Go back to Market Research
GMDH Inc. develops Streamline, a supply chain planning application for manufacturers, distributors, and retailers. It combines demand forecasting, replenishment, material requirements planning, supplier management, and production planning, with connections to enterprise systems for importing data and returning orders. The current product is primarily a cloud web application, with Linux deployment available for customers hosting it themselves. 1 2
Overall supply chain score: 4.9/10
| Dimension | Score |
|---|---|
| Supply chain depth | 4.4/10 |
| Decision and optimization substance | 4.4/10 |
| Product and architecture integrity | 5.0/10 |
| Technical transparency | 6.2/10 |
| Vendor seriousness | 4.4/10 |
Streamline has an identifiable planning engine, operational integrations, detailed public manuals, and customer evidence of inventory and administrative improvements. Its documentation supports scrutiny of the calculations and their limitations.
Its principal limitation is the economic model. Statistical forecasts feed safety stocks, ordering cycles, supplier rules, and approval workflows. The software accounts for several operational dependencies, but the published methods do not establish a joint selection of decisions by expected financial return under uncertain demand and supply.
The autonomous and agentic positioning goes further than this evidence. A connector to an external AI assistant does not improve the underlying replenishment model, and Streamline’s own manual explicitly says so. The most revealing gaps concern who maintains the planning assumptions, how simulations are validated, and who carries the consequences when recommendations fail.
Streamline versus Lokad
Streamline organizes planning around forecasts, replenishment parameters, and orders reviewed through a packaged application. Lokad’s inventory optimization service organizes it around recurring decisions ranked by expected financial contribution, with Supply Chain Scientists maintaining the economic logic.
The difference appears when resources are scarce. A service target may justify buying more of an item, while another purchase would yield greater expected margin from the same cash or warehouse space. Filling each item’s buffer, even with sophisticated forecasting, does not resolve that competition. Nor does filling a container proportionally to demand establish which mix is financially preferable.
Streamline provides substantial visibility into its policies and supports automated calculations and exports. The remaining question is whether the customer can improve those policies when the original assumptions stop matching the business, without recreating an extensive manual review process.
Lokad’s perspective, developed in Introduction to Supply Chain, treats this continuing reformulation as part of the service. The relevant comparison is the quality of decisions and the total labor needed to sustain them, including modeling, integration, and remediation.
The decision engine
Statistical forecasting is the documented AI core
The forecasting manual describes time-series decomposition into level, trend, seasonality, and adjustments, with an expert system selecting model components. It also describes handling stockout-affected history, choosing a relevant history window, and intermittent demand. 3 These are specific forecasting mechanisms, although the public material does not provide a reproducible benchmark establishing superiority over alternative methods.
The explanation also dismisses model competition as inherently prone to overfitting. That is too broad: validation design, model complexity, and the stability of the data determine whether selection works. Calling one approach human-like and another outdated does not settle the comparison.
The displayed uncertainty bands deserve separate scrutiny. The confidence-level manual derives them from squared residuals on the learning data, using a symmetric interval around each forecast. 6 Training fit is not an out-of-sample calibration test. Stable historical residuals do not establish the probability of future shortages after a promotion, a supplier disruption, or a change in customer behavior.
Supply uncertainty is recognized, but its maintenance remains external
For ordinary demand models, the safety-stock formula combines forecast residual variance with an imported lead-time variance, scaled by a service-level quantile. Missing lead-time variance is treated as zero. The distribution-center formula combines local protections through a square-root-of-summed-squares calculation. 4
This addresses more than demand uncertainty alone, but the published formula has no covariance term for disturbances shared across locations. A common promotion or delivery disruption can produce simultaneous shortages that a calculation that omits this dependence understates.
The current Suppliers manual makes an important distinction: the software calculates historical average lead time and variance for comparison, while planning continues to use imported or manually overridden parameters. Users are instructed to update the source system or override those values when the observed behavior differs. 5
The consequential gap is therefore between observation and decision. A supplier report can correctly identify deteriorating reliability while tomorrow’s recommendations still use obsolete parameters. The cost of detecting, approving, and propagating the change belongs in the ownership cost of the product.
A historical variance is also only a compressed description of supply risk. It does not retain partial-delivery behavior, dependence between delays and demand, or the different consequences of alternative suppliers. Those distinctions matter when deciding whether to expedite, split an order, substitute a product, or accept a shortage.
Order generation has operational substance and economic limits
The ordering manual specifies a cycle-based calculation: forecast consumption, safety stock, existing inventory, and scheduled movements determine the required quantity, followed by minimum-quantity and rounding adjustments. 7 The resulting outputs are dated purchases and transfers that can feed enterprise systems.
Streamline also supports constraints shared across products. The containers-and-groups reference documents supplier quantity and cost limits, weight and volume restrictions, pallet capacities, and manufacturing groups. Container filling is described as proportional to item demand. 8
These mechanisms make the software more than a collection of isolated forecast charts. Their economic limitations remain substantial. Demand volume is not the same as expected marginal profit. Proportionally filling a container can favor a low-margin, slow-to-liquidate item over a smaller but financially superior purchase.
Service targets carry a similar problem. The configuration guide asks users to maintain order cycles, lead times, safety-stock methods, and target availability. 34 These settings can produce an executable policy without establishing that its financial trade-offs are sound.
The future plan depends on the calculation path
The ordinary ordering plan applies group constraints to the first order, with only individual constraints applied to later orders. A separate constrained-plan calculation applies the group constraints across the horizon. Its documentation warns that larger calculations can take time and recommends overnight scheduling; projected inventory and shortage reports must be configured to use this result. 9
This distinction matters for the claimed year-ahead visibility. Two reports can portray different futures depending on whether the relevant constraints were propagated. A rapid forecast of stock levels is of limited value if the receipts behind it could not actually be ordered.
Capacity has another explicit boundary. The rough-cut capacity page compares manufacturing load against available capacity but says that overload resolution requires manual action. 10 A separate production-scheduling module does allocate orders to workstations from a constrained plan, using process steps, throughput, and shifts. It currently excludes process steps extending beyond one day. 11
Workstation scheduling addresses feasibility within these restrictions. The public methods do not establish joint selection of procurement, manufacturing, and inventory decisions under uncertain durations and demand.
Simulation and forecast accuracy
A simulated supply chain still needs predictive validation
Time Machine executes recommendations against a simulated enterprise system and can add white noise based on past forecast error. Its manual says this closely resembles future MAPE. 12 The October 2026 webinar presents digital twins and this simulation as a way to test demand changes and examine future orders and KPIs. 28
The missing step is validation of the simulated operational outcomes. A demand-error distribution does not establish the accuracy of projected inventory, expirations, service failures, or costs. Those states also depend on supplier behavior, execution delays, substitutions, order constraints, and the policy’s reactions to earlier events.
White noise is particularly limited when disturbances persist or affect multiple items together. Simulating independent surprises around the historical plan can produce reassuring results while omitting a common supplier failure or a sustained demand shift.
The relevant comparison is between predicted and realized outcomes at the level where decisions occur. It should also test policy changes: a model that reproduces the old policy’s stock trajectory may fail when purchasing frequency or supplier choice changes. Public materials do not provide that validation for the digital-twin claim.
The accuracy reports expose exclusions that headlines should preserve
The forecast-analysis manual documents saved forecast versions, several error measures, and an accuracy view calculated as 100% minus the selected error, floored at zero. It leaves percentage-error cells blank when sales are zero and leaves a row blank for an intermittent model or an inactive item. Aggregation and revenue-based weighting are configurable. 13
This makes the metric inspectable, but a portfolio accuracy figure must preserve its exclusions. Omitting intermittent items can remove precisely the products for which replenishment is difficult. Revenue weighting can conceal many low-volume errors, and longer aggregation periods can smooth the timing errors that create stockouts.
GMDH’s homepage advertises inventory availability above 99%, not 99% forecast accuracy. 32 These measures must remain distinct. High availability can be purchased with excessive inventory; it is not a demonstration that demand was predicted almost perfectly or that the policy maximized profit.
A credible result needs the product-location granularity, horizon, evaluation period, treatment of zeros and stockouts, and comparison against an operational baseline. The financial outcome must then be measured separately.
Productivity and the agentic claim
Streamline’s purchase-order guide describes a concrete review-and-execution workflow. Users inspect suggestions, change quantities or prices, submit them for approval or create orders directly, and synchronize the resulting transactions with the enterprise system. It warns that order creation is not reversible in that action. 14
Automated data refreshes, plan calculations, and exports can remove significant clerical work. They do not by themselves eliminate recurring judgment at the item level. The productive question is how much work remains after deployment, including reviewing exceptions, maintaining parameters, resolving failed integrations, and investigating economically poor recommendations.
The current MCP manual explicitly states that external reasoning does not change Streamline’s forecasting or calculation engine. The connector exposes supported actions within user permissions, but cannot run data updates or edit integrations. It remains a prerelease testing service. 15
This is a red flag when used to support a broad agentic-planning narrative. Providing conversational access to existing screens and actions is inexpensive compared with redesigning the decision process. It can preserve the planner’s old sequence of SKU investigations while making each interaction easier.
An external agent could perform a useful predefined workflow. The missing evidence is sustained production performance: fewer recurring interventions, financially better decisions, and controlled failure recovery. A connector specification does not supply those results.
The stronger test follows Holimization. When a recurring shortage exposes a missing dependency, the system should help identify the wrong assumption, revise the policy, and retest the affected decisions. Repeatedly adjusting forecasts or buffers can hide the defect while increasing the labor needed to keep the application operating.
Customer results and economic accountability
| Customer | Published evidence | Economic interpretation |
|---|---|---|
| Whalen | The case reports 36% lower inventory, elimination of two rented auxiliary buildings, and about $56,000 monthly storage savings. A spreadsheet update taking a day and a half for 60 items became a seconds-long update; the employee then spent time analyzing the data. 29 | The released warehouse expense is concrete. The updating improvement does not measure the complete planning workload or establish elimination of the employee’s work. |
| Transgold | After more than a year, a customer quotation reports 5–10% lower stockholding with minimal fill-rate impact and a substantially faster purchasing process. Implementation took about six weeks and included a one-way Micronet connector. 30 | A modest, operationally plausible deployment result. The published account does not reconcile implementation, subscription, and continuing labor with the gain. |
| Genomma Lab | The headline presents 40–50% inventory reduction. The body instead describes a materials-inventory reduction opportunity over the following four to six months, after a project spanning October 2019–March 2020. 31 | A projected opportunity is being presented as an achieved outcome. The page does not provide the subsequent measurement needed to establish the headline. |
These cases support real deployment and some meaningful improvements. They also show why a single percentage is insufficient: storage expense, stock value, employee capacity, and future opportunity are different quantities.
GMDH’s current homepage promises 100% ROI in the first three months, without publishing a common measurement method or customer-level reconciliation behind that statement. 32 Inventory released cannot simply be counted as recurring profit. Nor can time saved on one task be treated as payroll removed.
The MCP terms provide a particularly clear allocation of responsibility for the new interface. Effective September 2, 2026, they assign permissions and third-party-tool choices to the customer, disclaim result accuracy and reliability, cap liability at $100, and restrict disclosure of non-public testing results. These provisions apply to the prerelease MCP service. 16
Customers evaluating the prerelease service bear the operational consequences while their test findings can remain confidential, limiting independent scrutiny of its advertised benefits.
Configuration maintenance and total ownership cost
Streamline’s flexibility creates a substantial maintained model: forecast overrides, service targets, supplier parameters, item groupings, container rules, calendars, production processes, and permissions. Hierarchical inheritance reduces repetitive entry but also allows one change to affect many recommendations.
The lead-time workflow illustrates the recurring burden particularly well. Observed supplier behavior, planning parameters in the source system, application overrides, and incoming orders must remain consistent. A technically successful data refresh can preserve a stale purchasing assumption.
Integration also adds work beyond initial setup. The intermediate-database guide assigns the customer’s IT team responsibility for arranging the two-way exchange with its enterprise system. It recommends this separation when direct access is prohibited or heavy planning queries could affect production-system performance. 22
That separation improves the boundary, but someone must maintain the mapping when an ERP changes a status code, a subsidiary is added, or an order is rejected. Incorrect mappings can create plausible-looking forecasts from the wrong transactions.
Current pricing requires a tailored quotation. 33 A meaningful ownership-cost comparison therefore needs the subscription and implementation charges plus planner review, integration maintenance, parameter upkeep, model validation, and incident correction. The customer cases do not quantify that complete recurring workload.
Technology history and roadmap integrity
| Period | Evidence | Implication |
|---|---|---|
| 2021 | A November 30 vendor announcement already described GMDH’s product as AI-powered. 26 | Early AI branding referred to specialized forecasting and planning, not the later capabilities of language models. |
| 2024 | Streamline’s September comparison with Netstock promoted AI, dynamic simulation, and a digital twin as differentiators. 27 | The simulation narrative predates the current agent branding; its operational predictions still need validation. |
| 2025–2026 | The public changelog records continuing releases rather than only marketing announcements. In September 2026, MCP tools were generated from the product’s OpenAPI specification; October extended MCP to on-premises installations. 25 | The connector is identifiable engineering work, but chiefly exposes existing application capabilities. |
| October 2026 | The release notes say the previous intermittent model could produce excessive safety stock on long order cycles and announce a replacement. Existing items switch automatically. 25 | A substantive modeling correction also changes deployed purchasing behavior and requires economic regression checks. |
The intermittent-model change is more informative than another AI label. It acknowledges a concrete failure mode and records remediation. Yet the current forecasting and safety-stock manuals still describe the older log-normal formulation, while the release notes describe the replacement differently. 3 4 25
This leaves a version-specific documentation gap. A customer evaluating today’s recommendations cannot assume that every published formula matches the running model. Automatic migration also makes before-and-after checks necessary: changing safety stock changes cash commitment and shortage exposure.
The historical pattern is a persistent forecasting-and-replenishment product wrapped in increasingly expansive claims. Delivery is real; the evidence for autonomous economic optimization remains much thinner. Product seriousness should be judged by demonstrated changes to decisions and costs, including the consequences of corrected defects.
Bill of software materials
The current platform requirements establish a web application and Linux hosting option. Employer recruitment and a former developer’s public professional account identify parts of the engineering environment. 2 23 24
| Component or mechanism | Attributable role |
|---|---|
| C++ | Product engineering in Streamline’s August 2026 recruitment |
| Qt; CMake | The developer’s account of 2024–2026 work describes a Qt 5-to-6 migration and changes to the proxy application’s build |
| Node.js | The same account identifies the company’s OAuth authorization server |
| Python | Internal data-import and installer-packaging utilities in that account |
| Jenkins, later GitHub Actions | Reported delivery pipeline for the proxy application |
| ODBC and direct MySQL connectivity | Documented connections to customer databases 18 |
| Browser application, Linux deployment, enterprise-system connectors | Current deployment and integration boundary |
The customer databases are integration targets whose connectors add compatibility and upgrade work.
The concrete maintenance burden concerns the interaction of the planning engine, web interface, authentication, proxies, and ERP connectors. The disclosed Qt and build migrations illustrate work that continues after a feature is delivered. The public record does not quantify how much integration-specific code is maintained per customer.
Data can be exported, and documented schemas make migration more assessable. The forecasting logic, configuration conventions, and order-generation behavior would still require reconstruction in a replacement system. Portability of records does not imply portability of the decision process.
Security mechanisms and unsafe integration guidance
The current security page describes isolated customer application instances, encryption in transit and at rest, multifactor authentication, role-based access, and automatic backups. Its user manual further separates feature permissions from data-access catalogs. 17 21 These mechanisms provide a basis for assessment beyond the SOC 2 badge.
The database-connection instructions contain a more troubling detail. When SQL Server’s encrypted connection fails, the guide tells users to change connection encryption to Optional. Elsewhere it explains configuring certificates for encrypted connections. 18 The workaround weakens the default precisely when the connection’s security needs investigation.
The linked SQL-login guide, last modified in May 2021, instructs users to disable password-policy enforcement while creating a read-only database account. 19 Read-only access limits modification rights but still exposes business data. The age of the instruction does not remove its relevance while the current integration manual directs users to it.
Streamline’s SQL Assistant generates transformation queries against the intermediate database and asks users to preview results before applying them. 20 The October release also confines transformation scripts to their own connector’s tables. 25 The remaining questions concern database permissions, resource limits, and separation between generated transformations and production-system writes.
MCP adds another boundary: an external model can read data and invoke permitted operations as the user. Authentication establishes who authorized access; it does not establish that the model chose the intended action. Public material does not adequately describe how malicious instructions inside imported data are contained, or how economically consequential actions are bounded independently of the user’s broad permissions.
Supply chain score
Supply chain depth
| Criterion | Score |
|---|---|
| Economic framing | 4/10 |
| Decision end-state | 6/10 |
| Conceptual sharpness on supply chain | 4/10 |
| Freedom from obsolete doctrinal centerpieces | 4/10 |
| Robustness against KPI theater | 4/10 |
Average: 4.4/10
Economic framing — 4/10. Cost limits and inventory economics appear in the product, but order policies remain driven by service targets and replenishment rules rather than a disclosed joint financial objective. 4 7 9
Decision end-state — 6/10. The software produces and exports purchase, transfer, and manufacturing orders. Approval responsibilities and separately configured calculation paths limit the autonomous claim. 7 14
Conceptual sharpness on supply chain — 4/10. Timing, constraints, and supply variability receive concrete treatment. The public design is less convincing on correlated uncertainty and financial competition among decisions. 5 7 8
Freedom from obsolete doctrinal centerpieces — 4/10. Forecast-first planning, safety-stock targets, ABC-style configuration, and consensus reviews remain central. Dynamic simulation expands visibility without removing those policy limitations. 4 27 34
Robustness against KPI theater — 4/10. Error formulas and exclusions are documented. The projected Genomma opportunity and blanket ROI promise are presented more strongly than their economic evidence supports. 13 27 31
Decision and optimization substance
| Criterion | Score |
|---|---|
| Probabilistic modeling depth | 4/10 |
| Distinctive optimization or ML substance | 4/10 |
| Real-world constraint handling | 5/10 |
| Decision automation and labor productivity | 5/10 |
| Resilience under real operational complexity | 4/10 |
Average: 4.4/10
Probabilistic modeling depth — 4/10. Demand variability and supply variance affect protection, but dependence, predictive calibration, and automatic maintenance of planning uncertainty remain weakly established. 4 5 6
Distinctive optimization or ML substance — 4/10. The engine and formulas are identifiable. Public evidence does not establish distinctive risk-adjusted economic optimization beyond the documented forecasting and policy calculations. 3 4 7
Real-world constraint handling — 5/10. Shared order constraints, containers, materials, and workstation scheduling are documented. Different horizon calculations and manual capacity resolution leave important boundaries. 8 9 10 11
Decision automation and labor productivity — 5/10. Scheduled calculations, exports, and customer cases support administrative gains. Complete recurring labor and the incremental productivity of MCP remain unmeasured. 14 15 29 30
Resilience under real operational complexity — 4/10. Users can revise configurations and replay plans. The public evidence is weaker on diagnosing an inadequate decision model, reformulating it, and validating the correction across the affected operations. 12 25 34
Product and architecture integrity
| Criterion | Score |
|---|---|
| Architectural coherence | 6/10 |
| System-boundary clarity | 7/10 |
| Security seriousness | 4/10 |
| Software parsimony and ownership cost | 4/10 |
| Compatibility with programmatic and agent-assisted operations | 4/10 |
Average: 5.0/10
Architectural coherence — 6/10. A shared planning application connects forecasts, inventory, and orders. Multiple calculation modes and evolving model versions require more consistent documentation and validation. 3 7 25
System-boundary clarity — 7/10. Deployment, enterprise-system exchange, intermediate databases, and user permissions are documented. The division between external AI reasoning and the planning engine is explicit. 2 15 21 22
Security seriousness — 4/10. Isolation, authentication, and data permissions are described concretely. Unsafe integration workarounds and incomplete action-containment details weaken the security design presented to customers. 17 18 19 20 21
Software parsimony and ownership cost — 4/10. The identifiable components have operational roles, but configuration, connectors, migrations, and review work create an unquantified recurring burden. 22 23 24 33 34
Compatibility with programmatic and agent-assisted operations — 4/10. APIs and scheduled execution provide useful programmatic access. MCP mainly exposes the existing application and leaves the economic decision method unchanged. 14 15
Technical transparency
| Criterion | Score |
|---|---|
| Public technical documentation | 7/10 |
| Inspectability without vendor mediation | 7/10 |
| Portability and lock-in visibility | 5/10 |
| Implementation-method transparency | 6/10 |
| Evidence density behind technical claims | 6/10 |
Average: 6.2/10
Public technical documentation — 7/10. Manuals disclose calculations, restrictions, settings, and integration behavior. The mismatch between model releases and formula documentation prevents a higher score. 3 4 25
Inspectability without vendor mediation — 7/10. Many operational claims can be checked from public references. Model-selection internals and reproducible performance evaluations remain unavailable. 3 7 13
Portability and lock-in visibility — 5/10. Schemas and exports clarify data movement. Replacing the proprietary policy engine and its customer-specific configuration would still require substantial work. 14 22 34
Implementation-method transparency — 6/10. Data integration, permissions, configuration, and approval workflows are described in detail. The ongoing effort and costs are not comparably quantified. 14 18 21 34
Evidence density behind technical claims — 6/10. The record includes formulas, dated engineering changes, and named deployments. Autonomous decision quality and simulation validity are supported much less strongly. 4 25 29 30
Vendor seriousness
| Criterion | Score |
|---|---|
| Technical seriousness of public communication | 3/10 |
| Technology roadmap integrity | 6/10 |
| Conceptual sharpness | 4/10 |
| Economic accountability | 4/10 |
| Defensibility in an agentic-software world | 5/10 |
Average: 4.4/10
Technical seriousness of public communication — 3/10. The broad AI, autonomy, and ROI language exceeds the demonstrated mechanisms and outcomes. The more precise manuals frequently provide the necessary qualifications. 1 15 27 31
Technology roadmap integrity — 6/10. Detailed releases show delivery and concrete corrections. Documentation lag and automatically changing deployed models create additional verification work for customers. 3 25
Conceptual sharpness — 4/10. Streamline distinguishes forecasting, integration, and external AI in its documentation. Its commercial narrative blurs these into a much broader optimization claim. 3 7 15
Economic accountability — 4/10. Some cases identify tangible savings, but project economics are incomplete and one major inventory headline remains a projection. Prerelease MCP terms sharply limit vendor responsibility and public testing transparency. 16 29 30 31
Defensibility in an agentic-software world — 5/10. The maintained engine and operational integrations have substance. Conversational access and repeated planner reviews provide less durable differentiation than superior decisions with lower total labor. 7 14 15 30
Sources
Sources were checked on November 4, 2026. Customer results are vendor-reported. The release history and formula manuals are identified separately where they describe different model versions.
[1] Streamline product homepage
Streamline product homepage. Undated; accessed November 4, 2026.
System requirements. Undated; accessed November 4, 2026.
Statistical forecasting. Undated; accessed November 4, 2026.
Safety stock calculations. Undated; accessed November 4, 2026.
[5] Supplier parameters and performance
Supplier parameters and performance. Published or updated 2026-10-07; accessed November 4, 2026.
Confidence levels. Undated; accessed November 4, 2026.
[7] Ordering plan calculations
Ordering plan calculations. Undated; accessed November 4, 2026.
Containers and groups. Undated; accessed November 4, 2026.
Constrained ordering plan. Undated; accessed November 4, 2026.
[10] Rough-cut capacity planning
Rough-cut capacity planning. Undated; accessed November 4, 2026.
Production scheduling. Undated; accessed November 4, 2026.
Time Machine. Undated; accessed November 4, 2026.
[13] Forecast analysis reports
Forecast analysis reports. Undated; accessed November 4, 2026.
[14] Creating and sending purchase orders
Creating and sending purchase orders. Undated; accessed November 4, 2026.
Streamline MCP. Undated; accessed November 4, 2026.
[16] Streamline MCP Terms of Use
Streamline MCP Terms of Use. Published or updated 2026-09-02; accessed November 4, 2026.
Security policies. Undated; accessed November 4, 2026.
Database connection. Undated; accessed November 4, 2026.
[19] Creating an SQL Server login for Streamline
Creating an SQL Server login for Streamline. Published or updated 2021-05-06; accessed November 4, 2026.
SQL Assistant. Undated; accessed November 4, 2026.
Users and roles. Undated; accessed November 4, 2026.
Intermediate database. Published or updated 2026-09-29; accessed November 4, 2026.
[23] Streamline: Trainee Engineer
Streamline: Trainee Engineer. Published or updated 2026-08-04; accessed November 4, 2026.
[24] Former Streamline developer’s public professional résumé
Former Streamline developer’s public professional résumé. Published or updated 2026-10-12; accessed November 4, 2026.
[25] Streamline release history
Streamline release history. Published or updated 2026-10-08; accessed November 4, 2026.
[26] GMDH’s November 2021 award announcement
GMDH’s November 2021 award announcement. Published or updated 2021-11-30; accessed November 4, 2026.
[27] Streamline versus Netstock comparison
Streamline versus Netstock comparison. Published or updated 2024-09-10; accessed November 4, 2026.
[28] Predictive planning webinar recap
Predictive planning webinar recap. Published or updated 2026-10-12; accessed November 4, 2026.
Whalen furniture case. Undated; accessed November 4, 2026.
[30] Transgold replenishment case
Transgold replenishment case. Undated; accessed November 4, 2026.
[31] Genomma Lab inventory case
Genomma Lab inventory case. Undated; accessed November 4, 2026.
GMDH product homepage. Published or updated 2026-09-08; accessed November 4, 2026.
Streamline pricing. Undated; accessed November 4, 2026.
[34] Configuring replenishment parameters
Configuring replenishment parameters. Undated; accessed November 4, 2026.