# Industrial Robots: China’s Scale Advantage Meets the Economics of the Production Cell

A procurement research report on market structure, integration risk, production economics and the evidence required before a lower-cost robot becomes a lower-cost manufacturing system.

BaiyX Research | August 2026 · Updated November 2026

## Key takeaways

- China installed about 354,000 industrial robots in 2025; Chinese manufacturers supplied 55% of that domestic market. Scale justifies a wider supplier shortlist, not automatic technical qualification.

- The investable unit is the productive cell. Application engineering, guarding, tooling, recovery and local service can outweigh the price difference between robot arms.

- Buyers should qualify one representative application, price recovery capability explicitly, and expand only after sustained production evidence.

- The report includes a reproducible cell model, an acceptance framework, supplier evidence requirements and a 90-day qualification sequence.

## The procurement conclusion

Chinese industrial robots now deserve systematic consideration in international sourcing. That conclusion is narrower than saying a Chinese arm should replace every established brand. The useful procurement question is whether a specific combination of robot, integrator, tooling and service can deliver the required production outcome at a lower risk-adjusted cost. A large domestic market provides opportunities for manufacturing scale and application learning. It does not remove the need to establish who will restore a failed cell at the destination factory, how a replacement component is commissioned, or whether the quoted motion performance survives a full production shift.

For a buyer making a first purchase, we favour a bounded pilot with a repeatable task, controlled presentation of parts, a credible local integration partner and an explicit fallback. Machine tending, orderly palletising and other applications with stable inputs can provide a cleaner learning environment than a highly variable process dependent on difficult perception or delicate contact. This is a screening judgment, not a claim that all such applications are simple. A heavy pallet, hot workpiece or shared human workspace can make an apparently straightforward job demanding. Qualification should follow the actual hazards and process, rather than the familiar label attached to the application.

The commercial opportunity has two connected layers. New robot installations generate an equipment opportunity; their continued operation creates requirements for documented spare parts, consumables, inspection and recovery support. These layers fit BaiyX’s coverage of industrial equipment alongside aftermarket parts and maintenance consumables. They should not be confused with an existing service capability. Any supplier claiming rapid recovery must demonstrate its people, parts, access rights and procedures in the customer’s geography. Our recommendation is to sell and buy a verifiable production outcome, with a support boundary that both sides can inspect before paying for the system.

## What the market data establish

IFR’s World Robotics 2026 release reports approximately 354,000 installations in China in 2025, up 20%, representing 59% of global deployments. Chinese manufacturers supplied about 195,000 units in their home market, or 55%, compared with 57% in 2024. The preceding IFR release recorded approximately 295,000 Chinese installations in 2024. These are installation statistics, not the export value of Chinese robot manufacturers, and a domestic supplier share is not a measure of a country’s share of global robot revenue.

Evidence: [IFR, China World Robotics 2025 release, 25 September 2025; 2024 observations](https://ifr.org/downloads/press_docs/2025-09-25-IFR_press_release_China_in_English.pdf); [IFR, Five Million Robots now Operate in Factories Globally; World Robotics 2026, 2025 observations, accessed 9 November 2026](https://ifr.org/news/five-million-robots-now-operate-in-factories-globally/)

The distinction matters because three different investment narratives often get compressed into one headline. Growing domestic installations indicate a large arena in which suppliers can accumulate application experience. Rising domestic supplier volume indicates commercial acceptance in that arena. Neither observation, by itself, demonstrates an established overseas support network or qualification for a buyer’s specific process. An exporter can sell a capable machine and still leave the overseas customer carrying most of the integration risk. Conversely, an internationally familiar brand may offer little practical advantage when its nearest qualified engineer is unavailable during the customer’s production window.

A movement from 57% to 55% domestic share should also be read with the denominator in view. A supplier group can gain unit volume while losing share in a growing market. That is different from losing sales, and neither outcome says much about margins without price and product-mix data. We do not infer industry profitability from installation counts. Price concessions, warranty costs, channel inventory and payment terms can offset manufacturing scale. Buyers should therefore use the market figures to broaden competitive discovery, while using project evidence to determine technical approval and counterparty evidence to determine how much financial exposure to accept.

### Exhibit 1: China: installation scale and domestic supplier volume

thousand robots

| Item | Value |
|---|---|
| China total 2024 | 295 |
| China total 2025 | 354 |
| Local suppliers 2025 | 195 |

Rounded IFR observations. Supplier volume is a subset of 2025 installations; it is not an additional market.

## Scale creates learning, but learning is application specific

A robot manufacturer’s competitive advantage can come from volume procurement, common platform design, controller development, manufacturing quality or a dense network of integrators. Those advantages do not improve at the same rate. Purchasing more motors may reduce unit cost relatively quickly. Developing reliable recovery behaviour across many fault conditions depends on field experience, disciplined software release management and a willingness to feed incidents back into engineering. The buyer should ask which type of learning the proposed supplier has actually accumulated, rather than treating shipment growth as a universal proxy for maturity.

Application experience transfers imperfectly. A supplier with many palletising references may understand payload dynamics, cycle optimisation and gripper maintenance very well, yet have limited evidence for precision dispensing. A welding integrator may be excellent at path programming and torch maintenance but less familiar with a machining centre’s interlocks and chip environment. Useful references therefore match the process physics, workpiece variability, shift pattern and service geography. A reference that shares only the same robot model proves much less. We would prioritise one accessible, closely matched customer over a long list of unrelated installations with no operating data.

This suggests a different way to construct a shortlist. Begin with the application family and the required process capability. Identify integrators that have delivered that family successfully, then examine the robot platforms they can support. In parallel, ask robot manufacturers which local partners they authorise and what technical support those partners can access. The overlap is the practical sourcing universe. A low-priced arm without a viable integration route may remain an interesting future option, but it should not consume the same qualification budget as a solution whose team can already explain the complete cycle, the likely failure modes and the recovery sequence.

## The value chain behind the quoted arm

A robot cell combines mechanical structure, transmission, motors, drives, sensing, control software, end effectors, fixtures and process equipment. The commercial responsibility may be split across several companies. That division creates interface risk: a gripper drops a workpiece, the integrator blames acceleration settings, the robot manufacturer points to payload configuration, and the tool supplier questions the air supply. Procurement needs a lead party responsible for resolving the complete fault, even when the underlying component warranty belongs to another organisation.

OSHA’s technical manual distinguishes the manipulator, controller, teach pendant and end effector, and describes their integration with surrounding process equipment. This system perspective is useful beyond the United States: the purchased arm is only one element of the application, although local legal obligations must be checked separately. The practical implication is to put the boundary of supply on a drawing and a responsibility matrix. Every cable, network link, pneumatic connection, fixture datum and upstream handshake should have an owner. Undefined interfaces tend to reappear later as change orders or commissioning delays.

Evidence: [OSHA Technical Manual, Section IV Chapter 4: Industrial Robot Systems and Industrial Robot System Safety; accessed 9 November 2026](https://www.osha.gov/otm/section-4-safety-hazards/chapter-4)

For component diligence, a named manufacturer is a starting point rather than a verdict. A reducer’s suitability depends on its model, load, duty, lubrication and integration. A servo drive’s availability matters together with firmware compatibility and parameter backup. An imported component can still be misapplied, while a domestic component can perform well in a properly designed application. Ask for the controlled bill of materials, approved alternatives and the change-notification process. A contract that fixes the component brand but permits unspecified variants can preserve the appearance of consistency while allowing economically significant changes beneath it.

### Exhibit 2: Cell responsibility map: evidence to assign before award



| Item | Value |
|---|---|
| Robot and controller | Model, controlled configuration, recovery procedure and manufacturer support |
| Tooling and fixture | Payload assumptions, gripping evidence, wear parts and changeover method |
| Integration | State logic, interfaces, fault recovery and application acceptance |
| Buyer site | Utilities, upstream process, staffing and installation readiness |
| Local service | Named capability, parts access and restoration sequence |

BaiyX procurement framework; responsibilities must be assigned in the actual contract.

## Translate the process into a measurable duty

A useful request for quotation begins with the production task. Define the workpiece mass and range, the centre of gravity, the end-effector mass, the required orientation and the path through the cell. Then add the process constraints: insertion force, dwell time, allowable acceleration, surface protection, temperature, contamination and how defective parts are identified. Payload and reach alone leave too much ambiguity. A robot may reach the destination geometrically but lack comfortable motion margins, leave cables under stress or require a slower posture that destroys the intended cycle time.

The specification should identify normal variation and exceptional conditions separately. Normal variation might include several approved part sizes, predictable positioning error and different packaging lots. Exceptional conditions might include a missing component, a malformed tray, loss of air, interrupted communications or an operator stopping the cell during a transfer. The normal sequence determines throughput; the exception sequence determines how often the operator must intervene and how long recovery takes. Both belong in the acceptance plan. A demonstration consisting only of perfect parts at a favourable orientation says little about the true labour requirement of the installed system.

For each critical requirement, define the measurement boundary. Cycle time might start when a machine signals ready and finish when the next part is clamped, rather than when the robot completes its own motion. Scrap should include damage caused during handling, not only process defects downstream. Operator time should include replenishment, clearing faults, cleaning sensors and changing recipes. A supplier that optimises the arm’s fastest movement while excluding these tasks can produce an impressive demonstration and an uneconomic factory installation. The buyer’s task is to make the economically relevant boundary impossible to overlook.

## Repeatability, accuracy and process capability

Repeatability concerns the dispersion when a robot returns to a commanded condition; accuracy concerns closeness to the intended condition. A process may tolerate a consistent offset that can be taught out, yet fail when dispersion increases under load or temperature. Another process, especially one using offline paths or multiple interchangeable fixtures, may depend more strongly on absolute positioning. These distinctions should be translated into the application’s required measurement, rather than collapsing them into a single small number printed in a brochure. ISO 9283 provides a recognised framework for robot performance criteria and testing, but the applicable test conditions still need to be stated.

Evidence: [ISO TC 299 catalogue, ISO 9283:1998, robot performance criteria and test methods; catalogue accessed 9 November 2026](https://www.iso.org/committee/5915511/x/catalogue/)

The acceptance argument should connect robot behaviour to workpiece quality. For machine tending, the critical issue may be reliable seating in a fixture despite chip contamination. For dispensing, it may be bead placement and material flow during acceleration. For welding, it may be path consistency together with process setup and workpiece distortion. A robot can meet its own positioning specification while the final process fails because the gripper, fixture or process equipment contributes more variation. The inspection plan must therefore include the output of the cell, not only an isolated motion test of the arm.

Require raw observations where they are useful. A summary stating that all samples passed can conceal a process drifting toward a limit. Time-stamped results can show warm-up behaviour, deterioration during the shift or a recurring defect after a restart. Measurement equipment and methods should be suitable for the tolerance being assessed, with uncertainty considered when results approach a boundary. It is reasonable for a first pilot to focus on a limited set of decisive characteristics. It is not reasonable to substitute an unrelated precision claim for evidence that the customer’s actual product can be made consistently under the proposed operating conditions.

## Integration is the main source of hidden exposure

The price of integration reflects much more than writing robot motion commands. Work includes fixture design, electrical drawings, process interfaces, safety engineering, programming, testing, installation, training and documentation. It also includes the time needed to resolve interactions that become visible only when the equipment is assembled. A low quote may represent efficient reuse of a proven design, or it may represent work that has not been priced. Those possibilities cannot be distinguished by the total alone. Request a work breakdown, a list of assumptions and the evidence that the proposed design has already been used in a comparable application.

A disciplined integrator should be able to describe the state of the cell at each stage: waiting, loading, processing, unloading, blocked, faulted and recovering. Ask what happens when one device restarts while the others remain powered. Ask how an interrupted job is reconciled with production records. Ask whether a part is considered complete when a command was sent, when an acknowledgement was received or when physical completion was confirmed. These questions reveal whether the design treats the production cell as an integrated process or as a set of devices exchanging optimistic signals.

Integration scope should include changeover and maintainability. A cell that achieves a fast cycle but takes hours to change fixtures may suit long campaigns and disappoint a high-mix factory. A compact layout that blocks access to a filter, cable or lubrication point can convert routine work into prolonged downtime. Invite maintenance and production staff to review the design before fabrication. Their concerns often identify costs that a purchasing comparison misses. The resulting design may be less visually dramatic than a demonstration cell, but its economic value comes from repeatable operation under the factory’s actual staffing and production mix.

## Safety must follow the complete application

The presence of a collaborative robot does not establish that the application is safe for unrestricted human interaction. The workpiece, gripper, tooling, fixtures and surrounding equipment may introduce hazards independent of the arm. The appropriate safeguarding and operating modes must follow an application-specific assessment by competent people. OSHA’s technical manual emphasises system-level evaluation and the particular importance of tasks such as teaching, maintenance and recovery. Those are also the moments when a commercial system is most likely to be used in a way that differs from its normal automatic cycle.

Evidence: [OSHA Technical Manual, Section IV Chapter 4: Industrial Robot Systems and Industrial Robot System Safety; accessed 9 November 2026](https://www.osha.gov/otm/section-4-safety-hazards/chapter-4)

For procurement, the actionable requirement is a documented allocation of responsibilities. The supplier should state what assessment it performs, which information it needs from the buyer, what validation it supplies and what remains to be completed at the destination site. The buyer should provide actual operating practices, including how jams are cleared and who enters the cell. A generic certificate for a robot model cannot resolve these application questions. Equally, a buyer should not demand an undefined promise to comply with every worldwide standard; the intended market, applicable requirements and evidence expected at each milestone need to be specified.

Safety-related design changes can alter the business case. Additional separation, guarded loading or reduced operating speed may affect footprint, labour and throughput. It is better to discover this before commercial award than after the original payback has become an internal commitment. Qualification should therefore put safety feasibility before final cycle-time pricing. Where a proposed workflow depends on people working close to moving equipment, the test plan needs to reflect the validated operating arrangement. Comparing a guarded production cell with an unrestricted sales demonstration creates a misleading productivity benchmark and an avoidable conflict between the commercial promise and the usable installation.

## Software ownership and the cost of dependence

Controller software is a long-lived operating dependency. The buyer needs access to backups, configuration, program versions, communication settings and the documentation required to recover after hardware failure. Access should be practical, not merely stated in a contract. During acceptance, have the nominated maintenance team demonstrate a backup and restore procedure on an agreed safe test configuration. Confirm that required licences, passwords and vendor tools remain available when a service contract expires or a distributor changes. A machine that can be operated but cannot be restored independently carries an exposure that the original equipment price does not show.

Separate ordinary production software from remote support access. The supplier may reasonably need a controlled route to diagnose faults, but permanent unrestricted access is not the only way to provide service. The buyer’s information-technology and operational-technology teams should approve account ownership, access logging, update procedures and network segmentation. These are project requirements, not a claim that a particular national origin creates a particular cyber risk. The same diligence applies to all suppliers. The commercial question is whether the system can receive support without giving away uncontrolled operational authority or making production dependent on an undocumented external service.

Software updates also require a validation policy. A new release may fix one fault while changing timing or compatibility elsewhere. Define who approves updates, how the old configuration is retained and when regression checks are required. For a multi-cell fleet, maintain a configuration register rather than assuming identical-looking robots contain identical software. That register becomes especially valuable when spares are shared. A replacement controller sitting on a shelf is useful only if its hardware, firmware, licence and configuration can be brought into service within the recovery time assumed in the financial model.

## A worked cell investment case

Consider an explicitly hypothetical machine-tending project with 4,000 scheduled hours annually. Two technically plausible proposals target the same output. Proposal A costs $95,000 installed and Proposal B costs $125,000. The figures include the arm, tooling, guarding, integration and commissioning; they are invented analytical inputs, not quotations or market averages. Both initially assume annual gross benefits of $72,000 from avoidable overtime and redeployed labour that the factory can actually use. Annual incremental maintenance and support cost is $8,000 for A and $6,000 for B. Before reliability differences, their annual net benefits are therefore $64,000 and $66,000.

Simple payback would favour A at about 1.48 years, versus 1.89 years for B. That comparison is incomplete if the support arrangements differ. Suppose the buyer estimates 100 annual hours of production loss attributable to A and 40 hours for B, valued at $300 per hour of genuinely unrecoverable contribution. The annual penalties become $30,000 and $12,000, leaving annual net benefits of $34,000 and $54,000. Payback changes to approximately 2.79 years for A and 2.31 years for B. This does not prove that expensive equipment is better; it shows that a reliability assumption can reverse a price-based ranking.

The next diligence priority follows directly. Rather than spending another week negotiating a small equipment discount, establish whether the 60-hour downtime difference is credible. Obtain matched references, examine fault logs, validate local response capability and test recovery. If production can be rescheduled without losing contribution, the $300 hourly penalty should fall. If the cell constrains a bottleneck and missed deliveries create additional costs, the penalty may rise. A good model exposes the assumption that drives the decision and tells the procurement team what evidence is worth purchasing before it commits capital.

### Exhibit 3: Hypothetical annual net benefit after downtime

US$ thousand / year

| Item | Value |
|---|---|
| A before downtime | 64 |
| B before downtime | 66 |
| A after downtime | 34 |
| B after downtime | 54 |

Illustrative model: $72k gross benefit, maintenance $8k/$6k, downtime 100/40 hours at $300/hour. No supplier claims.

## Discounted cash flow and the break-even reliability premium

Extend the hypothetical case to five years using a 10% annual discount rate, constant year-end net benefits and no residual value. The five-year annuity factor is approximately 3.7908. Proposal A has a net present value of $34,000 multiplied by 3.7908, less $95,000, or about $33,887. Proposal B produces $54,000 multiplied by 3.7908, less $125,000, or about $79,703. Taxes, financing, working capital and inflation are excluded deliberately so the reader can reproduce the operating comparison. An investment committee should add its own treatment of those items before using the result for approval.

The premium for B is $30,000. Its annual maintenance advantage is $2,000. To justify the remaining premium over five years at the assumed discount rate, B requires additional annual operating benefit of approximately $5,914: $30,000 divided by 3.7908, less $2,000. At $300 per lost production hour, this corresponds to about 19.7 hours of annual downtime avoided. This is a more useful negotiation threshold than an unsupported statement that one brand is more reliable. The supplier can be asked to show why its service design plausibly avoids at least that amount of loss.

There is an important asymmetry in using averages. A small average annual downtime estimate can hide a low-probability outage lasting several weeks. If the buyer has contractual delivery penalties or a single dependent customer, the distribution of outcomes matters. A spare controller, a manual fallback or an alternative production route may be more valuable than a slight improvement in the robot’s average cycle. Evaluate the severe but credible recovery scenario separately from the base-case cash flow. Expected value is useful for comparing ordinary operating costs; it is not a substitute for deciding how much interruption the business can survive.

### Exhibit 4: Hypothetical five-year net present value

US$ thousand

| Item | Value |
|---|---|
| Proposal A | 33.887 |
| Proposal B | 79.703 |

10% discount rate, annual net benefit $34k/$54k, investment $95k/$125k; no tax, financing or residual value.

## Labour savings must be cashable or productive

Automation proposals frequently count every minute of displaced manual activity as a wage saving. That is rarely the correct starting point. A worker may remain necessary for loading, inspection, replenishment or other tasks on the same line. The project then releases time rather than removing a paid position. Released time has value if it reduces overtime, avoids a planned hire or supports additional profitable output. It has less immediate financial value if no one has identified how it will be used. The financial model should name the operational change that converts time into economic benefit.

A useful labour study records activities across the full shift, including low-frequency interventions. If the operator spends less time transferring parts but more time handling faults, cleaning sensors and managing small batches, the net improvement may be modest. Conversely, a cell may produce valuable unattended hours that the current workforce cannot cover, even if it changes daytime staffing very little. The analyst should distinguish direct labour reduction, capacity expansion, ergonomic improvement and quality improvement. Combining these benefits is legitimate only when the same improvement is not counted twice and when the resulting production can be sold or otherwise used.

The buyer should also budget training and ramp-up realistically. Initial weeks may require more engineering attention than the mature process. That temporary burden is not evidence of failure, but it is a real cash and capacity requirement. Set a ramp schedule with defined milestones rather than assuming full benefit from the installation date. If the project depends on one highly skilled employee, train a second person and document routine recovery. Otherwise the organisation has exchanged a manual labour constraint for a specialist availability constraint, which may be harder to see until the first holiday, shift change or employee departure.

## Spares and maintenance are a recovery system

A spare-parts list is useful only when connected to failure consequences and recovery procedures. Separate consumables used predictably, service parts with observable wear and rare components whose absence would stop the cell. The stocking decision should consider lead time, interchangeability, shelf conditions, obsolescence and the ability to diagnose the fault. Buying an expensive spare without the software or skill to install it may provide little protection. Equally, keeping no spare because the manufacturer promises international express delivery can leave a factory exposed to customs delays and configuration work that transport time does not capture.

For the first cell, ask the supplier to identify the smallest support package that meets the buyer’s target recovery time. That package may include a gripper wear kit, sensors, cables, batteries, documented settings and access to an exchange drive or controller. It need not include every major mechanical assembly. As the fleet grows, pooled stock can become economical, but only if configurations are controlled. The decision to standardise should therefore weigh spare-parts commonality and technician familiarity alongside purchase price. A portfolio of individually cheap but incompatible cells can be expensive to maintain.

Maintenance planning also needs a feedback loop. Record the symptom, root cause, part replaced, elapsed downtime and recurrence. A rising frequency of cable faults may justify a routing change rather than more cable inventory. Repeated gripper adjustments may indicate fixture or workpiece variation rather than poor robot precision. The strongest aftermarket relationship is one that reduces avoidable failures and makes necessary maintenance predictable. This is a commercially durable proposition for an industrial supplier because it ties repeat business to improved operating outcomes, rather than depending on the customer repeatedly buying parts for a problem that remains unresolved.

## How to qualify a supplier without confusing confidence with evidence

We propose four gates. The first is application feasibility: the supplier can explain the process, identify critical assumptions and demonstrate relevant experience. The second is technical evidence: the proposed configuration and test plan address the buyer’s actual requirements. The third is delivery capability: the manufacturing, integration and service teams have capacity and a credible schedule. The fourth is counterparty resilience: the contracting entity, warranty issuer and payment structure are understood. A supplier that fails an essential gate should not compensate by scoring well on unrelated presentation or price criteria.

Evidence should be proportional to the exposure. A low-value noncritical accessory does not justify the same diligence as an integrated production bottleneck. For a first robot cell, however, an organised supplier should be able to provide controlled drawings, a bill of materials, a preliminary acceptance plan, reference contacts and an escalation route. Site visits or remote demonstrations should be structured around unanswered questions, not a general tour. Ask to see how changes are approved, how failed tests are recorded and how finished systems are identified. The discipline surrounding a fault can be more revealing than a flawless demonstration prepared for visitors.

A scoring sheet can assist comparison, but it should show evidence quality separately from the numerical score. A claim supported by a brochure is different from a result witnessed on the buyer’s part. Use labels such as observed, documented, referenced and unverified. Then identify the few unverified assumptions that could reverse the investment decision. The purpose of diligence is to reduce uncertainty where it matters, not to create a large file. A concise record connecting each critical requirement to its evidence, owner and acceptance test is more valuable than hundreds of pages of generic certificates that never address the production task.

### Exhibit 5: Qualification gates and stop conditions



| Item | Value |
|---|---|
| Application | Stop if the supplier cannot explain the actual duty and exception handling |
| Technical | Stop if decisive requirements remain unsupported by relevant tests |
| Delivery | Stop if integration capacity or destination recovery has no credible owner |
| Counterparty | Stop if seller, warranty issuer or payment exposure remains unclear |
| Evidence labels | Observed / documented / referenced / unverified; do not treat them as equivalent |

BaiyX analytical framework. These are screening gates, not certification.

## Factory acceptance and site acceptance have different jobs

Factory acceptance should establish that the configured cell can perform the agreed functions before shipment. It should include representative parts, expected variations, safe recovery from defined faults, documentation and training readiness. The test needs a pre-agreed duration and a method for recording stoppages, interventions and defects. If the buyer cannot supply enough material for a meaningful production run, that limitation should be explicit. A short demonstration can confirm basic function; it cannot honestly establish long-term reliability or capability across all planned variants.

Site acceptance should establish performance in the destination environment. Power quality, air supply, foundations, networks, upstream equipment and operator practices may differ from the supplier’s facility. The contract should specify which buyer-provided conditions must be ready, how deviations are recorded and how their effects are separated from supplier defects. This avoids a familiar dispute in which every underperformance is attributed either to the factory or to the integrator without a shared measurement basis. It also allows commissioning to proceed methodically, with unresolved issues listed and owned rather than hidden beneath a ceremonial sign-off.

Commercial milestones should follow these evidence gates. Payment terms are negotiated matters, but a structure that releases nearly all funds before functional evidence leaves the buyer with limited practical leverage. Conversely, withholding payment for unspecified future perfection makes a reasonable supplier price in risk. Define measurable deliverables, a punch-list process and a bounded remedy path. Preserve configuration records from both tests so later performance changes can be investigated. The acceptance package becomes the baseline for maintenance and future replication, not merely the document required to close the capital expenditure file.

## The strongest argument against a new platform

The counterargument to adding a lower-cost robot platform is often compelling: standardisation itself has economic value. A factory with trained technicians, validated software, existing spares and a reliable incumbent integrator may rationally pay a premium to extend the installed base. The marginal cost of another familiar cell can be lower than the apparent equipment quotation suggests. A new platform requires learning, additional stock, new documentation and management attention. These transition costs are especially important when the buyer has only a small number of cells over which to spread them.

A second objection is process risk. Where the application sits on a critical bottleneck, has demanding validation requirements or cannot tolerate an extended learning curve, a less proven supplier may need to offer more than a modest discount. The alternative is not necessarily to reject the supplier permanently. It may be to qualify the platform in a less critical application first, build a record of service performance and preserve an option for later competition. This staged approach is valuable because it separates learning from a high-consequence production commitment. It also gives the new supplier a fair route to establish credibility.

Our recommendation changes when the evidence changes. A new platform with strong local support, a closely matched installed reference and a successful production pilot may deserve expansion. A familiar brand with weak support, closed recovery procedures or poor configuration control may deserve a challenge. Country labels and brand reputation are inputs to discovery, not substitutes for a project decision. The buyer should be willing to pay for demonstrated reduction in total operating risk, while refusing to pay an indefinite premium for reputation that no longer translates into measurable service, capability or production advantage at the relevant factory.

## Where the first application should earn its place

A first robot project should be selected partly for its learning value. The ideal pilot makes the important uncertainties visible without exposing the entire factory to them. For machine tending, examine how often the machine is genuinely waiting for a person, whether the door and chuck interfaces are accessible, and whether the workpiece can be presented consistently. If the cutting process dominates the cycle and the operator already tends several machines efficiently, an arm may create less benefit than expected. If an otherwise productive machine is idle overnight because staffing is unavailable, the same application can have a much stronger case.

Palletising creates a different set of constraints. The maximum package weight is only one input. The stability of the stack, variation in cartons, pallet condition, wrapping sequence and available floor area affect the design. A robot capable of moving the nominal load may still require a specialised end effector or slower motion for unstable products. The acceptance sample should include the least favourable approved package, not merely the easiest one. An awkward final layer or damaged incoming pallet can generate interventions that are invisible in a short video but material over a full shift.

For assembly, the central question is often whether variation has been controlled upstream. A person compensates for misalignment, burrs, soft components and ambiguous orientation almost unconsciously. Automating that compensation may require better fixtures, sensing or a redesign of part presentation. It can be cheaper to improve the upstream process than to purchase increasingly sophisticated perception. The buyer should compare these routes explicitly. Automation is most attractive when the task is stable enough for reliable repetition, or when the value of solving its variability clearly exceeds the engineering and maintenance burden introduced by the solution.

Welding and finishing require particular care because robot motion is only part of process quality. Tool condition, consumable management, material variation and process parameters may dominate the result. In such cases, a qualified process specialist is central to the supplier team. The RFQ should describe the required finished output and inspection method, while the supplier explains how robot control interacts with the process equipment. This protects the buyer from an attractive arm quotation that leaves the actual production technology undefined. It also gives a capable integrator room to propose a different fixture, sequence or cell architecture when that produces a stronger operating result.

## A compact operating dashboard for the pilot

The pilot dashboard should distinguish availability, speed and quality rather than hiding everything in a single utilisation figure. Record scheduled production time, time blocked by upstream or downstream equipment, faults attributable to the robot cell, actual cycle distribution and accepted output. A robot that appears underutilised may simply be waiting for its process. A cell that runs continuously may be producing defects. Separating these states prevents the wrong corrective action and makes supplier discussions more precise. The dashboard should also record operator interventions per hundred cycles and the total minutes required to resolve them.

Measure recovery from specific events. Useful examples include a deliberately rejected part, a normal controlled stop, a loss of a noncritical communication signal and a change of approved product recipe. Tests involving faults must be designed and authorised by the responsible engineering and safety personnel; the purpose is to evaluate planned behaviour, not to improvise hazardous interruptions. Record whether the operator could recover using the provided instructions and whether material status remained clear. A recovery that succeeds only when the integrator’s best programmer is present is not yet a transferable production capability.

Review trends weekly during ramp-up. If cycle time is improving but intervention frequency is not, the constraint may be part presentation rather than motion programming. If downtime falls only because the supplier keeps an engineer permanently on site, separate that temporary support from the expected steady-state arrangement. If output is stable while maintenance time rises, investigate whether the apparent reliability is being purchased through unsustainable attention. These distinctions turn the pilot into a decision instrument. Expansion should follow demonstrated repeatability of the operating model, including the people and support it requires, rather than a single record-breaking shift chosen for presentation to management.

### Exhibit 6: Pilot dashboard: separate the causes of lost output



| Item | Value |
|---|---|
| Availability | Scheduled time, robot-cell faults and elapsed recovery |
| Process flow | Upstream starvation and downstream blockage recorded separately |
| Performance | Actual cycle distribution using the complete cell boundary |
| Quality | Accepted output and handling-related defects |
| Human effort | Interventions per 100 cycles and total intervention minutes |

BaiyX proposed measurement framework; site personnel define safe test procedures.

## A 90-day decision programme

During the first month, establish the process baseline and narrow the application. Measure current output, interventions, quality loss and labour usage. Define the production boundary, identify critical hazards and map the interfaces with existing equipment. Create a common request for quotation and ask shortlisted teams to state their assumptions explicitly. The deliverable is not a broad market survey; it is a comparison that exposes why two proposals differ. If suppliers cannot quote against the same task, resolve the specification before attempting a price negotiation.

During the second month, close the assumptions most likely to change the investment decision. Use sample trials, reference calls, service demonstrations and a design review. Require a credible recovery plan and a configuration-controlled offer. Update the cash-flow model with observed cycle time, realistic operator involvement and the buyer’s own contribution-loss estimate. Run a downside case with slower ramp-up and longer recovery. A project that works only under the supplier’s most optimistic combination of assumptions should be redesigned or deferred, even when the headline payback is attractive.

The first purchase should also create a reusable evidence file. Preserve the original requirement, assumptions, accepted configuration, test results, service contacts and the actual ramp-up record in a format the buyer controls. When the next cell is proposed, compare it with this record rather than reopening every question from memory. That discipline reduces dependence on individual salespeople and engineers, makes supplier changes easier to assess and helps distinguish a genuine platform improvement from a different presentation of the same capability. It is a modest administrative investment with practical value across the equipment and aftermarket lifecycle.

During the third month, conclude the qualification decision and prepare the execution contract. Agree the acceptance protocol, responsibilities, milestone evidence, support package and training plan. Record which facts are established and which remain conditional until site commissioning. For BaiyX, the defensible commercial role is to make these technical and commercial interfaces easier to verify across borders. The long-term opportunity is not defined by the cheapest arm on a marketplace. It lies in helping an overseas customer acquire a productive system, maintain it with traceable parts and make its next equipment decision with a better operating record than it had before.

## Method and sources

Data last reviewed: 9 November 2026. Market statistics retain their stated observation years. Buyer frameworks and financial scenarios are BaiyX analysis, not supplier performance claims or market forecasts. Illustrative inputs must be replaced with measured project data before investment approval.

- [IFR, China World Robotics 2025 release, 25 September 2025; 2024 observations](https://ifr.org/downloads/press_docs/2025-09-25-IFR_press_release_China_in_English.pdf)

- [IFR, Five Million Robots now Operate in Factories Globally; World Robotics 2026, 2025 observations, accessed 9 November 2026](https://ifr.org/news/five-million-robots-now-operate-in-factories-globally/)

- [OSHA Technical Manual, Section IV Chapter 4: Industrial Robot Systems and Industrial Robot System Safety; accessed 9 November 2026](https://www.osha.gov/otm/section-4-safety-hazards/chapter-4)

- [ISO TC 299 catalogue, ISO 9283:1998, robot performance criteria and test methods; catalogue accessed 9 November 2026](https://www.iso.org/committee/5915511/x/catalogue/)