether customer teams can safely revise and improve the decision model with less total effort, including review, regression checks, and operational follow-through. (9, 26)
The numerical model: forecasts and scenarios before economic decisions
Inventory calculations are programmable, not a supplied optimization doctrine
Pigment documents iterative calculations across blocks, including an inventory example connecting beginning stock, incoming replenishment, and ending stock over time. Its current PREVIOUSOF mechanism permits up to ten metrics in an iterative configuration, with a common iteration dimension. These are useful computational primitives for expressing a customer’s logic. (29)
They do not choose the economic formulation. A stock equation can be correct while its replenishment rule ignores substitution, expiry, a supplier’s erratic delivery, or the opportunity cost of scarce capital. Adding dimensions or formulas does not automatically add the corresponding decision alternatives. As explained in Holimization, the important improvement may be a change in what the model considers, not a more accurate calculation within its existing assumptions.
The public supply chain offer emphasizes connected plans and rapid comparison of scenarios. This supports examining capacity or purchasing consequences across functions. It does not expose a native objective that ranks competing operational actions by their expected economic return, or a supplied method for jointly valuing demand, lead-time, and capacity uncertainty. That work remains a substantial part of the application design. (1, 5, 29)
Forecasting disclosure contains useful quantitative detail
Pigment’s Predictions feature outputs a baseline with lower and upper bounds for a 95% prediction interval. Its AutoML selection evaluates several models per series using MASE on the last 20% of historical data. The documented choices include AutoETS, Chronos-2, Prophet, and simpler methods; Chronos-2 is used without customer-specific fine-tuning. (2, 3)
The published benchmarks use MASE, an error measure relative to a seasonal-naïve baseline:
| Benchmark | Chronos-2 average MASE | Seasonal-naïve baseline |
|---|---|---|
| Monthly, 12-month horizon | 1.102 | 1.000 |
| Weekly, 52-week horizon | 0.900 | 1.000 |
| Daily, 365-day horizon | 0.718 | 1.000 |
The dataset emphasizes typical business metrics, particularly financial sales series, and contains no customer data. Monthly TimesFM 2.5 scores 0.981 in the same comparison. Thus Chronos-2 loses to the naïve baseline in that monthly test, despite the surrounding broad language about its superior performance. These results cannot be transferred to SKU-location replenishment or interpreted as percentages of correct forecasts. (3)
A prediction interval is useful uncertainty information; its nominal 95% coverage is not 95% forecasting accuracy. For inventory decisions, the missing step is a validated joint distribution of the quantities that affect the action. Marginal bands around individual series do not specify correlated demand across products, uncertain supplier lead times, or the cost of the resulting shortages. Pigment’s public documentation does not show those uncertainties combined into an action-value calculation. (2, 5)
External factors can predict without explaining an intervention
Pigment accepts up to ten external-factor metrics. The data must cover at least 90% of both historical and future periods; missing values are filled. A Granger test checks up to five lags and selects a factor when a p-value falls below 0.005. Prophet can expose an associated impact coefficient; the documentation does not offer that coefficient for Chronos-2 or AutoML. (4)
A marketing variable can help predict sales because the marketing team already knows something about demand. That does not establish what would happen if the company changed spending. Likewise, future weather or competitor prices must themselves be forecast unless known in advance. Their uncertainty and availability at the original decision date matter. The documented screening procedure does not establish causal promotion optimization or joint uncertainty propagation. (4)
Pigment’s August 2026 supply chain article explicitly argues for ranged forecasts and recognizes variable lead times. An anonymized retailer’s earlier case, however, organizes planning around a single agreed demand number and scenarios for hitting targets. The platform can accommodate both practices; the customer’s methodology determines whether uncertainty informs the decision or is compressed into a consensus plan. (5, 6)
Agent productivity: building the model is the substantive advance
Model construction can accelerate, while validation remains necessary
The Modeler Agent creates and edits blocks, formulas, dimensions, and calendars. Its October 2026 documentation describes Ask, Build, and Plan & Build modes, with optional automatic acceptance except for deletions. Changes are permanent and attributed to the initiating member in application history. (9)
Pigment reports that the agent creates 70% of new metrics and formulas. It also reports an Unilever North America overhead application built and launched in 25 days, compared with an estimated four-to-six-month alternative. That is evidence about construction speed in financial planning. The comparison does not provide a measured counterfactual, defect rate, subsequent maintenance effort, or supply chain decision outcome. (26)
Model maintenance consumes expensive specialist labor: reconciling definitions, tracing dependencies, adjusting formulas, and rebuilding reports. An agent can reduce that work. The economic gain depends on the quality and longevity of the resulting application, not the fraction of code or formulas generated automatically. A large volume of unnecessary metrics would increase maintenance rather than productivity.
Daily planning still has a large human component
Cheerz reports reducing its monthly operations-planning cycle from a week to less than a day and cutting inventory by 30%. Its purchasing application imports daily data from the stock system and models components through a prepared bill of materials. Danone’s infant-milk and medical-nutrition planning case describes a strategic-plan preparation task falling from two weeks to a day, conditional on receiving inputs on time. Its customized S&OP and strategic model was built with Statera Consulting in three months. (7, 8)
These cases identify meaningful workload reductions. Their denominators remain particular processes, rather than the total cost of keeping operational decisions correct. Faster consolidation also permits more planning cycles, as Cheerz reports. More frequent review can improve responsiveness while leaving the business dependent on the same human approvals and model corrections. Neither case establishes autonomous replenishment across the customer’s full operation. (7, 8)
MCP is a connector; model-changing tools need their own assessment
Pigment’s MCP server authenticates the user, applies existing permissions, and permits blocks to be excluded from AI access. Its October tool changelog includes raw-data retrieval, dependency inspection, and performance investigation; ranking and thresholds for retrieved data now run in the external AI tool. (11, 28)
The connector alone is a low-effort integration feature, not evidence of a supply chain decision engine. Moving a user’s question-and-report workflow into an external assistant preserves the underlying manual operating model. Pigment’s more substantive agentic work lies in changing and maintaining the application, with controlled execution and documented tools. Autonomous economic planning still lacks delivered evidence. (9, 15, 25)
Security architecture and the cost of changing models safely
Generated Python is treated as hostile code
Pigment’s June 2026 engineering article describes the Analyst Agent’s generated-Python runtime. Execution uses dedicated workers, per-run isolation, blocked network access, no mounted credentials, resource limits, a read-only root filesystem, disposable temporary storage, and gVisor. Each sandbox pod serves one execution. The article explicitly rejects regex checks as a sufficient security boundary. (15)
For larger datasets, a trusted service fetches the data and pushes it into the sandbox, preserving the sandbox’s lack of cloud credentials. This is a material design response to malicious prompts, unsafe generated code, and accidental data exposure. Security is enforced outside the generated program. The article gives a concrete architecture to examine and test. (15)
A separate April article explains administrative impersonation: a signed read-only state reaches API middleware, and every endpoint must declare whether it permits such sessions. A custom build-time analyzer rejects missing declarations. This reduces the risk that a new endpoint silently bypasses the control. The guarantee still depends on correct classification of allowed endpoints; it is not a database-enforced prohibition on every possible write. (16)
Formula permissions can undermine data permissions
Pigment documents how a modeler without security-administration rights can nevertheless expose restricted data by changing a formula used in access rules. Its mitigation places vulnerable blocks in a Security Folder accessible only to security-authorized users. The article initially recommends trustworthy modelers, then explains this technical separation. (14)
The mechanism matters because model code participates in defining security policy. A containment design must therefore cover the dependency chain that computes permissions, not merely the final access-rights metric. Correctly identifying and protecting those dependencies becomes continuing configuration work as models evolve. Faster agent-written changes increase the importance of this boundary; confidence in the person issuing the prompt does not make every generated change safe. (14)
Deployment and recovery have consequential exceptions
Pigment’s Test & Deploy separates structural changes from production use, but data is generally not synchronized between environments. Its documentation warns that emergency production access-rule changes will be overwritten by the next deployment unless repeated in development. It also warns that deploying a formula can discard existing inputs when overrides are disabled. These are concrete failure paths requiring review and testing. (13)
The Modeler documentation currently limits managed Test & Deploy applications to Ask mode. That restricts its advertised model-writing capability in the governed application workflow. Frames, the newer generated-interface feature, also lacks Test & Deploy support: authors duplicate, iterate, and publish instead. (9, 10)
Frames does have a documented execution boundary. Pigment’s public authoring specification describes JavaScript running in an opaque-origin sandboxed iframe with network access, external code loading, browser storage, and parent access blocked. Declared data sources and sinks mediate model interaction. Its product documentation separately allows security-authorized users to disable write-confirmation dialogs. Browser isolation reduces code exposure; it does not validate the business meaning of an allowed write. (10, 12)
Block recovery is also more involved than an undo command. The documented process usually offers daily backups from the preceding seven days, restores blocks as new objects, and initially restricts them to application administrators. Returning their data to the live model can require copying or formula-based reconciliation. That protects recovery from accidental broad disclosure, while leaving meaningful restoration work after a bad model change. (19)
Software bill of materials and ownership cost
| Layer | Disclosed components | Evidence |
|---|---|---|
| Application | C# ASP.NET Core on Linux, Go services; React and TypeScript interface | Current SRE recruitment specifies the production stack. (18) |
| Data and messaging | PostgreSQL through Cloud SQL and CloudNativePG, SingleStore, Elasticsearch, RabbitMQ, Temporal | The same posting identifies these as deployed platform components. (18) |
| Infrastructure | Google Kubernetes Engine, Terraform, ArgoCD, Istio, Vault | Current infrastructure responsibilities. (18) |
| Predictions | Python, Dask, Google Cloud Storage, Avro transfers | September 2025 engineering architecture. (17) |
| Agent analysis | LangGraph/LangChain in the 2024 design; a hardened Python execution service documented in 2026 | Dated engineering descriptions. (15, 23) |
This is a substantial software estate. Multiple data systems, queues, a workflow service, a service mesh, and a distributed forecast runtime create upgrade, observability, access-control, and incident-recovery obligations. The cost question is whether each layer pays for itself in delivered customer capability. Pigment provides unusually useful evidence of those engineering obligations, but no cost per maintained supply chain decision or comparable workload.
The Dask article explains that the original forecast service exhausted single-machine resources. The replacement stages data through cloud storage and distributes work. Pigment encountered an autoscaling defect that could leave more than 200 workers allocated, supplied a patch, and added its own handling for worker shutdown and cancellation of spawned subtasks. This is engineering beyond simply advertising an imported open-source name, and a concrete example of the maintenance cost such a dependency creates. (17)
The Python sandbox has another explicit cost: pre-warmed placeholder pods remain allocated to reduce startup latency. The article reports startup around one to two seconds and identifies paid idle capacity as the trade-off. Those timings concern execution startup, not a complete optimized supply chain calculation. (15)
Customer configuration adds a separate cost. Formula dependencies, security rules, scenario definitions, imports, and generated interfaces all require ownership. Predictions also consume an annual subscription quota by number of series computed; deleting or expiring a result does not restore quota. Frequent recalculation and experimentation therefore have a commercial cost dimension alongside compute and review effort. (2, 9, 13)
Pigment’s security page specifies multi-zone operation and a tested disaster-recovery objective of six hours with up to 24 hours of data loss. Those are explicit recovery objectives. Application-level restoration and reconciliation still determine how quickly a planning team can resume trustworthy decisions after a damaged model or incorrect import. (19, 20)
Technology history and roadmap integrity
| Period | Public development | Assessment |
|---|---|---|
| 2020–2021 | Series A in December 2020 and $73 million Series B in November 2021; the product sought to replace spreadsheet-based collaborative planning. | The early proposition centered on modeling and shared business data. (21) |
| May 2023 | Pigment AI announced conversational access and faster business insights. | The generative-AI positioning followed the ChatGPT moment; it should be judged against subsequent delivery. (22) |
| October 2024 | Engineering described a LangGraph analyst, tool use, human interaction, and testing difficulties. | The article treated unsolicited broader business-optimization advice as a bug in the analyst’s scope. (23) |
| March 2025 | Analyst, Modeler, and Planner agents announced as a roadmap, with private testing. | Three announced roles did not mean three generally available products. (24) |
| September 2025 | Distributed predictions architecture and dependency problems documented. | Specific implementation work and failure handling are visible. (17) |
| March–October 2026 | Modeler launched; containment and impersonation designs published; model-building usage reported. | Real delivery supports the productivity direction. (9, 15, 16, 26) |
| October–November 2026 | $250 million Series E announced; Planner still appears on the roadmap. | Funding does not close the remaining gap in delivered operational decision capability. (25, 27) |
The development record includes delivered capabilities and published accounts of the engineering problems they created. The unfinished Planner remains important because recommending economic operational actions is the capability most directly relevant to this review. (9, 15, 17, 25)
The conceptual tension is between a platform that helps people build and discuss plans and a system that continuously computes better actions. Pigment’s own customer cases emphasize alignment, scenario exploration, and more effective planning meetings. Its more recent uncertainty guidance is stronger than a single-number planning doctrine, but the technical link from that guidance to a validated operational decision policy remains unestablished. (5, 6, 8)
Economic accountability
Cheerz’s inventory reduction and the named planning-time improvements give the buyer concrete outcomes to investigate. They do not isolate the contribution of Pigment from changed process frequency, improved master data, consultant work, or a newly constructed planning model. The application builder’s responsibility for the formulas makes that attribution especially important. (7, 8)
A serious acceptance test should connect the deployed decision logic to realized outcomes, with agreed baselines and explicit responsibility for correcting failures. Successful software execution, timely inputs, or widespread user adoption can coexist with an economically poor replenishment rule. Pigment’s public cases do not define a remedy for that situation or show a harmful operational recommendation traced through reformulation and prospective retesting. Its software-change controls support this work; they do not supply the economic evaluation. (8, 13, 19)
Supply chain score
Supply chain depth
| Criterion | Score |
|---|---|
| Economic framing | 5/10 |
| Decision end-state | 5/10 |
| Conceptual sharpness on supply chain | 6/10 |
| Freedom from obsolete doctrinal centerpieces | 4/10 |
| Robustness against KPI theater | 6/10 |
Average: 5.2/10.
Economic framing — 5/10. Financial and operational quantities can share a model. A native action-ranking objective that accounts for the full economic consequences of uncertain supply and demand is not demonstrated. (1, 5, 29)
Decision end-state — 5/10. Customers build purchasing and inventory applications, but plans and scenarios remain the dominant output. The dedicated Planner capability is still on the roadmap. (7, 25)
Conceptual sharpness on supply chain — 6/10. Recent guidance recognizes uncertain demand and supply. The public implementation evidence is weaker on converting that uncertainty into jointly valued operational choices. (4, 5)
Freedom from obsolete doctrinal centerpieces — 4/10. Consensus planning, target alignment, and repeated human scenario review remain central in customer deployments. Faster execution of those processes does not replace their methodological limits. (6, 8)
Robustness against KPI theater — 6/10. Forecast benchmarks use defined errors and disclose losing results. Productivity claims have identifiable task scopes, although broad agent-generation metrics and estimated alternatives provide limited evidence of net economic value. (3, 7, 26)
Decision and optimization substance
| Criterion | Score |
|---|---|
| Probabilistic modeling depth | 5/10 |
| Distinctive optimization or ML substance | 6/10 |
| Real-world constraint handling | 5/10 |
| Decision automation and labor productivity | 6/10 |
| Resilience under real operational complexity | 5/10 |
Average: 5.4/10.
Probabilistic modeling depth — 5/10. Forecast intervals and explicit discussion of uncertainty improve on a pure point-forecast offer. Joint calibration and propagation into demand-and-supply decisions are not shown. (2, 5)
Distinctive optimization or ML substance — 6/10. Forecast selection, computational modeling, and model-building agents have inspectable mechanisms. Integrated open-source models and flexible formulas do not establish distinctive supply chain optimization by themselves. (3, 9, 17, 29)
Real-world constraint handling — 5/10. Dimensional models and customer-built component calculations can encode operational detail. The quality of the action space, constraints, and economic approximations remains application-specific. (7, 29)
Decision automation and labor productivity — 6/10. Task savings and model-construction acceleration are substantive. Full operating and improvement labor, including validation and exception work, remains unmeasured. (7, 8, 26)
Resilience under real operational complexity — 5/10. Model inspection, deployment, and restoration support correction. A demonstrated cycle from economic failure through revised decision formulation and retesting is missing. (9, 13, 19)
Product and architecture integrity
| Criterion | Score |
|---|---|
| Architectural coherence | 8/10 |
| System-boundary clarity | 7/10 |
| Security seriousness | 8/10 |
| Software parsimony and ownership cost | 4/10 |
| Compatibility with programmatic and agent-assisted operations | 8/10 |
Average: 7.0/10.
Architectural coherence — 8/10. Computation, analysis, and model editing share explicit planning objects. Newer interfaces and editing capabilities still have exceptions to the established deployment lifecycle. (9, 10, 13)
System-boundary clarity — 7/10. Documentation distinguishes model logic, generated interfaces, external tools, and execution services. Several important controls depend on customer configuration and product mode. (10, 11, 12, 15)
Security seriousness — 8/10. Published runtime and API controls address concrete threat paths, with enforcement outside generated code. Formula-derived permissions and deployment exceptions leave material configuration risks. (13, 14, 15, 16)
Software parsimony and ownership cost — 4/10. The platform maintains a large set of stateful and orchestration components. Documented capacity and dependency problems make their maintenance cost tangible; customer-level total ownership cost is not quantified. (17, 18)
Compatibility with programmatic and agent-assisted operations — 8/10. Agents can inspect and modify the model, while generated code has defined execution boundaries. Governed production change and autonomous economic evaluation remain less complete. (9, 12, 15, 28)
Technical transparency
| Criterion | Score |
|---|---|
| Public technical documentation | 8/10 |
| Inspectability without vendor mediation | 6/10 |
| Portability and lock-in visibility | 5/10 |
| Implementation-method transparency | 8/10 |
| Evidence density behind technical claims | 7/10 |
Average: 6.8/10.
Public technical documentation — 8/10. Detailed product and engineering material exposes mechanisms, limits, and known failure paths. Current production behavior is still not independently reproducible from documentation alone. (3, 13, 15, 17)
Inspectability without vendor mediation — 6/10. Public formulas, interfaces, and benchmark tables permit meaningful technical scrutiny. Reproducing customer outcomes or the complete hosted calculation environment requires more access. (3, 7, 12, 29)
Portability and lock-in visibility — 5/10. Public integration and agent interfaces improve accessibility. Customer applications depend on Pigment-specific dimensions, formulas, permissions, and workflows whose migration cost is not established. (9, 12, 13, 29)
Implementation-method transparency — 8/10. Deployment, recovery, modeling controls, and named implementation work are well described. Complete recurring effort and economic acceptance criteria remain less visible. (8, 9, 13, 19)
Evidence density behind technical claims — 7/10. Engineering disclosures connect claims to specific implementations and defects. Supply chain automation and net economic benefit have thinner evaluation evidence than the platform mechanics. (7, 15, 17, 25)
Vendor seriousness
| Criterion | Score |
|---|---|
| Technical seriousness of public communication | 6/10 |
| Technology roadmap integrity | 7/10 |
| Conceptual sharpness | 6/10 |
| Economic accountability | 5/10 |
| Defensibility in an agentic-software world | 7/10 |
Average: 6.2/10.
Technical seriousness of public communication — 6/10. Technical articles acknowledge difficult engineering and security limits. Broad AI productivity language sometimes outruns the narrower benchmarks and deployment scope supplied. (3, 15, 17, 26)
Technology roadmap integrity — 7/10. The progression from collaborative modeling to analysis and model-building agents includes identifiable delivery. The continuing Planner roadmap leaves the most consequential supply chain promise unfinished. (21, 22, 23, 24, 25)
Conceptual sharpness — 6/10. Pigment recognizes that model construction and maintenance constrain planning productivity. Its public operating model remains less clear on replacing consensus forecasts and manual scenario choice with economic decision policies. (5, 6, 9)
Economic accountability — 5/10. Named cases report concrete benefits and implementation work. Responsibility for validating customer-built decision logic and remedies for poor returns are not resolved in those accounts. (7, 8, 26)
Defensibility in an agentic-software world — 7/10. A governed computation environment and tools that change the model offer more durable value than conversational access alone. The remaining challenge is demonstrating that faster application construction produces better recurring decisions with lower total effort. (9, 13, 15, 28)
Sources
[1] Supply Chain Planning (SCP) Software | Pigment
Supply Chain Planning (SCP) Software | Pigment. current; accessed November 4, 2026. Product scope and connected planning.
[2] Machine Learning Forecasting
Machine Learning Forecasting. 2026-05-13; accessed November 4, 2026. Intervals, forecasting workflow, and subscription quota.
[3] Choose your Prediction model
Choose your Prediction model. 2026-05-05; current body; accessed November 4, 2026. Model selection and published benchmark tables.
[4] Use External Factors in your Predictions
Use External Factors in your Predictions. current; accessed November 4, 2026. Factor screening, completeness rules, and impact coefficients.
[5] Maintaining credible supply chain plans through demand and forecasting uncertainty
Maintaining credible supply chain plans through demand and forecasting uncertainty. 2026-08-07; updated 2026-09-03; accessed November 4, 2026. Demand ranges and supply variability.
[6] How a Fashion Retailer Transformed Supply Chain with Pigment
How a Fashion Retailer Transformed Supply Chain with Pigment. undated; accessed November 4, 2026. Anonymized deployment, consensus forecasting, and scenario practice.
[7] How Cheerz Reduced Inventory Levels by 30% with Pigment
How Cheerz Reduced Inventory Levels by 30% with Pigment. undated; accessed November 4, 2026. Named inventory and workload results; application construction.
[8] How Danone Transformed Demand Planning with Pigment
How Danone Transformed Demand Planning with Pigment. undated; accessed November 4, 2026. Named strategic planning implementation and conditional time saving.
Run the Modeler Agent. 2026-03-04; updated 2026-10-19; accessed November 4, 2026. Operating modes, permanent edits, permissions, and deployment restrictions.
[10] Create experiences with Frames
Create experiences with Frames. current; accessed November 4, 2026. Publication workflow, write confirmation, and deployment limits.
MCP Server. 2026-03-06; updated 2026-11-02; accessed November 4, 2026. Authentication, data access, and external-client integration.
[12] Building Pigment Frames: public agent instructions
Building Pigment Frames: public agent instructions. current; accessed November 4, 2026. Vendor-authored public specification for generated JavaScript and sandbox restrictions.
[13] Understand Test & Deploy for Pigment Features
Understand Test & Deploy for Pigment Features. 2026-03-05; updated 2026-10-01; accessed November 4, 2026. Environment behavior, production overrides, and security-rule deployment.
[14] Use Security Folder to Enable Trustless Modeling and Data Protection
Use Security Folder to Enable Trustless Modeling and Data Protection. 2025-08-22; accessed November 4, 2026. Documented formula exposure paths and mitigation.
[15] Sandboxing LLM-generated code execution | Pigment’s Tech Blog
Sandboxing LLM-generated code execution | Pigment’s Tech Blog. 2026-06-10; accessed November 4, 2026. Threat model, production runtime isolation, dataset transfer, and capacity costs.
[16] Impersonation Done Right: Tokens, Read-Only Guarantees, and Audit Trail | Pigment’s Tech Blog
Impersonation Done Right: Tokens, Read-Only Guarantees, and Audit Trail | Pigment’s Tech Blog. 2026-04-08; accessed November 4, 2026. API enforcement and build-time coverage checks.
[17] How We Turbocharged Our Prediction Pipeline Using Dask | Pigment’s Tech Blog
How We Turbocharged Our Prediction Pipeline Using Dask | Pigment’s Tech Blog. 2025-09-02; accessed November 4, 2026. Distributed forecast architecture and dependency fixes.
[18] Senior SRE — public Pigment recruitment posting
Senior SRE — public Pigment recruitment posting. current; retrieved public Lever API; accessed November 4, 2026. Explicit production stack, retrieved through the public Lever posting API.
[19] Restore a Block from Backup
Restore a Block from Backup. updated 2025-11-10; accessed November 4, 2026. Restoration scope, access restrictions, and reconciliation.
[20] SOC2 Type 2 Certified Data Safety | Pigment
SOC2 Type 2 Certified Data Safety | Pigment. current; accessed November 4, 2026. Disaster-recovery objectives and infrastructure scope.
[21] Pigment has announced a $73 million Series B
Pigment has announced a $73 million Series B. 2021-11-15; accessed November 4, 2026. Original planning proposition and early funding history.
[22] Pigment to deliver faster and smarter business insights through Pigment AI
Pigment to deliver faster and smarter business insights through Pigment AI. 2023-05-03; accessed November 4, 2026. Initial generative-AI announcement.
[23] Agentic AI for data analysis in Pigment | Pigment’s Tech Blog
Agentic AI for data analysis in Pigment | Pigment’s Tech Blog. 2024-10-31; accessed November 4, 2026. Dated agent implementation and limits.
[24] AI agents: How Pigment will change business planning again
AI agents: How Pigment will change business planning again. 2025-03-25; updated 2026-09-03; accessed November 4, 2026. Three-agent announcement and private-testing status.
[25] Pigment’s Business Planning Roadmap: What’s Next?
Pigment’s Business Planning Roadmap: What’s Next?. current; accessed November 4, 2026. Released capabilities and outstanding roadmap items.
[26] Unify, build, automate: Pigment’s framework for agentic planning
Unify, build, automate: Pigment’s framework for agentic planning. 2026-10-08; accessed November 4, 2026. Reported modeling usage and Unilever application example.
[27] Pigment raises $250M at $2.3B valuation
Pigment raises $250M at $2.3B valuation. 2026-10-21; accessed November 4, 2026. Current funding announcement.
[28] Tools and skills changelog
Tools and skills changelog. 2026-10-15; updated 2026-11-03; accessed November 4, 2026. Recent changes in external-agent tool responsibilities.
[29] Iterative Calculations across multiple Blocks using PREVIOUSOF
Iterative Calculations across multiple Blocks using PREVIOUSOF. current; accessed November 4, 2026. Iterative inventory example and computational limits.