AIoT Systems vs. Expensive Pilots (2026) | Research Report
AIoT Systems vs. Expensive Pilots (2026) | Research Report
AIoT Systems vs. Expensive Pilots (2026) | Research Report
Recently Updated on
September 17, 2026
Index
This report is written by Mujtaba Sheikh, AI-First Solution Architect and Fractional CTO at Phaedra Solutions. His 13+ years across IoT architecture, real-time monitoring, predictive maintenance, and connected product delivery inform the report’s focus on operational scalability rather than pilot-stage connectivity.
Connecting physical assets is no longer the primary challenge in an Internet of Things initiative. The more difficult task is converting device data into reliable operational decisions.
Phaedra Solutions reviewed current AIoT research alongside its experience delivering real-time monitoring, predictive maintenance, connected dashboards, and command-and-control systems. The evidence shows that AIoT systems scale when they support a specific operational decision, have a named owner, and are designed for secure lifecycle management. They stall when connectivity, dashboards, or model development are treated as the final outcome.
To support clarity throughout this report, we define three related terms at the outset.
Internet of Things system: A network of connected devices that collects, transmits, and exchanges operational data.
AIoT system: An IoT system in which artificial intelligence contributes to prediction, recognition, classification, recommendations, or operational action.
Connected operations: An operating model that connects devices, data, intelligence, people, workflows, and governance into a repeatable decision process.
An AIoT system does not require every device to run an AI model. It requires AI to make a material contribution to the operational decision loop.
Key Findings at a Glance
21.1 Billion Connected IoT Devices.
IoT Analytics estimated that the number of connected IoT devices worldwide reached 21.1 billion by the end of 2025 and could reach 39 billion by 2030. (1)
$153.9 Billion Industrial AI Market.
The global industrial AI market is forecast to grow from $43.6 billion in 2024 to $153.9 billion by 2030. (2)
3 Primary AIoT Functions.
A 2025 systematic review of 103 academic articles found that AIoT systems are primarily used for prediction, object and event recognition, and operational decision-making. (5)
30–40% Reduction in Unplanned Downtime.
Phaedra Solutions measured a 30% to 40% reduction in unplanned downtime across AIoT-enabled monitoring deployments compared with non-connected operating baselines.
Executive Summary: Why Some AIoT Systems Scale and Others Stall
The AIoT market has moved beyond the basic connectivity question. Independent forecasts differ in how they define devices and connections, but they consistently indicate that connected systems will expand into the tens of billions during the next decade. (1) (3) (4)
The more important question is whether organizations can convert that connectivity into measurable operational value.
The most scalable AIoT use cases are concentrated in a practical group of applications:
Command-and-control systems with defined response workflows
These use cases perform well because they support decisions that operations teams already need to make. They help determine when equipment should be serviced, whether a product meets quality standards, where an asset is located, whether an event requires escalation, or what action should follow an abnormal condition.
Across the research and Phaedra Solutions’ delivery experience, five conditions consistently influence whether an initiative progresses beyond a pilot:
A named operational owner is accountable for acting on the system’s output.
Sensor and operational data are validated before model deployment.
Cloud and edge responsibilities reflect actual latency and connectivity requirements.
Device, firmware, credential, and model lifecycles are managed from the beginning.
Security, integration, and response processes are treated as launch requirements.
In my experience, the most expensive AIoT failure is not always an inaccurate model. It is often an accurate prediction delivered into an operating process with no owner, response threshold, or authority to act. A prediction creates value only when the organization has defined what happens next.
Connectivity Is Scaling Faster Than Operational Value
IoT Analytics estimated that connected IoT devices reached 21.1 billion worldwide at the end of 2025, with the total expected to reach 39 billion by 2030 and exceed 50 billion after 2035. (1)
GSMA Intelligence separately forecasts 40.8 billion IoT connections by 2030 and 52.9 billion by 2035. The figures are not directly interchangeable because each organization applies different definitions, but both forecasts indicate sustained growth. (3)
Cellular IoT is also expanding. Ericsson reported approximately 4.5 billion cellular IoT connections at the end of 2025. It also reported that 5G Reduced Capability, or RedCap, had been launched by 14 service providers, with additional operators investing in the technology. (4)
RedCap supports mid-range IoT applications that require more capability than narrowband technologies but do not need full mobile-broadband performance. This expands the connectivity options available for industrial monitoring, wearables, video-based applications, and other moderate-bandwidth deployments. (4)
The industrial AI market is developing in parallel. IoT Analytics estimates that the market will grow from $43.6 billion in 2024 to $153.9 billion by 2030. Its research identifies quality inspection as a leading industrial AI use case and edge AI as an important area of future growth. (2)
These figures show that connectivity and AI adoption are expanding. They do not, however, confirm that organizations are extracting equivalent operational value.
Connected-device counts are infrastructure measures, not business outcomes. The more useful measures are how many failures were prevented, how quickly incidents were resolved, how much inspection effort was reduced, and whether operational decisions improved because the system existed.
Industry Trends and Market Insights: Connected Operations in 2026
1. Enterprise IoT Is Entering the Connected-Operations Phase
IoT Analytics reports that the enterprise IoT market grew 13% in 2025 to approximately $324 billion and projects 14% growth in 2026, with AI among the main drivers. Its 2026 market framing describes a shift from isolated connectivity toward autonomous and cross-ecosystem connected operations. The relevant business trend is not simply more devices; it is tighter integration between device data, operational systems, AI, and controlled action. (10)
2. Edge AI Is Expanding, but Hybrid Architecture Remains Essential
The 2026 smart-manufacturing roadmap identifies industrial data management, heterogeneous sensing and control integration, explainability, reliability, and trustworthy operation as core barriers to scale. These requirements support a hybrid model: time-sensitive inference can move closer to equipment, while model training, fleet-wide comparison, policy management, and historical analysis remain centralized. (11)
3. AIoT Is Converging With Robotics and Physical AI
A 2026 review of AI, IoT, and robotics integration describes an emerging architecture in which smaller models operate near devices while larger cloud models support coordination and complex reasoning. This direction is still early, but it indicates that future connected systems will increasingly combine perception, local inference, workflow orchestration, and controlled physical action rather than stopping at alerts and dashboards. (12)
4. Lifecycle Cost Is Replacing Device Cost as the Key Business Measure
As deployments expand, the cost of provisioning, identity, firmware, credentials, model distribution, observability, incident response, and retirement becomes more important than the unit price of a sensor. For decision-makers, total lifecycle cost and the operational outcome supported by each device will be more useful measures than the number of connected endpoints.
The market is therefore moving toward platforms and delivery models that can manage devices, data, models, policies, and operational responses as one governed system. AIoT providers will increasingly be judged on secure fleet operations and measurable decision outcomes, not on connectivity demos.
Research Methodology
This report combines published industry research with delivery observations from Phaedra Solutions’ AIoT and connected-systems work.
It is intended to identify recurring operational and architectural patterns rather than provide a statistically representative assessment of the entire AIoT market.
External Evidence Review
The report reviewed findings and forecasts from:
IoT Analytics’ connected-device forecasts and industrial AI research (1) (2)
GSMA Intelligence’s IoT market outlook (3)
Ericsson’s IoT connectivity forecasts (4)
A 2025 systematic review of 103 AIoT studies (5)
Deloitte’s 2026 State of AI in the Enterprise research (6)
Microsoft Azure’s IoT and edge-AI architecture guidance (7) (8)
The NIST IoT Device Cybersecurity Capability Core Baseline (9)
Deloitte’s 2026 research included responses from 3,235 business and technology leaders across 24 countries. (6)
Phaedra Solutions Delivery Review
Phaedra Solutions’ observations are based on connected-systems work involving:
The review considered the operational and technical conditions that influenced whether systems delivered measurable value or remained limited to monitoring and proof-of-concept use.
Evaluation Criteria
The systems and use cases reviewed were assessed against the following criteria:
Operational decision supported
Named owner of the decision
Cloud, edge, or hybrid deployment model
Latency sensitivity
Connectivity environment
Data quality requirements
Device and model update requirements
Security and compliance requirements
Integration with existing operational systems
Reported or observed business outcome
Study Limitations
Public case studies tend to represent successful deployments more often than failed pilots. Organizations are less likely to publish projects that delivered limited returns, experienced security problems, or did not progress beyond an initial trial.
Phaedra Solutions’ delivery experience is weighted toward monitoring, predictive maintenance, dashboards, and command-and-control applications. Findings for other AIoT applications should be interpreted in the context of their industry, operating environment, and risk profile.
Industry sources also use different definitions of an IoT device, IoT connection, cellular connection, and industrial AI market. Their forecasts demonstrate scale and direction but should not be compared as identical measures. (1) (3) (4)
Where AIoT Systems Create Operational Value
The strongest AIoT use cases typically involve narrow and repeated decisions supported by reliable data.
A systematic review of 103 AIoT studies found that prediction, object and event recognition, and operational decision-making were the most common AI functions. (5)
IoT Analytics also identifies quality inspection as a leading industrial AI application. (2)
AIoT Use Cases, Decisions, and Scale Requirements
AIoT Use Case
Operational Decision
Typical Architecture
Requirements for Scale
Predictive Maintenance
When an asset should be inspected, serviced, or taken offline
Edge inference with cloud aggregation
Historical failure data, reliable sensor inputs, maintenance capacity, and a named response owner
Machine-Vision Quality Inspection
Whether a product should pass, be reworked, or be rejected
Whether an asset should be located, rerouted, recovered, or escalated
Hybrid or cloud-led
Integration with ERP, warehouse, transport, or logistics systems
Operational Anomaly Detection
Whether unusual behavior requires investigation or intervention
Edge or hybrid
Reliable baselines, severity thresholds, escalation paths, and alert ownership
Edge-Based Classification
Whether a local event or object matches a defined operational category
Edge
Sufficient local computing capacity, controlled model updates, and fallback behavior
Command-and-Control Systems
Which event requires attention and what action should follow
Cloud aggregation with local edge alerts
Prioritization rules, user permissions, operating procedures, and clear authority to act
Microsoft’s edge-AI guidance identifies image classification, object detection, and related visual-analysis tasks as common edge scenarios. It also notes that incremental model updates are important in narrow-bandwidth environments, where repeatedly transferring complete model packages can increase cost and deployment downtime. (7)
Predictive Maintenance
Predictive maintenance creates value when it enables intervention before an asset fails.
The model requires reliable historical data, suitable sensor placement, maintenance capacity, and a clear financial basis for acting on an alert. Model accuracy alone does not produce a maintenance outcome.
A complete implementation should connect predictions with:
Maintenance scheduling
Work-order creation
Parts availability
Asset criticality
Inspection requirements
Operational authority
Phaedra Solutions measured a 30% to 40% reduction in unplanned downtime across AIoT-enabled monitoring deployments where predictions were connected to defined maintenance processes.
Machine-Vision Quality Inspection
Machine vision can reduce manual inspection effort and improve consistency, particularly on stable production lines.
Performance may decline when products, lighting, camera positions, materials, packaging, or defect definitions change. The system therefore requires a process for detecting model drift and collecting new labeled examples.
The operating process must also define the action that follows a detection. Depending on the environment, this may include rejecting an item, stopping equipment, notifying an operator, or routing the product for secondary inspection.
Asset and Shipment Visibility
Visibility systems are most valuable when they support faster exception handling.
Knowing the location of an asset is useful, but operational value comes from determining whether it is delayed, misplaced, damaged, exposed to unsuitable conditions, or likely to disrupt a downstream process.
These systems should integrate with the operational platforms through which action is taken, such as:
Enterprise resource planning systems
Warehouse management systems
Transport management systems
Maintenance platforms
Service-management workflows
A standalone tracking dashboard may improve visibility without materially improving operational performance.
A related example is Phaedra Solutions’ AI Inventory Management Software, which combines real-time inventory tracking, AI-driven reporting, and centralized web and mobile dashboards.
Operational Anomaly Detection
Anomaly detection can identify abnormal conditions before they develop into failures. It can also create excessive alerts if the system does not distinguish between unusual activity and operationally important activity.
A scalable anomaly-detection system requires:
Defined severity levels
Escalation rules
Alert-suppression logic
Named responders
Response-time targets
Feedback from operational teams
Measurement of false positives and false negatives
A dashboard becomes operationally useful only when important signals have a threshold, an owner, a response window, and an action.
Five Reasons AIoT Pilots Stall Before Scale
The evidence reviewed for this report indicates that most stalled AIoT initiatives fail because of operating-model and architecture weaknesses rather than hardware limitations.
1. No Operational Owner or Business Metric
Many pilots have a technical sponsor but no operations leader accountable for the result.
The system may demonstrate that it can detect an anomaly or forecast a failure, but nobody owns the response time, downtime reduction, asset-loss rate, inspection cost, or maintenance outcome.
Without that accountability, there is no clear owner for operational adoption or further investment.
2. Weak Sensor and Operational Data
AI cannot compensate for unreliable measurements, incomplete maintenance records, inconsistent timestamps, changing sensor calibration, or labels that do not represent real operating conditions.
Data validation should assess:
Completeness
Acceptable operating ranges
Timing and synchronization
Duplication
Calibration
Sensor health
Environmental context
Asset operating mode
A model trained on weak data may perform well during a controlled pilot and fail when operating conditions change.
3. Missing Device and Model Lifecycle Management
A small pilot can be supported through manual configuration. A scaled deployment requires repeatable processes for:
Device provisioning
Device identity
Firmware updates
Credential rotation
Configuration management
Model versioning
Model rollback
Health monitoring
Device replacement
Secure retirement
NIST’s IoT cybersecurity baseline identifies device capabilities needed to support common security requirements across connected environments. (9)
Device and model lifecycle management must be included in the initial system design.
4. Incorrect Cloud-Edge Partitioning
Sending all data to the cloud may be appropriate for centralized analysis, but it can create problems where latency, connectivity, bandwidth, privacy, or safety requirements demand local processing.
Running all functions at the edge creates other challenges, including limited computing capacity, fragmented observability, model-distribution complexity, and higher field-maintenance requirements.
The architecture should be based on operating conditions rather than a default cloud-first or edge-first approach.
5. Security and Integration Are Deferred
Pilots are often built in controlled environments with temporary credentials, limited device counts, isolated datasets, and simplified integrations.
These assumptions do not remain valid at production scale.
Security requirements include:
Device identity
Authentication
Encryption
Access control
Update integrity
Audit logging
Incident response
Secure device retirement
Integration requirements may include operational technology, ERP, maintenance, logistics, identity, ticketing, and reporting systems.
Deferring these requirements frequently results in significant redesign before the system can be expanded.
Cloud, Edge, or Hybrid AIoT Architecture
The deployment model should reflect where the decision must occur and how quickly the system must respond.
Cloud-Led Architecture
A cloud-led architecture is suitable when:
Connectivity is reliable
Decisions can tolerate network latency
Centralized aggregation is the main requirement
Local processing requirements are limited
Fleet-wide analysis is more important than immediate action
The system primarily supports reporting, optimization, or planning
Typical applications include long-term asset-performance analysis, fleet reporting, energy analysis, and centralized monitoring.
Edge-Led Architecture
An edge-led architecture is suitable when:
Decisions must be made in real time
Connectivity is intermittent or unavailable
Continuous data transfer would be costly
Safety or uptime requires local operation
Privacy or policy restricts data movement
Local equipment or protocols require on-site processing
Microsoft identifies industrial and operational-technology environments as common candidates for edge-connected architectures, particularly where equipment cannot connect directly to the public internet. (8)
Hybrid Architecture
For many connected-operations systems, a hybrid architecture provides the most appropriate balance.
The edge can manage:
Local data collection
Immediate inference
Time-sensitive alerts
Temporary data buffering
Continuity-sensitive decisions
The cloud can manage:
Fleet-wide visibility
Long-term storage
Model training
Comparative analytics
Central policy management
Model and configuration distribution
Edge computing is not inherently more advanced than cloud computing. The design question is which intelligence must remain local to protect response time, resilience, safety, privacy, and cost. Functions that benefit from centralization should remain centralized.
The AIoT Maturity Curve
Phaedra Solutions uses a four-stage model to evaluate whether a connected-operations initiative is prepared to progress beyond a limited pilot.
Stage 1: Connected
Devices transmit data reliably.
The organization can confirm whether an asset is online and collect basic telemetry, but there is no analytical or decision layer.
Primary Requirement: Reliable data collection and device identification.
Stage 2: Monitored
Dashboards, thresholds, and alerts provide operational visibility.
Decisions remain reactive and human-led. Operators respond after an event becomes visible.
Primary Requirement: Reliable identification and escalation of abnormal conditions.
Stage 3: Predictive
Models forecast failures, defects, delays, or anomalies before their full operational impact occurs.
A named owner is accountable for reviewing the output and acting within a defined response period.
Primary Requirement: Timely intervention based on validated predictions.
Stage 4: Governed Action
The system recommends or initiates actions within defined operational limits.
Actions may include:
Opening a maintenance ticket
Adjusting an operating threshold
Isolating equipment
Rerouting an asset
Escalating an incident
Requesting human approval
The system also includes audit records, human override, rollback, model monitoring, and fallback procedures.
Stage 4 should not be interpreted as unrestricted autonomy. In most operational environments, the appropriate target is bounded automation. The system can act within narrow, tested limits and escalate decisions outside those limits to an accountable person.
The AIoT Readiness Scorecard
Before expanding an AIoT pilot, operations and engineering leaders should assess seven conditions.
Score each condition as:
1: Defined, implemented, and tested
0: Missing, informal, or untested
1. Operational Decision
Is the exact decision supported by the system clearly defined?
2. Named Owner
Is one role accountable for reviewing or acting on the output?
3. Response Process
Are thresholds, escalation routes, response times, and fallback procedures documented?
4. Data Readiness
Are sensor quality, calibration, completeness, labeling, and contextual data continuously validated?
5. Device and Model Lifecycle
Can devices, credentials, firmware, configurations, and models be securely updated, monitored, rolled back, and retired?
6. Architecture Fit
Does the cloud-edge design reflect the deployment’s latency, connectivity, bandwidth, privacy, and safety requirements?
7. Security and Integration
Is the system integrated with the necessary operational platforms and protected by a defined security baseline?
AIoT Use Cases by Scalability
Use Cases That Scale Reliably
Predictive maintenance connected to a maintenance response process
Machine-vision inspection on stable production lines
Asset visibility integrated with operational platforms
Anomaly detection supported by severity rules and named responders
Edge inference where latency or connectivity requires local processing
Cross-site optimization using standardized device and data models
Agentic workflows operating within restricted permissions
Physical AI operating in structured and controlled environments
Deloitte reports that physical AI adoption is most advanced in manufacturing, logistics, and defense. Its 2026 enterprise research also identifies substantial interest in agentic AI across supply chain management, research and development, knowledge management, and cybersecurity. (6)
Common Failure Patterns
Dashboard projects with no operational decision attached
Predictive models with no person accountable for responding
Cloud-only systems deployed in unreliable network environments
Pilots based on unvalidated sensor data
Device fleets without update, credential, and retirement processes
Systems that require manual integration after every alert
Security controls postponed until after deployment
Future Direction of Connected Operations
AIoT systems that scale successfully will increasingly be hybrid, edge-aware, updateable, integrated, and governed by design.
Selective Growth of Edge AI
Organizations will use edge inference where latency, resilience, privacy, safety, or bandwidth requirements justify local processing.
Training, historical analysis, fleet coordination, and central management will generally remain in cloud or data-center environments.
Convergence of Device and Model Operations
Organizations will need to manage model versions, device configurations, credentials, firmware, and operating policies as parts of one system lifecycle.
A model update should not be deployed without confirming device compatibility, network requirements, rollback behavior, and operational impact.
Growth of Bounded Automated Action
More systems will move from detecting events to recommending or initiating actions.
Automation should be expanded according to:
Risk
Model confidence
Reversibility
Operational impact
Human-review requirements
Availability of fallback procedures
The appropriate objective is controlled authority rather than unrestricted autonomy.
IoT Analytics describes agentic AI as an emerging area within industrial AI but notes that practical deployment remains at an early stage. (2)
Decision-Level ROI Measurement
Organizations will increasingly evaluate AIoT systems according to operational outcomes rather than connectivity or dashboard activity.
Relevant measures include:
Unplanned downtime
Alert-to-resolution time
Inspection effort
Production throughput
Defect rates
Asset loss and delay
Energy consumption
Manual intervention
Service continuity
The next stage of AIoT will not be defined by the number of sensors deployed or the size of the underlying models. It will be defined by the quality of the decision architecture: clear ownership, reliable data, controlled automation, and systems that can be securely updated throughout their operating life.
Predictions for the Next Phase of AIoT
The following predictions are my interpretation of the evidence in this report and the operating patterns I expect business and engineering leaders to encounter as AIoT deployments mature.
1. Businesses Will Buy Decision Outcomes, Not Device Counts
I expect procurement conversations to move away from the number of sensors, dashboards, or connected sites.
Buyers will ask how much downtime was avoided, how quickly exceptions were resolved, how inspection yield changed, and whether energy, labor, loss, or safety performance improved. A device without a measurable decision outcome will become difficult to justify.
2. Device Operations and Model Operations Will Merge
Device identity, firmware, credentials, configuration, model versions, operating policies, and rollback will be managed as one lifecycle.
Organizations that keep IoT operations and machine-learning operations in separate teams and toolchains will face slower updates, weak accountability, and more production risk.
3. Small Edge Models Will Handle Real-Time Decisions
I expect smaller, specialized models to perform local classification, anomaly detection, and safety-sensitive inference, while larger cloud models handle training, fleet-wide comparison, root-cause exploration, and complex reasoning.
The winning architecture will not be edge-only or cloud-only; it will place each form of intelligence where it is operationally justified.
4. Dashboards Will Become Exception-Management Interfaces
Passive dashboards that show every signal will lose value.
Future interfaces will rank the events that require action, explain why they matter, identify the responsible role, and show the response window and recommended next step. The best dashboard will contain less information and create faster action.
5. Automation Will Expand Through Bounded Authority
Organizations will automate narrow, reversible, observable actions first.
High-impact or difficult-to-reverse decisions will continue to require approval. This approach will let businesses gain speed without handing unrestricted authority to a model operating in a physical environment.
6. Data Rights and Interoperability Will Become Procurement Terms
Contracts will increasingly define ownership of device data, labels, model outputs, event histories, and export formats.
Businesses will also require documented pathways for changing cloud providers, replacing hardware, or moving to a different operational platform. Vendor lock-in will be treated as an operating risk rather than a technical inconvenience.
Recommendations for Operations and Engineering Leaders
1. Begin With the Operational Decision
Define the decision, accountable owner, response period, required action, and business measure before selecting devices or models.
2. Validate Data Before Developing Prediction Models
Assess sensor health, missing data, timing, calibration, context, and label quality before optimizing model performance.
3. Design Lifecycle Management From the Beginning
Include device provisioning, identity, credentials, firmware, model updates, monitoring, rollback, replacement, and retirement in the initial architecture.
4. Match Architecture to Site Conditions
Base cloud and edge responsibilities on actual latency, bandwidth, connectivity, privacy, resilience, and safety requirements.
5. Integrate the Operational Response
Connect alerts and recommendations to the systems through which work is completed, including maintenance, logistics, ERP, identity, ticketing, and incident-management platforms.
6. Measure Operational Outcomes
Track downtime, response time, inspection effort, throughput, defect rates, asset losses, energy use, and service continuity. Dashboard views and alert volumes are not sufficient measures of value.
7. Expand Automation Gradually
Begin with recommendations and human approvals. Automate only actions that are narrow, observable, reversible, and supported by tested fallback procedures.
Conclusion: Connected Operations Require More Than Connectivity
AIoT systems create value when intelligence is connected to a defined operational decision.
The most effective system is not necessarily the one with the most devices, the newest model, or the most complex dashboard. It is the system that can reliably collect data, interpret it within the correct operating context, deliver it to an accountable owner, support an appropriate action, and remain secure and maintainable at scale.
For operations leaders, the central question is no longer: Can we connect the asset?
It is: Can we build a governed decision process that remains useful, secure, and maintainable after the pilot ends?
Sources and Methodology Notes
This report combines publicly available market forecasts, academic research, architecture guidance, cybersecurity frameworks, and Phaedra Solutions’ connected-systems delivery observations.
Book a free call and let’s identify where AIoT can improve operational performance without creating unnecessary integration, lifecycle, or governance risk.
Mujtaba turns product ideas into working software — fast. As a Fractional CTO and solution architect with 13+ years of experience, he leads AI-first development teams that ship MVPs in under 10 days, cut product rework by 40%, and build digital infrastructure that holds up at scale.
His work spans UX design, full-stack development, blockchain integration, and IoT — all engineered with AI-assisted tooling to reduce build time and operational cost by 30–60%.
Oops! Something went wrong while submitting the form.
Cookies Settings
We use cookies to provide you with the best possible experience. They also allow us to analyze user behavior in order to constantly improve the website for you.
