Last updated: November 4, 2026
Go back to Market Research
Colibri supplies cloud software for demand planning, replenishment, and sales and operations planning. Its VISION, FLOW, and PILOTE modules support forecasts, purchasing and distribution plans, capacity scenarios, and financial projections. The company belongs to the VISEO group. (1, 18)
Overall score: 3.8/10
| Dimension | Score |
|---|---|
| Supply chain depth | 3.8/10 |
| Decision and optimization substance | 3.6/10 |
| Product and architecture integrity | 4.4/10 |
| Technical transparency | 3.8/10 |
| Vendor seriousness | 3.4/10 |
Colibri documents a conventional replenishment process with a substantial planner workload. Users configure product groups and logistics rules; scheduled calculations generate plans and alerts; planners revise and approve suggestions before exporting orders to an ERP. The software automates calculations within that process, while people maintain its policies and decide how to resolve exceptions. (2)
The AI claims are considerably broader than the demonstrated decision mechanisms. Super Best Fit promises the best possible forecasts, Data Sensing adds external variables, and assistants learn or execute user actions. Public evidence does not establish a joint model of demand and supply uncertainty, an economic objective governing operational decisions, or the total labor eliminated by the new features. (4, 1)
The most material questions concern the policies that drive purchases, the recurring cost of keeping them correct, and responsibility when the resulting decisions lose money. Customer cases establish deployed workflows and some forecasting improvements. They do not establish the stronger proposition of economically optimized supply chain decisions running with little human intervention.
Colibri versus Lokad
Colibri organizes the planning work of sales teams, buyers, and supply chain managers around shared forecasts, policy settings, alerts, and scenarios. This can replace fragile spreadsheets and make information easier to exchange. It also preserves a recurring role for people who interpret the forecast and approve the resulting plan.
Lokad’s quantitative supply chain approach evaluates operational decisions by their expected financial consequences. Purchases, allocations, production, and other actions are considered against uncertain demand, replenishment delays, and the economic costs of shortages, excess stock, and constrained resources. The objective is to run decision logic repeatedly and improve it when experience reveals a missing factor. Introduction to Supply Chain develops the underlying perspective.
For example, classifying a product as important does not determine how much capital should be committed to it. A low-volume spare part can justify more protection than a fast-moving accessory if its shortage interrupts an expensive operation. A policy based on sales classification must be supplemented with the relevant costs and uncertainty before it can make that trade-off.
The service boundary also matters. Colibri documents implementation and ongoing support, while planners own several policy settings. Lokad’s managed inventory optimization assigns a Supply Chain Scientist to data preparation, decision economics, monitoring, and production maintenance. Both incur maintenance costs; the useful comparison is the total cost and financial result of operating and revising their decision logic.
The replenishment mechanism and its economic limits
Forecasts become configurable policies and planner approvals
Colibri’s five-step supply-planning description identifies the normal path into execution. Planning groups combine products and locations. A planner sets safety stocks, reorder points, calculation rules, horizons, and calendars. Calculations run daily or weekly, producing suggestions and alerts. The planner then modifies plans, contacts suppliers, creates or validates purchase suggestions, and exports approved orders to the ERP. (2)
This is a concrete operating model. Its economic weakness is the number of decisions displaced into configuration. A safety-stock level determines which shortages are tolerated; a planning horizon determines which future costs are visible; a grouping determines where distinct products receive similar treatment. Unless these choices are calculated from financial consequences, apparently reasonable settings can systematically favor the wrong inventory.
Colibri also documents supplier and container planning, multiple suppliers, and capacity calculations. (17) These are relevant constraints. The public material does not identify how the system evaluates a combined purchase when a container threshold, substitution, uncertain arrivals, and competing uses of capital interact.
Safety-stock automation preserves the policy problem
The safety-stock article describes ABC categorization, decreasing ratios applied across references, and dynamic stocks reflecting product life cycles and variability. IZIPIZI uses VISION and FLOW for its forecasts and stock projections. (3)
The exposed mechanism adjusts inventory policy. It does not establish how a unit of additional stock is valued against alternative uses of money. Two items can have similar sales variability and very different shortage losses, substitution possibilities, or obsolescence risks. Smoothing a classification into finer ratios improves the continuity of the policy; it does not supply the missing economics.
Supply uncertainty is equally consequential. A forecast of sales plus a fixed lead time can underestimate exposure precisely when a supplier becomes unreliable. The reviewed product explanations do not show a joint probability model of demand, delivery time, and other relevant uncertainties entering a monetary decision calculation.
Super Best Fit and Data Sensing
The March 2025 release describes Super Best Fit as selecting statistical, machine-learning, or deep-learning algorithms. It promises the best models and forecasts without effort from the user. Data Sensing incorporates variables such as weather, promotions, prices, and competitors’ shortages through correlation analysis. (4)
Automatic model selection can remove model-selection work. Its quality depends on the candidate methods, training data, evaluation windows, loss function, and information available when the forecast was issued. None of those details is supplied with the best-possible guarantee.
The sensing label adds no distinct technical substance. A competitor’s shortage may influence demand, but its observation must arrive before the purchasing decision and remain useful over the replenishment horizon. Correlation on revised or late-arriving data cannot establish that benefit. The relevant comparison uses the same information cutoff and refresh schedule, then measures the financial consequences of the resulting orders.
Historical promises and delivery
| Period | Public evidence | What followed |
|---|---|---|
| 2013 development origin | A historical recruitment page says development began on Azure inside VISEO. | The product developed as a cloud forecasting and replenishment application. (10) |
| 2019 | Vitacuire described a deployed collaborative forecasting process. | Sales staff, account managers, and a planning manager retained distinct preparation and validation roles. (8) |
| 2021–2022 | Colibri documented safety-stock policies and reported a Vitacuire forecasting improvement. | The evidence concerned policy configuration and forecasts; no net economic result was published with the improvement. (3, 9) |
| March 2025 | Super Best Fit, external-variable analysis, and automation were announced as available; assistants were forthcoming. | The current website markets conversational assistance, application control, and learning of user actions. (4, 1) |
| December 2025–2026 | Colibri expanded its regional implementation/support organization and reported commercial growth. | This establishes continued product and service activity, while the incremental decision-labor benefit of the AI features remains unmeasured. (14, 15) |
The history shows continuity in the underlying work: improve forecasts, configure planning policies, coordinate users, and validate plans. The newer assistant positioning does not demonstrate that this operating model has changed. Learning a user’s repeated actions can reproduce a workflow without establishing whether those actions improve the economics.
Colibri’s own 2026 technical commentary creates a sharper inconsistency. Its February article warns that advanced ML requires specialist analysis and configuration, can overfit, and does not benefit every business equally. (5) Those qualifications are sensible, but they leave the earlier effortless best-possible forecast promise unsupported. The vendor needs to identify which work the selector eliminates and which expert work remains.
The February article, repeated in a March version, also describes a 2020 forecasting competition in which 17 ML methods supposedly fell below a naïve baseline, followed by a breakthrough a year later. (5, 6) The organizers’ M5 paper records the competition in March–June 2020 and reports that all 50 leading methods used ML. (7) Colibri’s chronology misrepresents the benchmark history it invokes to explain its technology. Benchmark results deserve precise attribution, especially when they support a product’s account of recent progress.
Productivity, configuration maintenance, and ownership cost
The customer must maintain a planning application
Planning groups, logistics rules, alert thresholds, calendars, and supplier settings are executable business logic. Someone must revise them when demand patterns, procurement options, product substitutions, or service costs change. A configurable interface gives that person access to the settings; it does not establish that the change is economically sound.
Colibri offers prototypes using customer data and iterative configuration. (19) Its December 2025 service announcement describes support through advice, setup, launch, and subsequent evolution, with regional product and deployment specialists. (14) These commitments address an identifiable need: keeping the application useful after its first deployment.
The recurring expense comprises both this specialist work and the planners’ operational reviews. They solve different problems. One changes the decision logic; the other repeatedly intervenes in its output. If a recurring exception always requires the same judgment, a lasting productivity gain requires that judgment to become reliable, tested logic.
The public material does not explain how customers version policy changes, replay them against historical decisions, compare financial outcomes, or roll back an incorrect configuration. These omissions make the cost of safely evolving the application difficult to assess.
The customer evidence measures only part of the benefit
Vitacuire’s 2019 account describes forecasts feeding production and purchasing requirements, with several human validation roles. (8) An April 2022 follow-up reports forecast reliability increasing from 50–60% to more than 70% over two years, promoted as a 20% improvement. It does not define the error formula, aggregation, forecast origin, or whether the improvement is relative or measured in percentage points. (9)
The numbers cannot establish how much stock, waste, or margin changed. Seasonal manufacturing forecasts also cannot be compared directly with daily retail item forecasts. The evaluation needs the decisions, their horizon, and their financial consequences.
Other customer accounts on the current homepage describe monthly forecast review at GGB and support for changing substitution dates at Puressentiel. (1) These provide evidence of useful deployment and subsequent adaptation. They also identify continuing review and maintenance work. No published before-and-after labor measurement separates those workloads from the newer assistants’ contribution.
Colibri reports €3.9 million revenue and 19 new customers for 2025. (15) This establishes commercial activity. It does not measure customer returns or resolve the missing productivity evidence.
Software materials and security
A Microsoft application stack, with important details undisclosed
| Component | Product-specific evidence | Operating implication |
|---|---|---|
| Microsoft Azure | Colibri identifies Azure as its hosting platform and longstanding technical partner. | Compute, storage, backup, and service operations remain part of the recurring cost. (11) |
| C#/.NET and SQL Server | A historical Colibri-suite development vacancy requests these skills. | It identifies a Microsoft application and database development lineage. (10) |
| jQuery, AngularJS, and VSTS | The same undated recruitment page names the web and development tools. | These are historical product-team requirements; the current replacement and support status is not published. (10) |
| Salesforce add-on | Colibri’s partner page describes an integrated add-on for sales users. | Sales inputs cross an additional application boundary whose timing and mappings require maintenance. (11) |
| ERP integration | The supply workflow exports approved orders; customer accounts describe forecast transfer into ERP planning. | Correct quantities, identifiers, approval states, and retries must survive the handoff. (2, 8) |
The documented software stack is comparatively compact. The more visible cost lies in configured planning policies, integration semantics, specialist support, and repeated review.
The implementation details still matter. Public material does not expose transaction handling for failed order exports, end-to-end replay, or the boundaries between forecasting computation and operational storage. Those are material gaps when a planning system is expected to feed purchases reliably.
Security claims stop short of failure containment
Colibri’s published security description lists HTTPS, OAuth2 authentication, optional two-factor authentication, customer-data isolation, external testing, and data rollback within 30 days. (12)
The rollback horizon is relevant, but the description does not establish which state is recoverable, who can alter backups, how restoration is tested, or how a corrupted configuration is detected before it generates bad orders. Likewise, naming isolation does not expose the privilege boundary that prevents one compromised account or service from accessing other customers’ data.
The engineering test is what remains protected after an account, dependency, or administrator fails. Public security material provides too little detail on enforced privileges, protected recovery state, and patching behavior to support a strong assessment. Hosting and audit credentials do not answer those questions.
Economic accountability: the forecast promise and service obligation diverge
The published English service terms disclaim guarantees concerning the accuracy or reliability of results and suitability for particular expectations. They undertake to correct major defects preventing use when attributable exclusively to Colibri, cap direct liability at the previous twelve months’ fees, and exclude specified losses including profit and turnover. The text states that the French version prevails. (13)
These public terms frame responsibility around access and software defects. They do not turn the forecast-performance promise into a defined economic obligation. A functioning application can still produce purchases that tie up capital or cause shortages.
The gap should be closed before attributing a failed outcome to customer competence. Implementation must identify data and staffing prerequisites, define how the policy will be evaluated, and assign responsibility for diagnosing forecast errors, policy errors, overrides, and execution failures. The reviewed customer material does not define a net-return baseline, failure threshold, or remediation commitment.
Supply chain score
The overall score is the equally weighted mean of the 25 criteria below.
Supply chain depth
| Criterion | Score |
|---|---|
| Economic framing | 4/10 |
| Decision end-state | 5/10 |
| Conceptual sharpness on supply chain | 4/10 |
| Freedom from obsolete doctrinal centerpieces | 2/10 |
| Robustness against KPI theater | 4/10 |
Average: 3.8/10
Economic framing — 4/10. Inventory, cash, service, and capacity are visible business concerns. The documented planning rules do not expose a monetary objective that selects among competing actions under uncertainty. (17)
Decision end-state — 5/10. The replenishment process ends in approved order exports to an ERP. That establishes an operational output, while retaining planner validation and leaving the financial quality of the configured policy unresolved. (2)
Conceptual sharpness on supply chain — 4/10. Product life cycles, containers, suppliers, and distribution are addressed. The public explanation does not show how their interactions change the preferred decision when demand and delivery outcomes vary. (17)
Freedom from obsolete doctrinal centerpieces — 2/10. A shared forecast, planning classifications, exception reviews, and S&OP coordination remain the center of the design. The white paper explicitly promotes a single shared future for organizational alignment. (16) This leaves uncertainty and decision selection outside the principal organizing concept.
Robustness against KPI theater — 4/10. Vitacuire’s before-and-after figures supply some context but no reproducible metric or economic bridge. The best-possible forecast promise is substantially stronger than the evidence used to support it. (9, 4)
Decision and optimization substance
| Criterion | Score |
|---|---|
| Probabilistic modeling depth | 2/10 |
| Distinctive optimization or ML substance | 3/10 |
| Real-world constraint handling | 5/10 |
| Decision automation and labor productivity | 4/10 |
| Resilience under real operational complexity | 4/10 |
Average: 3.6/10
Probabilistic modeling depth — 2/10. The available explanations concern forecast selection, variability, and inventory rules. No documented joint probability model connects uncertain demand and supply to financial decision evaluation. (3, 4)
Distinctive optimization or ML substance — 3/10. Model selection and external-variable analysis identify computational tasks. Public descriptions do not expose the candidate models, validation procedure, decision objective, or optimization mechanism needed to establish the stronger claims. (4, 5)
Real-world constraint handling — 5/10. Supplier, container, calendar, and capacity features address concrete planning restrictions. How they interact in a shared economic calculation remains unclear, particularly when the future violates their deterministic inputs. (17)
Decision automation and labor productivity — 4/10. Scheduled replenishment calculations remove arithmetic work and produce order suggestions. Published outcomes do not establish total labor savings after exceptions, maintenance, and support, or an incremental gain attributable to assistants. (2, 1)
Resilience under real operational complexity — 4/10. Customer feedback and implementation services support application changes. The public record does not show a systematic method for converting costly operational failures into tested improvements to objectives, constraints, and available actions. (19, 14)
Product and architecture integrity
| Criterion | Score |
|---|---|
| Architectural coherence | 5/10 |
| System-boundary clarity | 5/10 |
| Security seriousness | 3/10 |
| Software parsimony and ownership cost | 5/10 |
| Compatibility with programmatic and agent-assisted operations | 4/10 |
Average: 4.4/10
Architectural coherence — 5/10. The three modules cover related stages of a planning process and share a stated cloud offering. Public material does not expose enough internal execution detail to assess consistency across forecasting, replenishment, and strategic scenarios. (1, 11)
System-boundary clarity — 5/10. Cloud hosting, CRM access, and ERP exports have identifiable roles. Transaction guarantees, reconciliation after partial failure, and ownership of integration faults remain insufficiently specified. (11, 2)
Security seriousness — 3/10. The security description names controls and a recovery horizon without explaining their enforced boundaries or tested failure behavior. That leaves damage containment difficult to assess. (12)
Software parsimony and ownership cost — 5/10. The product-specific materials suggest a relatively compact application lineage. Recurring policy maintenance, integrations, and specialist support are visible costs; no customer evidence measures their total against the benefits. (10, 14)
Compatibility with programmatic and agent-assisted operations — 4/10. Scheduled calculations and ERP exchange support automation. Publicly inspectable execution and configuration interfaces are too limited to establish dependable agent-driven application maintenance or unattended economic decisions. (2)
Technical transparency
| Criterion | Score |
|---|---|
| Public technical documentation | 4/10 |
| Inspectability without vendor mediation | 3/10 |
| Portability and lock-in visibility | 4/10 |
| Implementation-method transparency | 5/10 |
| Evidence density behind technical claims | 3/10 |
Average: 3.8/10
Public technical documentation — 4/10. The operating sequence and some policy mechanisms are documented. Algorithm definitions, uncertainty models, runtime architecture, and safe configuration-promotion methods remain sparse. (2, 3)
Inspectability without vendor mediation — 3/10. Product articles expose what users do, while detailed numerical evaluation still depends on a vendor engagement. The claimed avoidance of a black box is not matched by a reproducible forecast or decision example. (5, 4)
Portability and lock-in visibility — 4/10. Order exports provide one exit path. Published terms describe a limited data-extraction period after eligible terminations, at customer expense. (13) Portability of configured policies and application behavior remains unexplained.
Implementation-method transparency — 5/10. Prototyping, planner settings, and regional support responsibilities are described. Ongoing regression testing and maintenance workload are insufficiently documented to evaluate lifecycle cost. (19, 14)
Evidence density behind technical claims — 3/10. Customer cases substantiate use of the application and selected workflow changes. They do not validate best-possible forecasts, joint optimization, or the economic contribution of the current AI features. (8, 9, 4)
Vendor seriousness
| Criterion | Score |
|---|---|
| Technical seriousness of public communication | 3/10 |
| Technology roadmap integrity | 4/10 |
| Conceptual sharpness | 4/10 |
| Economic accountability | 3/10 |
| Defensibility in an agentic-software world | 3/10 |
Average: 3.4/10
Technical seriousness of public communication — 3/10. The forecast guarantee outruns its technical support, and the vendor’s benchmark chronology is incorrect. Its more careful discussion of ML limitations has not reconciled those claims. (4, 5, 6, 7)
Technology roadmap integrity — 4/10. Deployed planning workflows and continued releases establish delivery. The transition from policy automation to assistants lacks measured evidence of a corresponding reduction in decision labor or improvement in economic results. (8, 4, 15)
Conceptual sharpness — 4/10. Colibri recognizes configuration needs and the limits of advanced forecasting. It still treats forecast improvement, user-action automation, and optimized supply chain outcomes as more closely connected than the demonstrated mechanisms justify. (5, 2)
Economic accountability — 3/10. Published obligations address software access and serious defects while disclaiming result accuracy. The public proposition does not define an economic failure or a bounded responsibility for fixing the decision logic that caused it. (13)
Defensibility in an agentic-software world — 3/10. Replacing spreadsheets and helping planners review alerts address longstanding administrative work. The current evidence does not establish the larger productivity shift toward improving and running decision logic across the network with minimal recurring review. (1, 2)
Sources
Sources accessed November 4, 2026.
Colibri product overview. Current. Product modules, current assistant claims, and customer descriptions of review and adaptation.
[2] Supply planning: five operating steps
Supply planning: five operating steps. 2026 page. Planner configuration, scheduled calculations, alerts, approval, and ERP order export.
[3] Safety-stock policy and IZIPIZI
Safety-stock policy and IZIPIZI. 2021-11-22; current text. Published safety-stock policy and IZIPIZI deployment.
[4] Super Best Fit, Data Sensing, and assistants
Super Best Fit, Data Sensing, and assistants. 2025-03-27. Original vendor release; available modules and forthcoming assistants are distinguished.
[5] Machine learning: value or marketing?
Machine learning: value or marketing?. 2026-02-02. Specialist configuration requirements, ML limitations, and benchmark chronology.
[6] Machine learning: revised article
Machine learning: revised article. 2026-03-16. Subsequent article repeating the benchmark chronology.
[7] M5 accuracy competition results
M5 accuracy competition results. 2022 journal; competition 2020. Organizers’ research paper identifying the 2020 competition and leading methods.
Vitacuire deployment. 2019-10-03. Original deployed workflow and human validation responsibilities.
[9] Vitacuire two-year results
Vitacuire two-year results. 2022-04-04. Two-year follow-up with forecast-reliability figures.
Developer Web App .NET. Undated historical recruitment. Historical product-development vacancy; current deployment versions are unspecified.
Technical partners. Current. Azure hosting and the Salesforce add-on.
Application security. Undated. Published control summary and recovery horizon.
Published service terms. Undated English terms. English service conditions, result disclaimers, liability scope, and data extraction.
[14] Regional service organization
Regional service organization. 2025-12-11. Implementation, configuration, and ongoing service organization.
2025 commercial results. 2026-01-20. Vendor-reported revenue and new-customer activity.
[16] Why do sales and operations planning?
Why do sales and operations planning?. 2022-era white paper. Organizational planning doctrine and shared-forecast framing.
[17] Cloud planning product functions
Cloud planning product functions. 2023-era article. Supplier/container planning, safety stocks, capacity, and financial scenarios.
Company ownership notice. 2024-04-25. VISEO affiliation.
[19] Configuring supply chain processes
Configuring supply chain processes. 2021-era article. Customer-data prototypes and iterative configuration.