An AGV or AMR proof of concept should not stop at one question: Can the robot move?
The useful question is whether the project can produce enough evidence for the next investment decision.
Where is operating capacity being lost? How much time is lost? Is the constraint path choice, right of way, shared equipment, task release, or something else? What should the customer fund next?
Rovnou is looking for companies that want to answer these questions using operational data they are authorized to provide. Our primary focus is AGV and AMR operations in manufacturing and e-commerce logistics.
The short answer
A useful PoC starts by designing a decision, not a demonstration.
Before selecting an algorithm or integration method, agree on five things:
- A recurring operational problem
- Data that can reveal that problem
- One primary decision metric
- The person and decision that follow the result
- Ownership boundaries for data, safety, and control
If these are missing, a technically successful PoC can still fail to reach production.
The MHI Intralogistics Robotics Survey identifies finding and achieving ROI as a leading robotics adoption challenge. A survey by the Japan Institute of Logistics Systems also identifies investment limits, uncertainty about outcomes, and internal or external alignment as obstacles to logistics transformation. The gap is not only access to technology. It is the absence of decision-grade evidence.
Separate the demo, PoC, pilot, and production stages
Ambiguous labels create ambiguous expectations.
| Stage | Purpose | Output to production robots |
|---|---|---|
| Product demo | Understand the idea and basic behavior | None |
| Historical log review | Reconstruct current operations and assess data | None |
| Counterfactual evaluation | Compare alternative decisions in a replay environment | None |
| Live shadow | Read live inputs and compare recommendations | Recommendations are not sent |
| Limited pilot | Measure field behavior within an agreed boundary | Limited |
| Production | Operate continuously | According to agreed ownership |
Replaying historical logs is not live shadowing. A counterfactual result is not a measured improvement in a live facility. Keeping these stages separate makes the result easier to trust and the next decision easier to make.
Start with a read-only operational log review
Rovnou's proposed first phase does not send commands to an existing FMS or robot fleet.
We begin with exported data that the customer is authorized to provide. A production control connection is assessed separately. We do not send commands to a site before reviewing the interface, update rate, control authority, and safety ownership.
The minimum useful data usually includes:
- timestamp;
- robot ID and, where relevant, fleet or model;
- operating state such as moving, waiting, stopped, charging, or fault;
- a position, node, or zone;
- task ID, destination, and waiting reason where available; and
- timestamps for incidents the operations team already considers important.
The data does not need to be perfect. Missing events, clock drift, and inconsistent state definitions are themselves important findings when deciding whether a later integration is viable.
Who is a good fit for this PoC?
Fleet size alone is a poor qualification rule. A 30-robot site with daily stoppages, usable logs, and accountable owners can be a better validation partner than a 300-robot site that cannot provide data.
We want to work with companies where several of these conditions are true:
- repeated waiting at intersections, narrow aisles, chargers, lifts, doors, or other shared resources;
- material operator effort spent on manual recovery, redispatch, or priority changes;
- additional AGVs or AMRs have not produced the expected throughput;
- multiple fleets or vendors share infrastructure;
- it is difficult to distinguish an FMS, WCS, WES, equipment, or task-generation problem;
- operational events or logs can be exported; and
- operational, data, and budget owners can review the result.
Multiple vendors and a large fleet are useful signals, but they are not mandatory. The essentials are a real problem, observable data, and someone who will use the answer.
Choose one primary metric
Too many metrics make it easy to select only a favorable result. We therefore agree on one primary KPI before analysis.
Possible metrics include:
- waiting time attributed to traffic;
- number of manual recoveries or operator minutes;
- dwell time in shared zones;
- task completion time;
- completed tasks per hour; or
- operational headroom before adding more robots.
We then examine several known incidents and test whether the problem is observable. A conclusion that the site should change equipment logic or task release rather than path coordination can still be a valuable result.
Why customer-specific work is paid
The initial fit conversation is free. Normalizing customer data, reconstructing operations, comparing causal hypotheses, and producing a decision-ready report is customer-specific work and is normally paid.
Payment does not mean promising a positive result. It means both sides commit to a defined evaluation:
- the customer provides data it is authorized to share and explains the operating context;
- Rovnou works against the agreed scope, metric, and deliverables;
- no fixed improvement rate is assumed in advance;
- a negative fit conclusion is reported rather than hidden; and
- the go and no-go conditions are agreed before the work begins.
Price and timing are proposed after the data format, scope, replay requirements, and security conditions are understood.
Example deliverables
Depending on the agreed scope, deliverables may include:
- A data-quality review covering fields, gaps, clocks, and state definitions
- A time and location breakdown of waiting, stoppages, and manual recovery
- A timeline reconstruction of incidents selected by the customer
- Hypotheses covering routes, right of way, equipment, tasks, or communications
- A comparison of current operations and selected alternatives
- Evidence for a live shadow, limited pilot, or no-go decision
We prioritize the output required for a decision instead of building a polished dashboard first.
Companies we want to work with
AGV and AMR operators
Manufacturers, logistics operators, 3PLs, and e-commerce fulfillment teams that need to decide how to reduce waiting and recovery work.
System integrators and WCS or WES providers
Teams that want to quantify a customer's fleet problem before selecting hardware or making a production connection.
AGV and AMR vendors
Vendors that want to use post-deployment operating data to review layout, shared infrastructure, routing, or task-release constraints with their customers.
Rovnou is an operations optimization product that coordinates paths and movement timing across multiple AMRs. If an initial validation establishes a fit, live inputs, standard or proprietary APIs, and a bounded field pilot can be assessed as separate stages.
From first conversation to a decision
- Spend 20 to 30 minutes reviewing the incident, available data, owners, and next decision
- Confirm data authority, confidentiality, retention, and safety or control boundaries
- Propose the primary metric, scope, deliverables, price, and stop conditions
- After agreement, receive the data and reconstruct current operations and selected incidents
- Review the result together and make the next go or no-go decision
You do not need a complete specification for the first conversation. One recent incident and a list of available log fields are enough to assess fit.