Manhattan Associates is recasting warehouse and fulfilment systems from long-cycle projects into an AI-enabled execution platform that runs continuously at the edge of the network.
In Brief
- Embedded AI agents inside Manhattan’s WMS, TMS and store stack shift automation from bespoke projects to platform-native capabilities.
- Simulation and promising tools move fulfilment design from annual rulesetting to scenario-based, seasonal optimisation embedded in operations.
- Cloud economics and ramped ARR show how multi-year warehouse and transport programmes are being industrialised and monetised over time.
A Structural Break: WMS Stops Being a One-off Project
Manhattan’s recent disclosures point to a clear break from the traditional view of warehouse management as a capital project that is designed, deployed and then stabilised for years. The company is building an always-on supply chain platform where warehouse, transport, order management and store systems are updated quarterly, overlaid with AI agents and optimisation services.
The shift is visible in three elements of the operating model:
- Active Warehouse and Active Transportation are sold and deployed as a single platform rather than separate stacks, with customers explicitly valuing end-to-end optimisation of inbound and outbound flows.
- Cloud revenue is growing at over 20 percent year-on-year, while on-premise license and maintenance revenues are deliberately allowed to decline, with management flagging a four-point revenue headwind from this mix shift in 2026.
- Manhattan is hiring ahead of demand, onboarding around 100 new services associates in January 2026, specifically to support cloud migrations and Agentic AI deployments.
Ramped ARR above 600 million dollars, up 23 percent year-on-year, underlines how the economic model now mirrors the operational cadence. Contracts ramp to full pricing over four years as distribution centres, transport lanes and stores are brought onto the platform in phases. The introduction of a four-year ramped ARR metric acknowledges that for most customers, Manhattan’s software is no longer a single go-live, but a multi-year transformation of the physical network.
How AI Agents Are Being Embedded Into Execution
The most visible new layer is Active Agents, Manhattan’s Agentic AI capability built directly into its platform. Rather than sitting in a separate analytics stack or data lake, the agents are embedded into the transactional systems that run warehouses, transportation and stores.
Operationally, this matters for three reasons:
- The agents work off the same master data, orders and inventory positions that drive WMS, TMS, OMS and POS, avoiding latency and integration complexity.
- Base agents are preconfigured for common workflows in warehouse, transportation, contact centre and stores, and can be activated with minimal configuration.
- Agent Foundry lets customers extend or build agents for specific operational problems, without needing to stand up separate AI infrastructure.
Manhattan reports that early adopters have seen increased automation, simpler user experiences and higher productivity. Although the company has not disclosed quantitative benchmarks, the intended outcome is clear: fewer manual exceptions, faster decision-making and more consistent application of network-wide rules.
In operational terms, this kind of embedded AI usually targets three layers of warehouse and transport execution:
- Task orchestration: generating and sequencing work for pick, pack, replenishment or loading based on live constraints rather than static priorities.
- Exception handling: triaging late inbound loads, short picks or inventory mismatches, proposing remedies within service thresholds and labour constraints.
- Advisory workflows: guiding supervisors and store associates to adjust labour, slotting or fulfilment options based on current sales and backlog signals.
By packaging these capabilities as an uplift to existing modules, priced similarly to labour management or slotting, Manhattan is trying to normalise AI as another optimisation component within the core platform, rather than a separate innovation programme that competes for budget.
Turning Fulfilment Design Into Continuous Simulation
Alongside the agent layer, Manhattan is adding tools that change how fulfilment strategies are designed and deployed. A new fulfilment optimisation simulation capability sits on top of Active Omni, Manhattan’s order management system.
The intent is to allow operators to test different fulfilment strategies before putting them into production. Examples described by management include:
- Prioritising speed of delivery during peak trading periods.
- Shifting the objective function toward clearing distressed inventory at the end of a season to avoid markdowns.
In operational terms, this requires Manhattan’s OMS to expose and parameterise key allocation and sourcing rules, such as:
- Node eligibility and ranking logic across DCs, stores and drop-ship partners.
- Inventory reservation and safety stock policies for different channels.
- Cost models for transport modes, lanes and carrier options.
The simulation engine then runs historical or synthetic order flows through alternative rule sets, comparing outcomes on lead time, cost and inventory ageing. Manhattan positions this as a sandbox to ‘ensure the system is ready to pivot fulfilment strategies when the business calls for it.
At network level, this moves fulfilment design from annual rulesetting into something closer to a quarterly or even monthly planning cadence. For a retail or wholesale network, it implies tighter coordination between merchandising, inventory planning, transport planning and operations, because each campaign, season or macro shock can justify a different trade-off between speed, cost and inventory risk.
Pushing AI and Optimisation To The Edge
Unlike hyperscale AI deployments in data centres, where TE Connectivity and Schneider Electric are building out interconnects, liquid cooling and power infrastructure, Manhattan’s AI is deployed at the operational edge: distribution centres, transport control towers, contact centres and stores.
This edge focus has structural implications for supply chain governance:
- Data ownership and latency: because agents sit inside the platform, there is no requirement to replicate operational data into external lakes, which reduces integration overhead and keeps decision latency low.
- Change management: AI-driven changes are delivered through Manhattan’s quarterly cloud releases. Customers that adopt a quarterly update cadence receive a ‘steady dose’ of services and improvements, rather than episodic upgrades.
- Role redesign: Manhattan is explicit that forward-deployed engineers and domain specialists will accompany 90-day AI pilots, helping customers learn standard agents and build at least one or two custom ones.
In operational terms, this kind of shift typically requires a redesign of warehouse and transport roles. Supervisors move from manually assigning tasks and chasing exceptions to validating and tuning agent behaviour. Analysts spend more time on scenario definition and post-implementation measurement than on ad hoc reporting. This is consistent with broader shifts seen at companies like Danone and Nissan, where teams are reoriented around orchestration and simulation rather than pure execution.
From Separate Stacks To Unified Flow Control
Manhattan is also explicit about a structural preference shift on the customer side: ‘customers no longer want to select separate stacks for warehouse and transportation.’ This is a material break for many organisations that historically ran independent WMS and TMS programmes, often from different vendors.
When WMS and TMS sit on a unified platform, several network-level changes become feasible:
- Inbound and outbound flows can be planned against shared capacity models, with yard, dock and labour schedules aligned to transport schedules rather than treated as separate constraints.
- Cost-to-serve can be computed across warehouse handling and transport legs, allowing different routing options to be compared on a full landed basis, not just linehaul rates.
- Inventory placement strategies can be evaluated with explicit consideration of transport lead times, carrier performance and slotting implications.
To operationalise this, master data and planning processes need to converge. Location hierarchies, calendar definitions, carrier and service level definitions must be consistent across modules. Allocation logic in OMS, wave planning in WMS and load building in TMS have to work off the same service thresholds and priority rules.
Manhattan’s fixed-fee, fixed-timeline offers for rolling out the next set of distribution centres, using a ‘recipe’ built on previous deployments, indicate that it is trying to industrialise this convergence. The more repeatable the template, the easier it becomes to deploy the same flow-control logic across ten or more nodes without reinventing design decisions each time.
A Sticky, Multi-year Economic Model With Real Friction
The platform shift is underpinned by an economic model that is long-dated and relatively sticky. Average contract durations sit between 5.5 and 6 years; around 38 percent of RPO is expected to be recognised in the next 24 months. Renewals account for roughly 18 percent of bookings today and are expected to remain a meaningful contributor, with management signalling opportunities to bring some renewals earlier to three-year cycles to reprice as value is proven.
At the same time, there are constraints and trade-offs that temper the narrative:
- Maintenance and license attrition is a deliberate drag, shaving around four points off headline revenue growth in 2026 as on-premise customers migrate.
- Services margin is expected to remain flat year-on-year as Manhattan absorbs the cost of new headcount to support AI and cloud programmes.
- AI revenue is treated as upside to current guidance rather than baked into the base, suggesting that adoption risk and customer readiness are still material unknowns.
These frictions matter. Moving WMS, TMS and OMS to a unified, cloud-native stack touches every warehouse, store and carrier relationship. Even with fixed-timeline offers and standardised recipes, the execution load on both Manhattan and its customers is high. The decision to structure AI as an uplift and to run 90-day pilots reflects an attempt to phase this load rather than stack transformational change all at once.
Editorial Synthesis: WMS as an Operating System
Taken together, Manhattan’s actions show a clear directional move: from selling warehouse and transport systems as standalone applications to running a supply chain operating system that sits at the edge, is updated quarterly, and comes with embedded AI and optimisation agents.
For customers, this operating model creates a different set of obligations and options. It allows warehouse and transport design to evolve continuously through simulation and agents, tied into a single data and rule base. It reduces the need for separate AI stacks and integration-heavy projects, but it also requires disciplined master data, tighter governance across functions and a willingness to redesign roles around an AI-assisted execution model.
For Manhattan, the model locks in multi-year revenue ramps and strong cash generation, but exposes the company to the execution reality of being woven into customers’ physical networks. The platform is now close enough to the dock door and the store till that its performance will be judged less on features and more on how reliably it shapes flow, cost and service in day-to-day operations.
