Go back to Market Research
Algonomy supplies retail software for demand forecasting, replenishment, merchandising, supplier collaboration, customer data, and ecommerce personalization. Order Right generates purchasing and transfer recommendations; Vendor Link connects retailers with suppliers. ADA completed its acquisition of Algonomy in July 2026 and is bringing the combined business under the ADA brand. (1, 2)
Overall supply chain score: 3.9/10
| Dimension | Score |
|---|---|
| Supply chain depth | 3.6/10 |
| Decision and optimization substance | 4.4/10 |
| Product and architecture integrity | 4.2/10 |
| Technical transparency | 4.2/10 |
| Vendor seriousness | 3.2/10 |
Order Right follows an identifiable process: forecast product demand, apply replenishment constraints, and return order recommendations. Public material describes retail constraints and operational integrations, but leaves the central economic calculation unexplained. It does not establish how competing orders are valued under joint demand and supply uncertainty. A large collection of forecasting algorithms cannot fill that gap.
The communication makes the weakness more consequential. ADA advertises 97% demand forecast accuracy without the definition needed to interpret it. Algonomy’s history also stretches from a human-like analytics assistant announced in 2017 to today’s agentic positioning, with little measured continuity between the promise and the work eliminated. The clearest productivity evidence concerns supplier administration, while the numerical replenishment engine remains difficult to inspect. (3, 5, 10, 15)
Algonomy versus Lokad
The practical difference concerns who writes and maintains the logic deciding what to buy, move, or discount. Algonomy supplies packaged forecasting and replenishment modules, configured by its engineers and supported by data, application, and modeling specialists. Retailers supply operating policies and can amend recommendations. Its merchandising offering adds analytical and conversational interfaces for category teams. (2, 13)
Lokad’s quantitative supply chain approach organizes the engagement around executable decisions valued in money under uncertainty. A Supply Chain Scientist prepares data, formalizes the decision logic, implements monitoring, and maintains the production process within the managed service. The economic priorities and available actions become part of that maintained computation.
Consider fresh products sharing shelf space and a supplier delivery. More frequent orders can reduce spoilage while increasing transport and receiving costs; discounting one product can divert demand from another. The engineering problem is to compare these coupled actions and their uncertain financial consequences. Algonomy documents forecasting inputs, order constraints, and merchandising functions, but does not expose a common calculation resolving those trade-offs.
Both offerings require maintenance as the business changes. The relevant cost is the effort to correct a faulty decision rule, test it across the assortment, and deploy the correction. Lokad’s Introduction to Supply Chain explains this economic perspective; its implementation must still justify its service cost through better decisions and lower recurring workload.
What Order Right computes
Order Right’s FAQ describes multivariate SKU-location forecasting with hierarchical methods for sparse series and automatic selection among candidate models. Outputs cover store purchases, warehouse purchases, and transfers, with delivery through Excel or an ERP purchase-order interface. (2)
The 2022 datasheet supplies more detail about the decision stage. Forecasts feed proprietary replenishment algorithms that consider expiration, shelf life, lead time, minimum order and display quantities, and ordering frequency. Users choose a risk level; potential waste can enter model selection. These are useful controls, but the document gives no monetary objective, probability model behind the risk setting, or algorithm for reconciling scarce resources across products. (3)
This distinction matters when the forecast is reasonable and the recommendation is still bad. A case pack can exceed the economically desirable quantity. Accepting a small availability loss may save more in spoilage than it costs in lost margin. A shared truck or budget can make the best individual order an inferior use of resources. A replenishment engine needs explicit valuations for those decisions; improving the forecast alone cannot supply them.
The grocery product page describes promotion shifts, weather predictors, configurable delivery calendars, and product-level expiry alerts. The forecasting page also claims to predict cannibalization between similar products. No published experiment separates incremental promotion demand from sales transferred between products. That separation determines whether stocking a promoted line creates profit or merely moves the same sale. (4, 5)
Supply uncertainty remains unresolved
The public FAQ identifies contractual lead time and expected arrival dates as inputs. The Southeast Asian grocery case says the application dynamically models supply parameters and coordinates inventory across several echelons. Neither explains distributions of actual receipt dates, correlated shortages, or how upstream uncertainty changes downstream purchasing. (2, 6)
A late supplier delivery can affect several stores simultaneously, while a promotion changes both demand and substitution. Calculating a demand forecast for each location and updating a lead-time parameter does not establish an economic treatment of those joint risks. The missing evidence is a calibrated model of possible outcomes connected to the order calculation.
Algonomy advertises probabilistic demand models in its Databricks consulting offering. This bespoke service supplies no public distributional or calibration detail and does not identify the model used inside Order Right. The connection between those models and production order calculations remains unexplained. (9)
Accuracy claims that conceal the planning problem
ADA’s forecasting page displays 97% demand forecast accuracy. It supplies no error formula, forecast horizon, aggregation, weighting, exclusions, or out-of-sample evaluation. A second case study claims more than 95% accuracy across store, product, and SKU levels, again without defining the measurement. These are serious deficiencies in a product sold on forecasting quality. (5, 23)
The historical M5 retail competition illustrates why the missing definition matters. Recalculating 100 × (1 − WAPE) on the final 28 days for the organizers’ top 50 point-forecast submissions gives roughly 22.1–25.1% at SKU/store/day, versus 95.8–98.2% when products and stores are pooled by day. The overall winner reaches approximately 99.8% if those days are pooled as well. These are diagnostic recalculations of the published forecasts, not M5’s official scoring metric. (8)
Aggregation can make an impressive percentage coexist with large errors where replenishment decisions occur. It cancels shortages against surpluses that cannot physically cancel: extra yogurt in one shop does not satisfy a customer seeking bread in another. Algonomy’s 97% figure cannot be located within this spectrum, and therefore establishes neither usable SKU-level precision nor economic returns.
The Southeast Asian case makes a different claim: forecasting improved for 90% of SKUs. That measures the share of series improved, not 90% forecast accuracy. The case also describes reduced stockouts and inventory, but gives no reproducible evaluation connecting the forecast changes to those outcomes. (6)
A useful validation would freeze forecasts before the relevant ordering deadline, evaluate them at the action’s product-location horizon, and trace the resulting orders into margin, waste, availability, and handling costs. A high accuracy percentage stripped of that context encourages precisely the wrong procurement criterion.
Technology history: from conversational intelligence to agents
2017: Maya and the human-like analytics promise
Manthan announced Maya in January 2017 as an AI-powered conversational agent for business analytics. The release described interaction with enterprise analytics as if with a human, role-sensitive responses, continuous learning, and Amazon Alexa integration. The announced presentation was at the NRF retail convention. (10)
The mechanism disclosed was a conversational entrance to analytics. The release did not document reliable task coverage, failure rates, autonomous business decisions, or measured labor savings. Its intelligence language considerably exceeded the evidence offered. Today’s LLM capabilities cannot validate those earlier claims retrospectively.
2021–2022: a unified platform and thousands of algorithms
The 2021 RichRelevance–Manthan merger announcement promoted a unified retail decision platform and more than 300 prebuilt algorithms across the business. Its concrete examples centered on personalization, customer activation, analytics, and merchandise processes. (11)
In January 2022, Algonomy announced an intended acquisition of Linear Squared’s forecasting business, subject to approvals. Later that year, Order Right’s datasheet described automated forecast-driven replenishment. (12, 3) This established a more specific supply chain product proposition, but did not publish the optimization formulation or financial decision rule.
The present FAQ now promotes more than 2,000 candidate forecasting algorithms and a general advantage of at least 20 percentage points over traditional methods. (2) Algorithm counts reveal neither diversity of methods nor performance on unseen data. Selecting from more candidates can increase validation and maintenance work, especially when many product histories contain little information.
2026: agentic language, familiar operating steps
ADA’s current supply-planning page describes Maya answering business questions and recommending actions. The replenishment page now calls its forecasting GenAI-powered, while still describing the sequence of integrating data, forecasting demand, generating orders, and acting on alerts. No published technical account identifies the generative model’s contribution to numerical demand prediction. (13, 25)
Supplier collaboration is separately promoted through Vendor Link and ATLAS: onboarding, shared forecasts, purchase-order exchange, shipment validation, invoice matching, and promotion collaboration. (14) These activities can be automated, but attributing additional productivity to agents requires a comparison with the workflow automation already delivered.
The historical problem is unresolved accountability for successive claims. Maya’s conversational premise survives nine years later; the evidence still does not show how much total planning work it removes. The transition from machine learning to generative and agentic labels is much more explicit than the corresponding change in the replenishment computation.
Productivity, configuration maintenance, and ownership cost
There is concrete evidence of administrative work being removed. The Keells case describes KANE, its implementation of Algonomy’s supplier platform, replacing repeated data extraction, email exchanges, and manual reconciliation with shared data and electronic workflows. It reports 92% faster supplier onboarding and 85% faster product onboarding. (15)
Those gains matter in a retailer coordinating hundreds of suppliers. They concern transaction administration and were documented before the current ATLAS positioning. They do not establish a reduction in the human work required to maintain forecasting and purchasing logic.
For Order Right, the user defines ordering frequency. The grocery offering also exposes scheduling rules, model-selection criteria, and per-product alert thresholds. (2, 4) Each becomes a maintained assumption. A change to supplier calendars can alter stock coverage, expiry exposure, and the number of alerts. Improving one product’s setting can increase workload elsewhere.
The cost therefore includes the people who investigate data changes, revise constraints, check recommendation quality, and maintain ERP mappings. Vendor specialists can perform part of that work; their recurring effort still belongs in total cost of ownership. The public material does not quantify the remaining planner hours or the engineering effort per change.
A conversational answer can accelerate an investigation. The larger productivity gain comes when the correction becomes tested executable logic applied across all affected products. Algonomy’s public planning material does not expose versioned policy changes, historical replay of revised decisions, or controlled rollback of the numerical engine. This limits the evidence for inexpensive continuous improvement.
Software bill of materials
The evidence identifies several distinct product and delivery environments.
| Product or activity | Documented materials | Date and scope |
|---|---|---|
| Order Right | Microsoft Azure; customer-specific tenant provisioning | Current product FAQ. (2) |
| Product deployment and analytical warehousing | SQL procedures, database tuning, ETL; T-SQL/PL-SQL and SSIS among requested skills | Employer Database Programmer listing accessed November 2026; module unspecified. (16) |
| Real-time CDP engineering | Java/Scala services and AWS services including DynamoDB, SQS, S3, Lambda, and EC2 | Historical Algonomy recruiter post for the rCDP team. (17) |
| Data and AI consulting | Databricks on Azure, Delta Lake, Unity Catalog, Workflows, and vector search | Current service architecture, including customer data-platform projects. (9) |
| Active Content extensions | JavaScript, external HTTP APIs, and internal Data Model Extensions | Current CodeFusion developer documentation. (19, 20) |
The operational burden lies in the boundaries between these environments. Retail transactions must retain consistent product identities, units, promotion dates, and stock semantics as they move into analytics and replenishment. Each additional copy or transformation needs monitoring, recovery, and an owner when figures disagree.
The consulting stack introduces a separate data-platform lifecycle: ingestion, storage, access policies, compute orchestration, and model deployment. Its cost must be justified by the decisions improved. Databricks’ infrastructure and capabilities belong to Databricks; Algonomy’s engineering contribution must appear in its retail models, integrations, and measured operating improvements.
Public architecture material does not trace a complete promotion-to-order calculation across the customer-data and supply-chain products. A shared brand and exchange of data leave the critical question open: which component owns the economic decision when demand generation, inventory, and markdown priorities conflict?
Security: extension code needs enforced boundaries
The customer-profile API documents OAuth tokens with resource scopes and expiration. These are specific controls over data access. (18) Order Right’s tenant statement provides deployment context, but gives little detail about privileged support access, recovery after destructive changes, or isolation enforcement.
Active Content introduces a more concrete concern. Its CodeFusion editor lets users write JavaScript, access internal data models, and call external APIs. The creation guide shows a customer-profile retrieval example; the testing guide instructs users to inspect results and then save and publish. (19, 20, 21)
The documented network and data access makes containment essential. A mistaken script or compromised integration can disclose customer information even if it never changes a business record. The public guides do not explain enforced destination restrictions, separation of secrets, per-script data privileges, tenant execution isolation, resource limits, or independent production approval.
The POST API guide says its purpose is retrieval rather than modifying remote systems. It does not explain enforcement of that restriction against the destination service, and it exposes request/response logging features. (22) Remote services determine the effects of incoming requests; a stated intended use cannot substitute for access boundaries.
This concern applies to Active Content’s extension surface. The documentation does not establish that Order Right users execute JavaScript. It does, however, show why the broader suite’s programmability requires a stronger security account than testing instructions and user care.
Customer results and economic accountability
The Middle Eastern grocery case is unusually useful because it separates the pilot from the headline extrapolation. Three departments at one flagship store reportedly reduced inventory by 21% and out-of-stock occurrences by 9%. The $60 million headline is projected across more than 300 stores. (7)
That is a potentially valuable pilot, but the chainwide figure is not a realized return. The selected departments had experienced particularly large inventory increases, so a broader rollout needs evidence that the initial improvement generalizes. The account does not publish the observation period, financial baseline, implementation and operating costs, or a method separating software effects from other changes.
The same case dates readiness to two weeks after data integration. That excludes a potentially substantial part of implementation from the advertised interval. An assessment of time and cost must include data preparation and continued maintenance.
The multiformat retailer case reports markdown and transfer improvements alongside its undefined 95% accuracy. Moving inventory to stores with demand is a meaningful action, but the reported value of transferred stock is not profit; neither is estimated sales loss avoided. (23) Incremental margin must account for transfer expense, displaced sales, markdowns, and inventory that remains unsold.
Algonomy’s public accounts do not define a net economic failure condition, bounded customer obligations, or responsibility for correcting disappointing results. That makes the economic proposition difficult to falsify. It also prevents the published successes from supporting a reliable estimate of expected return across deployments.
Supply chain score
Supply chain depth
| Criterion | Score |
|---|---|
| Economic framing | 3/10 |
| Decision end-state | 6/10 |
| Conceptual sharpness on supply chain | 4/10 |
| Freedom from obsolete doctrinal centerpieces | 4/10 |
| Robustness against KPI theater | 1/10 |
Average: 3.6/10
Economic framing — 3/10. Waste and inventory costs feature in the proposition, but the order engine’s monetary objective is unpublished. Conflicting uses of cash, shelf space, and transport therefore lack an inspectable valuation rule. (3)
Decision end-state — 6/10. Purchase and transfer recommendations have an execution path into ERP. This is a practical operational output; public evidence does not establish how much approval and exception handling remains across deployments. (2)
Conceptual sharpness on supply chain — 4/10. Shelf life, pack constraints, promotions, and substitution receive specific treatment. Their treatment as inputs and features leaves the systemwide economic interactions unresolved. (3, 4)
Freedom from obsolete doctrinal centerpieces — 4/10. Automated order generation reduces dependence on spreadsheet planning. The central design still prioritizes forecasting followed by configured replenishment, with policy choices carrying much of the economic burden. (2, 25)
Robustness against KPI theater — 1/10. The undefined 97% and 95% accuracy claims obstruct meaningful evaluation. They cannot substantiate precision at the SKU-location ordering horizon or improved economic decisions. (5, 23)
Decision and optimization substance
| Criterion | Score |
|---|---|
| Probabilistic modeling depth | 2/10 |
| Distinctive optimization or ML substance | 4/10 |
| Real-world constraint handling | 6/10 |
| Decision automation and labor productivity | 6/10 |
| Resilience under real operational complexity | 4/10 |
Average: 4.4/10
Probabilistic modeling depth — 2/10. A consulting offer mentions probabilistic demand models. No public Order Right account establishes calibrated joint demand and supply outcomes feeding order valuation. (9, 3)
Distinctive optimization or ML substance — 4/10. Hierarchical modeling and automated model selection provide an identifiable forecasting approach. Candidate counts and a proprietary optimizer leave the distinctive numerical contribution difficult to assess. (2)
Real-world constraint handling — 6/10. The replenishment documentation covers several concrete retail restrictions. The unresolved issue is how the software trades off soft constraints and allocates common resources when individually feasible plans compete. (3)
Decision automation and labor productivity — 6/10. Order generation and Keells’ documented supplier workflows remove recurring work. The newer assistants lack measurements showing additional labor savings beyond that existing automation. (2, 15)
Resilience under real operational complexity — 4/10. Product lifecycle settings and configurable predictors provide adaptation paths. Changes to objectives, data meaning, or available actions remain dependent on vendor intervention without a public account of reproducible testing and deployment. (2)
Product and architecture integrity
| Criterion | Score |
|---|---|
| Architectural coherence | 4/10 |
| System-boundary clarity | 5/10 |
| Security seriousness | 3/10 |
| Software parsimony and ownership cost | 4/10 |
| Compatibility with programmatic and agent-assisted operations | 5/10 |
Average: 4.2/10
Architectural coherence — 4/10. Product-specific documentation identifies working components. Their different data and execution environments lack a published account of shared economic decision logic. (16, 17)
System-boundary clarity — 5/10. The order handoff and supplier transaction roles are understandable. Responsibility for data reconciliation and competing merchandising decisions across modules remains less explicit. (2, 14)
Security seriousness — 3/10. Scoped authentication is documented for customer data. Practitioner-authored extension code has a poorly explained containment boundary, while planning-specific recovery and privileged-access mechanisms remain opaque. (18, 19, 20, 21)
Software parsimony and ownership cost — 4/10. Deployment and service descriptions reveal substantial data engineering and configuration work. The public case material does not quantify that continuing work or justify its cost against incremental decision value. (2, 9)
Compatibility with programmatic and agent-assisted operations — 5/10. APIs, exports, and scripted commerce integrations provide automation surfaces. The planning engine’s objectives and revision process remain much less accessible for programmatic improvement. (18, 20, 24)
Technical transparency
| Criterion | Score |
|---|---|
| Public technical documentation | 5/10 |
| Inspectability without vendor mediation | 4/10 |
| Portability and lock-in visibility | 4/10 |
| Implementation-method transparency | 5/10 |
| Evidence density behind technical claims | 3/10 |
Average: 4.2/10
Public technical documentation — 5/10. Customer-data and Active Content APIs are substantially documented. Supply-chain methods receive product FAQs and datasheets rather than comparable numerical documentation. (18, 2, 3)
Inspectability without vendor mediation — 4/10. Integration behavior can be examined publicly. Order valuation, uncertainty calibration, and algorithm changes cannot be independently reproduced from the published material. (2, 18)
Portability and lock-in visibility — 4/10. Order exports and documented customer-data destinations facilitate moving records. Exporting the learned replenishment logic, configuration dependencies, and complete decision history is not similarly explained. (24)
Implementation-method transparency — 5/10. Product documentation and customer accounts describe data integration and workflow changes. They provide less visibility into continuing maintenance, acceptance tests, and responsibility for remediation. (2, 15)
Evidence density behind technical claims — 3/10. Operational examples support specific capabilities. The most ambitious accuracy and intelligence assertions carry little experimental evidence. (6, 7, 5)
Vendor seriousness
| Criterion | Score |
|---|---|
| Technical seriousness of public communication | 3/10 |
| Technology roadmap integrity | 3/10 |
| Conceptual sharpness | 3/10 |
| Economic accountability | 3/10 |
| Defensibility in an agentic-software world | 4/10 |
Average: 3.2/10
Technical seriousness of public communication — 3/10. Useful product descriptions coexist with undefined accuracy, algorithm counts, and unexplained generative forecasting claims. Those claims prevent careful evaluation of the product’s central proposition. (5, 11, 25)
Technology roadmap integrity — 3/10. The dated record establishes delivered replenishment and supplier products. Successive conversational-intelligence promises lack a measured account of how capability and productivity advanced. (10, 13, 14, 15)
Conceptual sharpness — 3/10. The public narrative repeatedly treats predictive precision, next-action suggestions, and economic optimization as interchangeable. Their different requirements are central to a dependable purchasing engine. (13, 25)
Economic accountability — 3/10. Customer cases report changes in operational measures, but net returns and failure conditions remain undefined. Pilot projections and transferred-stock values cannot discharge the broader economic promise. (7, 23)
Defensibility in an agentic-software world — 4/10. Installed integrations and domain workflows retain value. A conversational interface is increasingly easy to reproduce; durable differentiation requires superior decisions and cheaper maintenance of the logic that produces them. (14, 15)
Sources
All sources accessed November 4, 2026. Publication dates are stated where available; undated documentation reflects the public version reviewed.
ADA acquires Algonomy. 2026-07-30. Completed acquisition and ADA branding.
Order Right FAQs. undated; accessed 2026-11-04. Forecasting, execution paths, configuration, support, and deployment.
Order Right datasheet. 2022-06 URL directory. Forecast-driven replenishment, constraints, risk selection, and waste criteria.
Order Right for grocery. undated. Grocery scheduling, promotional effects, and alert configuration.
[5] Demand forecasting with hyperlocal precision
Demand forecasting with hyperlocal precision. undated. Undefined 97% accuracy claim and forecasting workflow.
[6] Southeast Asian grocery retailer
Southeast Asian grocery retailer. undated. Coverage of forecast improvements and multiechelon replenishment case.
[7] Middle Eastern grocery retailer projects $60M inventory savings
Middle Eastern grocery retailer projects $60M inventory savings. undated. Single-store pilot, reported improvements, and chainwide projection.
[8] M5 competition methods and submissions
M5 competition methods and submissions. fixed repository commit. Organizers’ forecast submissions used for the aggregation comparison.
[9] Data and AI decision intelligence services
Data and AI decision intelligence services. undated. Consulting-specific probabilistic forecasting and Databricks architecture.
Manthan unveils Maya. 2017-01-16. Contemporaneous Maya announcement and Alexa interface.
Algonomy launch. 2021-01-19. Merger completion, platform scope, and algorithm claims.
[12] Intent to acquire Linear Squared business
Intent to acquire Linear Squared business. 2022-01-05. Proposed acquisition and intended forecasting expansion.
Supply Planning and Maya. undated. Current Maya and merchandising workflow.
[14] Supplier Collaboration and ATLAS
Supplier Collaboration and ATLAS. undated. Vendor Link and ATLAS process claims.
[15] Keells supplier collaboration case study
Keells supplier collaboration case study. 2024-06 URL directory. Named deployment, supplier administration, and workflow changes.
[16] Algonomy Database Programmer recruitment
Algonomy Database Programmer recruitment. undated; accessed 2026-11-04. Employer requirements for analytical warehousing and product deployment.
[17] Algonomy rCDP engineering recruitment
Algonomy rCDP engineering recruitment. historical recruiter post. Team-specific historical cloud and software recruitment evidence.
[18] Customer profile API overview
Customer profile API overview. current documentation. Customer-profile authorization scopes and token behavior.
[19] Creating CodeFusion Script
Creating CodeFusion Script. current documentation. User-authored JavaScript and publication workflow.
[20] Developing CodeFusion Script
Developing CodeFusion Script. current documentation. Internal data access and external API capabilities.
[21] Testing CodeFusion Script
Testing CodeFusion Script. current documentation. Manual test, save, and publication steps.
CodeFusion POST API. current documentation. Declared retrieval-only purpose and logging behavior.
[23] Multiformat retailer merchandise planning
Multiformat retailer merchandise planning. undated. Accuracy, markdown, and inventory-transfer claims.
[24] Amazon S3 destination settings
Amazon S3 destination settings. current documentation. Data export destination configuration.
[25] Current demand forecasting and replenishment
Current demand forecasting and replenishment. undated. Current generative-AI label and replenishment workflow.