Go back to Market Research
Asper.ai is a Fractal-owned software business serving consumer packaged goods companies. Its current platform combines demand planning with revenue growth management: pricing, promotions, assortment, product mix, and trade investment. An underlying data platform combines commercial information, while the AskAsper conversational interface lets users explore scenarios and recommendations. Existing enterprise resource planning and trade promotion management systems retain transaction records and receive approved plans and decisions. 1
Overall supply chain score: 4.4/10.
| Dimension | Score |
|---|---|
| Supply chain depth | 4.6/10 |
| Decision and optimization substance | 4.6/10 |
| Product and architecture integrity | 4.4/10 |
| Technical transparency | 4.0/10 |
| Vendor seriousness | 4.2/10 |
Asper addresses commercially meaningful problems. Pricing, promotion expenditure, and product substitution belong in an economic assessment of demand. Its older case studies also report forecast automation and planner productivity improvements. The weakness is the passage from those capabilities to the current promise of coordinated, agentic growth: the public evidence establishes forecast assistance and commercial recommendations much more clearly than a repeatable system that improves profit after implementation and operating costs. 1 6 7
Patents describe probabilistic action recommendations and the evaluation of planner adjustments, without establishing their implementation in the current commercial platform. Deployment descriptions span several cloud services, and the current offering includes consulting, adoption, and capability development. The cost of maintaining this combination weighs directly on the promised returns. 10 11 12 13 15
Asper.ai versus Lokad
Asper starts with the commercial planning organization: combine its forecasts, analyze revenue opportunities, simulate pricing and promotions, and route decisions through existing approval processes. Lokad starts with the decisions themselves: which units to buy, produce, allocate, or discount, given uncertain demand and supply, operational constraints, and the financial consequences of each alternative. Asper’s current promise to preserve existing workflows is convenient for adoption; it also preserves much of the organizational machinery whose productivity should be questioned. 1 See Lokad’s inventory optimization approach.
The distinction matters when a promotion looks attractive. More predicted sales can coincide with lost full-price sales, costly stock transfers, insufficient replenishment, or leftovers after the event. Asper explicitly covers pricing and promotion effects. Its public materials do not expose how uncertain supplier receipts, competing uses of inventory, and those financial consequences enter a common decision calculation. Its deviation-management patent provides a useful starting point—probabilities, action alternatives, constraints, and gains or losses—but does not establish the deployed treatment of that particular supply chain problem. 1 10
Lokad’s Introduction to Supply Chain explains the broader design perspective: automate numerical decisions, express their economics, and improve the decision logic as knowledge accumulates. The relevant productivity comparison includes the people maintaining data mappings, models, exceptions, and commercial rules. A shorter forecast review meeting captures only part of that work.
From demand anticipation to agentic RGM
The UK entity was incorporated in November 2019 as Samya.ai Limited and adopted the Asper.ai name in November 2022. Fractal acquired Samya in June 2021. 2 4
| Period | Public claim or product direction | What the record establishes |
|---|---|---|
| 2020 | Dynamic Demand AI used 200 external factors and automated machine learning for demand anticipation. | The launch described consultation with more than 100 planners and proof-of-value trials with five CPG companies. It did not publish a controlled evaluation of the forecast or revenue claims. |
| 2021 | Fractal acquired Samya, promoting deep learning, reinforcement learning, probabilistic methods, and a tenfold advantage. | The acquisition announcement did not define the quantity, baseline, or experiment behind that advantage. |
| 2022 | Asper launched with an autonomous decisioning engine across demand, inventory, pricing, and promotions. | The announcement simultaneously described the demand planning and pricing/promotion modules as being in limited test release. |
| 2023 | Case studies described low-touch demand planning and improvements to planner productivity. | The documented workflows retained planner adjustments and separated high-risk cases for human attention. |
| 2025 | Fractal announced a phased $20 million investment. | The stated perimeter still covered demand, revenue growth management, inventory planning, and sales execution. |
| Current | Agentic RGM combines commercial agents, AskAsper, and a data platform. | The current description centers on governed commercial decisions and preserving existing enterprise workflows. |
The early AI language was already expansive. In 2022, “autonomous” described a suite whose advertised modules were still in limited test release. The subsequent customer documents offer a narrower and more credible account: automate part of forecast production and help planners handle the remainder. The current agentic presentation needs evidence of what changed in the numerical decision system, beyond the addition of a conversational interface. 5 6 7
There is continuity in the commercial problem being addressed. Demand anticipation, external drivers, planner intervention, and revenue effects recur throughout the record. There is much less continuity in public technical documentation: current product, resource, and AI-philosophy links largely lead to a coming-soon page, while useful case studies survive in archived PDFs. Product maturation should make mechanisms and outcomes easier to inspect. 1 6 7
What the numerical methods disclose
Probabilistic recommendations in the patent record
A Samya patent application published in 2021 describes a deviation-management system. A user specifies a use case and metrics; interconnected models predict deviations, assign probabilities, trace their effects, and recommend actions. The claims explicitly include distributions over action outcomes, user-specified operational constraints, and quantified gains from taking an action or losses from declining it. A feedback mechanism records actions to improve subsequent predictions. 10
This is substantive evidence of an attempt to move beyond a single forecast followed by an alert. The published design connects uncertainty to alternative actions and linked metrics. However, a configurable metric can represent revenue, forecast error, or another target. The patent does not settle whether the deployed product prices all relevant consequences consistently, learns causal effects reliably, or coordinates scarce inventory across competing decisions. Those are implementation questions with direct economic consequences. 10
A second patent, granted in 2025 with a 2020 priority date, concerns forecast adjustments. It compares adjusted forecasts with actual outcomes, identifies systematic biases, and assigns skill scores to the entities making adjustments. It then recommends improvements to subsequent forecasts. This can expose harmful planner intervention, but its central output remains an adjusted forecast. An improvement in prediction quality does not, by itself, determine the best order quantity or price. 11
Demand and supply uncertainty
The commercial product emphasizes demand drivers, pricing elasticity, and scenarios. Neither its current description nor the marketplace listings explains a joint distribution of demand, supplier lead times, available quantities, and other supply uncertainties. The deviation patent is broad enough to discuss multiple metrics; the missing evidence is a concrete supply chain implementation that combines those uncertainties before choosing an action. 1 10 12 13
This gap becomes material when commercial and physical decisions interact. A promotion changes demand precisely when replenishment may be unreliable. A price recommendation that treats future availability as a fixed input can look profitable while shifting costs into expediting, lost sales elsewhere, or residual stock. More demand signals do not resolve uncertainty about what can actually be supplied.
Asper also needs to distinguish predicted sales from the effect of changing a price or promotion. Historical promotions are selected by people responding to expected demand, competitor behavior, and inventory positions. A model that fits those observations can reproduce that selection without estimating the incremental effect of a different action. The public product descriptions identify elasticity and promotion returns, but do not expose the counterfactual evaluation method or its error. 1 13
What the customer evidence measures
The archived pet nutrition case gives useful operational detail: 87 customers, 23 warehouses, multiple trade channels, and a four-month rolling forecast. Low-touch cases were automated; planners adjusted the baseline in high-touch cases. Its results distinguish approximately 60% of the portfolio being automated from 42% of product volume being driven by automated decisions. The document does not establish that purchase orders, production releases, or prices were executed automatically. 6
Its accuracy figures also need their original denominators. Seventy percent of the portfolio improved; that is not a claim of 70% forecast accuracy. One-third of the portfolio reportedly improved by more than 20 percentage points, and the case reports a 6% improvement across three channels. The starting discussion mentions mean absolute percentage error above 25%, but the results do not provide a consistent formula, weighting scheme, final error, and out-of-sample protocol. A buyer cannot reconstruct the improvement from the published numbers. 6
The packaged-food case is stronger on elapsed time. It reports 6.8% improvement in forecast accuracy for two categories in the first eight weeks, more than 5% across categories in year one, 8.2% in year two, and 40% higher planner productivity. The implementation included a discovery workshop, process mapping, a proof of value, and configuration of the future planning process. High-risk situations continued to receive planner attention. 7
These are reported customer benefits, with meaningful scope information. They remain anonymous cases without the forecast metric, control population, productivity denominator, or total staffing and service costs needed to reproduce the assessment. The packaged-food case does not say whether 40% productivity means fewer hours, more items per planner, or a different activity mix. Neither case provides realized net profit after the implementation and recurring work. 6 7
The marketplace claims broaden the promises. AWS describes a commitment to a 5–15% accuracy lift; Microsoft describes a 10–20-point lift and up to 80% portfolio automation. Percentages and percentage points are different measures, and the listings do not establish common baselines or product scope. Category-level accuracy also cannot establish SKU-level replenishment quality: aggregation removes much of the variation that makes individual inventory decisions difficult. 12 13
Software materials and operating boundaries
The most concrete public component descriptions come from vendor-authored marketplace listings.
| Product description | Named components | Ownership implication |
|---|---|---|
| AWS demand forecasting listing | Redshift, Amazon EMR, and ElastiCache | Warehouse, distributed data processing, and caching each require data synchronization, capacity choices, and operational maintenance. |
| Microsoft application listing | Cosmos DB, Azure Database for PostgreSQL, Kubernetes Service, Azure OpenAI, and virtual machines | Two database services, container orchestration, model serving, and an external AI service create several interfaces to operate and secure. |
| Current platform overview | Integration with customers’ existing cloud and data platforms | Customer data infrastructure remains part of the operating arrangement rather than disappearing into the application. |
The AWS and Azure descriptions are separate offerings; the public record does not map them to versions of the current Agentic RGM platform. Even the AWS listing combines a statement that the product is built using AWS services with delivery metadata classifying it as professional services and marking “Deployed on AWS” as “No.” This weakens the listing’s usefulness as a deployment specification. Buyers need the actual topology, service responsibilities, and upgrade path for their installation. 12 13
An Asper lead data scientist vacancy describes building and maintaining model pipelines, deploying models, monitoring their behavior, and updating them with engineering and business teams. Python and SQL feature among the skills sought. The concrete point is the continuing engineering workload: new external signals, changing products, and deteriorating models require maintained pipelines and deliberate revisions. 14
Asper’s description of itself as a decision layer above ERP and trade promotion management is clear. Approval and result synchronization still require failure handling: which system owns a decision, what happens when a write fails, and how a repeated request avoids duplicate execution. Public materials do not provide an API contract, retry semantics, model versioning scheme, or operational recovery procedure. 1
Security boundaries
The Azure listing names Kubernetes and calls the application secure. The website privacy policy describes general protection of personal information. Neither explains the isolation of customer workloads, the privileges of commercial agents, access to production credentials, outbound network restrictions, or recovery from a compromised model or integration. These mechanisms matter when recommendations can become approved commercial actions. 13 16
The current platform promises guardrails and auditability, but does not describe how a price corridor or trade rule is enforced independently of the agent proposing the action. A trace of an unsafe action does not prevent it. The security assessment therefore remains weak until execution permissions, enforcement, and recovery are technically inspectable. 1
Configuration maintenance, productivity, and accountability
Asper’s data platform embeds product attributes, promotion mechanics, and commercial taxonomy. Its agents operate with price corridors, pack ladders, and trade rules. These encode real business knowledge, but they also create maintained configuration. New packs, retailer agreements, promotions, and market structures change the assumptions behind the calculations. The cost of reviewing and safely updating that logic belongs in total ownership cost. 1
The current Value Enablement offering makes the human contribution explicit: expert engagements, adoption support, cross-functional alignment, and sustained RGM capability. PricingOne’s own services include strategic work, implementation support, analytics, and training. Asper’s 2023 white paper likewise called for organizational and process changes. The product is sold together with a substantial change in how commercial teams work. 1 15 8
That qualifies the 2025 announcement’s assertion that the platform eliminates the barriers associated with in-house AI talent, infrastructure, and model training. Outsourcing those duties can be worthwhile. It does not remove their cost, the customer’s data work, or the commercial expertise required to maintain the implementation. The reported proof-of-value period also ends before a full rollout and its recurring obligations. 9 1
The post-2022 productivity question is broader than whether a planner can query an assistant. How many people keep the decisions running, and how quickly can the organization improve the logic? Asper’s case studies report reductions in forecast work. The current public materials do not show a versioned, testable process for changing commercial rules, replaying decisions on historical data, measuring economic effects, and rolling back a bad change. Preserving approvals and adding conversations can accelerate individual tasks while leaving the main improvement process labor-intensive. 6 7 1
Asper’s AWS listing does make an outcome commitment: forecast accuracy lift. It does not publish the contractual metric, acceptance test, remedy, or allocation of responsibility when the target is missed. The relevant economic account should cover realized margin and inventory consequences after software, consulting, and client labor costs. An improvement in a forecast score provides no automatic answer to that account. 12
Supply chain score
Each subscore ranges from 0 to 10. The overall score is the unweighted average of the 25 subscores.
Supply chain depth
| Criterion | Score |
|---|---|
| Economic framing | 6/10 |
| Decision end-state | 4/10 |
| Conceptual sharpness on supply chain | 5/10 |
| Freedom from obsolete doctrinal centerpieces | 4/10 |
| Robustness against KPI theater | 4/10 |
Average: 4.6/10.
Economic framing — 6/10. Pricing, promotion expenditure, mix, and margin are material economic variables. The public product stops short of exposing a common calculation that includes inventory and supply risks alongside those commercial effects. 1 10
Decision end-state — 4/10. The current platform promises approved commercial execution, but the documented cases establish forecast production and planner assistance. Their automation percentages do not identify executed replenishment or production decisions. 1 6 7
Conceptual sharpness on supply chain — 5/10. The deviation patent connects uncertainty, alternative actions, operational constraints, and linked metrics. The commercial implementation of joint demand and supply uncertainty remains unexplained. 10 12
Freedom from obsolete doctrinal centerpieces — 4/10. The offering retains a unified forecast, existing S&OP workflows, planner adjustments, and approval processes. Economic decision automation is less developed publicly than the machinery supporting those practices. 11 12 13
Robustness against KPI theater — 4/10. Customer cases preserve useful scope and timing, but accuracy lifts and portfolio automation lack the definitions needed to connect them to profit or staffing. 6 7 13
Decision and optimization substance
| Criterion | Score |
|---|---|
| Probabilistic modeling depth | 5/10 |
| Distinctive optimization or ML substance | 5/10 |
| Real-world constraint handling | 4/10 |
| Decision automation and labor productivity | 5/10 |
| Resilience under real operational complexity | 4/10 |
Average: 4.6/10.
Probabilistic modeling depth — 5/10. The patent describes distributions over deviations and action outcomes. Public deployment evidence does not explain their calibration, dependency structure, or application to uncertain supply. 10
Distinctive optimization or ML substance — 5/10. The interconnected-metrics and forecast-adjustment patents disclose mechanisms beyond generic AI labels. Their implementation, comparative performance, and role in the current product are not inspectable. 10 11
Real-world constraint handling — 4/10. The current product identifies commercial guardrails, and the patent allows use-case constraints. The joint handling of physical capacity, uncertain receipts, inventory competition, and commercial commitments is not specified. 1 10
Decision automation and labor productivity — 5/10. The pet nutrition and packaged-food cases report actual automation scope and planner productivity benefits. Neither measures the full recurring workload or the productivity of changing decision logic. 6 7
Resilience under real operational complexity — 4/10. Multi-channel and multi-warehouse cases establish relevant operating conditions. They do not disclose performance during supply disruption, failed integration, or substantial changes in assortment and business rules. 6 7
Product and architecture integrity
| Criterion | Score |
|---|---|
| Architectural coherence | 5/10 |
| System-boundary clarity | 6/10 |
| Security seriousness | 3/10 |
| Software parsimony and ownership cost | 4/10 |
| Compatibility with programmatic and agent-assisted operations | 4/10 |
Average: 4.4/10.
Architectural coherence — 5/10. The current data and decision layers form a coherent commercial concept. The relationship between the AWS offering, Azure application, and current platform lacks a public version or deployment map. 1 12 13
System-boundary clarity — 6/10. Asper explicitly leaves transaction records with ERP or trade promotion management and describes synchronization of plans and results. Failure semantics and ownership of partially completed actions remain unspecified. 1
Security seriousness — 3/10. The public descriptions identify cloud infrastructure and general safeguards, but do not expose tenant isolation, agent privileges, credential protection, or enforceable execution boundaries. 1 13 16
Software parsimony and ownership cost — 4/10. Named cloud services, maintained model pipelines, commercial configuration, and specialist services create a substantial operating obligation. Public pricing and customer cases do not quantify that obligation. 12 13 14 15
Compatibility with programmatic and agent-assisted operations — 4/10. AskAsper and approved execution provide an agent-facing product direction. Public API contracts, testable rule changes, and versioned rollback of decision logic are absent from the product materials. 1
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 | 5/10 |
| Evidence density behind technical claims | 5/10 |
Average: 4.0/10.
Public technical documentation — 4/10. Patents and marketplace listings reveal proposed algorithms and components. Current product links frequently lead to a placeholder, leaving a large gap between those records and an implementation guide. 1 10 11 13
Inspectability without vendor mediation — 3/10. No public runnable model, evaluation dataset, or decision API accompanies the commercial claims. The customer PDFs report results without a reproducible measurement procedure. 6 7 1
Portability and lock-in visibility — 3/10. The platform works above existing customer infrastructure, but does not document export of trained models, commercial configuration, action history, or executable decision logic. 1
Implementation-method transparency — 5/10. The packaged-food case describes discovery, process mapping, proof of value, and configuration. Current materials identify specialist adoption work, but not the staffing and maintenance schedule of a full deployment. 7 1 15
Evidence density behind technical claims — 5/10. Two substantive patent records, cloud component descriptions, and detailed archived cases support a meaningful technical assessment. The current agentic product has much less mechanism-level evidence. 10 11 6 13 1
Vendor seriousness
| Criterion | Score |
|---|---|
| Technical seriousness of public communication | 4/10 |
| Technology roadmap integrity | 4/10 |
| Conceptual sharpness | 5/10 |
| Economic accountability | 4/10 |
| Defensibility in an agentic-software world | 4/10 |
Average: 4.2/10.
Technical seriousness of public communication — 4/10. The early tenfold advantage and autonomous-suite claims outstrip their disclosed measurements and release status. More concrete customer documents remain available through archives. 4 5 6 7
Technology roadmap integrity — 4/10. Demand and commercial planning provide a consistent business focus. The public history does not trace how the earlier autonomous engine became today’s agentic platform or which numerical capabilities changed. 3 5 9 1
Conceptual sharpness — 5/10. Linked commercial decisions and measurement of planner bias address concrete problems. The stronger analytical ideas in the patents are only loosely connected to current product explanations. 10 11 1
Economic accountability — 4/10. The forecast-lift commitment is explicit, but its acceptance conditions and remedies are unpublished. Customer results omit realized net profit and the combined vendor, partner, and client cost. 12 6 7
Defensibility in an agentic-software world — 4/10. Commercial data preparation and domain knowledge can retain value. A conversational layer over inherited workflows provides a weaker advantage unless it also reduces the labor needed to operate and improve the numerical decisions. 1 14 15
Sources
The following sources were checked on November 4, 2026. Archived case studies and the white paper retain the historical product descriptions and reported results.
[1] Agentic RGM and platform overview
Agentic RGM and platform overview. Current functional scope, commercial guardrails, AskAsper, data platform, integration boundaries, Value Enablement, and proof-of-value timeline.
[2] Asper.ai Limited company record
Asper.ai Limited company record. UK entity incorporation and former company name.
[3] Samya.ai introduces Dynamic Demand AI
Samya.ai introduces Dynamic Demand AI. 2020 product launch, external signals, planner consultation, and proof-of-value trials.
Fractal acquires Samya.ai. Acquisition and the numerical-method and tenfold-advantage claims.
[5] Fractal announces launch of Asper.ai
Fractal announces launch of Asper.ai. 2022 autonomous-suite positioning and limited test release status.
[6] Pet nutrition demand planning case study
Pet nutrition demand planning case study. Archived five-page case. Forecasting scope, manual intervention, and portfolio-versus-volume automation figures; results page checked visually.
[7] Packaged foods demand planning case study
Packaged foods demand planning case study. Archived packaged-food case. Discovery and configuration approach, staged accuracy figures, and planner productivity; results page checked visually.
[8] From Forecasting to Fulfillment white paper
From Forecasting to Fulfillment white paper. Archived 2023 white paper describing organizational, data, and process work required for implementation.
[9] Fractal invests $20 million in Asper.ai
Fractal invests $20 million in Asper.ai. March 2025 investment announcement, four product areas, and claims about outsourcing AI development burdens.
[10] Deviation prediction and interconnected metrics patent
Deviation prediction and interconnected metrics patent. Published patent design for interconnected deviations, probabilistic action outcomes, constraints, and feedback. A patent does not establish deployment in the current product.
[11] Forecast adjustment and cognitive attributes patent
Forecast adjustment and cognitive attributes patent. Patent granted March 25, 2025, with 2020 priority, addressing forecast-adjustment bias and skill scores.
[12] AWS Marketplace demand forecasting listing
AWS Marketplace demand forecasting listing. Vendor-authored AWS listing: component names, forecast-lift commitment, pricing route, and delivery metadata.
[13] Microsoft Marketplace application and Azure components
Microsoft Marketplace application and Azure components. Vendor-authored Microsoft listing: Azure services, unified forecasting and RGM, automation and accuracy claims.
[14] Lead Data Scientist vacancy
Lead Data Scientist vacancy. Employer vacancy describing model development, pipeline maintenance, production integration, and monitoring.
[15] PricingOne services and capability development
PricingOne services and capability development. Partner services, commercial expertise, and capability development. Its general client results are not attributed to Asper.
[16] Asper website privacy policy
Asper website privacy policy. Website privacy practices; contains no detailed specification of the commercial platform’s execution or isolation controls.