01Deployment guide
From an operational problem to a fleet in production
An eight-step method to decide whether a robot solves your problem, get ready, choose a vendor, design a measurable pilot and scale safely and compliantly.
Last reviewed: 2026-11Methodology & sources →
The criteria, checklists and questions in this guide are our own editorial framework, built on the site's taxonomy. Regulatory and market facts carry their source alongside.
01
From problem to use case
Start from the problem, not the robot. A use case makes sense when the task repeats, the environment can be bounded and the data or the avoided risk justify the investment.
A
Dull, dirty or dangerous task
Inspection rounds, gauge reading, cargo transport, work at height or in confined spaces. The more human risk you remove, the easier the case is to justify even when direct savings are modest.
B
Boundable environment (ODD)
The operational design domain (ODD) describes where and when the robot must work: zones, schedules, weather, lighting, surfaces, people, connectivity and GNSS. If you cannot describe it, you cannot test it.
C
Value of the data
Many robots pay back through the data they capture (thermography, point clouds, periodic readings) more than through the labor they replace. Define who will use that data and in which decision.
D
Volume and frequency
A task that happens daily across several shifts pays back better than an occasional one. Estimate how many hours per year the robot would cover before talking to vendors.
Describe the ODD before requesting quotes
- Zones and routes (floor plans, stairs, ramps, doors, elevators)
- Operating hours and shifts
- Weather, temperature, dust, water and lighting
- Presence of people, vehicles and other robots
- Network coverage (Wi-Fi, private 4G/5G) and GNSS availability
- What to do outside the ODD: stop, return to base or request intervention
Red flags
- The task depends on expert judgment nobody can specify.
- The environment changes so much that no stable ODD can be described.
- Nobody in the organization will own operating and maintaining the system.
- The main goal is an innovation image, not an operational KPI.
02
Readiness assessment
Mark the status of each item. The score shows where the blockers are before a pilot; it is not a certification.
03
Buy, integrate or RaaS
The procurement model changes the risk profile as much as the choice of robot. Compare the three routes against the same criteria.
| Criterion | Buy | Robot-as-a-Service (RaaS) | Integrate or build |
|---|---|---|---|
| Upfront investment | High (capex) | Low; periodic fee (opex) | High: hardware + engineering |
| Time to value | Medium | Short | Long |
| Maintenance & updates | Yours or a separate contract | Usually included | Yours |
| Obsolescence risk | Yours | The vendor's | Yours |
| Customization | Limited to the product | Limited | Maximum |
| Vendor lock-in | Medium | High (data, software, contract) | Low, but you depend on your team |
IFR notes that flexible models such as robot-as-a-service help overcome adoption barriers and expects them to gain weight.[International Federation of…, 2026]
Total cost of ownership (TCO) structure
- 1.Hardware: robots, docks, payloads and spares
- 2.Integration: engineering, mapping, site adaptation and system connections
- 3.Software: licenses, fleet management, analytics
- 4.Operations: remote supervision, interventions and staff hours
- 5.Maintenance: preventive, corrective, batteries and sensor calibration
- 6.Insurance, regulatory compliance and cybersecurity
- 7.Training and change management
- 8.End of life: decommissioning, battery recycling and data migration
04
Evaluate vendors
Questions for the request for proposal (RFP), grouped by the taxonomy axes. Ask for written answers and, where possible, data from comparable real deployments.
- In which ODD is the system validated, and with what real operating data do you show it?
- Which autonomy level does it reach in our environment, and which tasks need human intervention?
- How many interventions per 100 operating hours do you record in comparable deployments?
- How does the system detect it has left its ODD, and what does it do then?
- Which sensors does it use to localize, and what is the fallback if one fails (GNSS, LiDAR, cameras)?
- How does it behave in low light, dust, fog, rain or on reflective surfaces?
- How often must sensors be calibrated, and who does it?
- If the system uses AI models, what data were they trained on and how are updates validated?
- Which tasks continue without connectivity and which are suspended safely?
- What latency does teleoperation need, and what happens when it is exceeded?
- For fleets, what happens if the central server goes down or the network partitions?
- How are events logged for post-incident analysis?
- Which standards does the product meet (ISO 10218, ISO 3691-4, ANSI/A3 R15.08, ISO 13482), and with what evidence?
- Which safety functions does it have, and at what performance level (PL) or integrity level (SIL)?
- How does the product fit Machinery Regulation (EU) 2023/1230 from January 2027?
- For drones, which EASA scenarios or authorisations cover the proposed operation?
- Who owns the captured data, and in which open format can we export it?
- Where are the data stored and processed, and who has access?
- How are vulnerabilities handled and reported under the Cyber Resilience Act?
- How are images of people anonymized or minimized?
- Which availability do you guarantee, and which penalties apply if it is missed?
- What are the response and repair times, and where are the spare parts?
- What exactly does the fee (RaaS) or maintenance contract include?
- Which costs appear when scaling from one robot to a fleet?
- For how many years do you guarantee support, spares and software updates?
- What happens to our robots and data if you change product or stop operating?
- Which customers with a comparable case can we contact as references?
05
Design the pilot
A pilot exists to make a decision, not to prove the robot works. Before starting, fix what you will measure, against which baseline, and which result justifies going to production.
F1
Lab or test area
- Goal
- Validate basic technical capabilities and data integration.
- Exit criterion
- The robot completes the reference mission and the data reach the target system.
F2
Controlled pilot
- Goal
- Operate in the real environment, in a bounded zone, under close supervision.
- Exit criterion
- KPIs reach the agreed thresholds over a representative period with no serious incidents.
F3
Initial production
- Goal
- Embed the robot in the daily process with the final roles.
- Exit criterion
- The business case is confirmed with real data and the team operates without the vendor on site.
Pilot KPIs
| KPI | Definition | Why it matters |
|---|---|---|
| Mission success rate | Missions completed without intervention ÷ scheduled missions | Measures real autonomy, not the advertised one |
| Interventions per 100 h | Human interventions (remote or on site) per 100 operating hours | Determines how many staff the fleet needs |
| Availability | Operating hours ÷ planned hours | Includes charging, failures and maintenance stops |
| Data quality | Valid captures ÷ total captures, per the agreed definition | Unusable data has no value even if the mission completes |
| Cost per mission | Period TCO ÷ completed missions | Allows comparison with the current cost of the task |
| MTBF / MTTR | Mean time between failures / mean time to repair | Anticipates maintenance and spare-part cost |
| Incidents & near misses | Safety events recorded per 1,000 h | A gating condition, not just another metric |
| Adoption | Actual use by teams and their assessment | Without adoption, the fleet does not scale |
There are no universal thresholds: they depend on the task, the risk and the baseline. Agree them in writing with the vendor before the pilot.
Decision points (go / no-go)
- G1Were KPIs met over a representative period, including varying shifts and weather?
- G2Does the business case hold with real operating and maintenance costs?
- G3Is there a risk assessment and regulatory conformity for production operation?
- G4Can the internal team operate the fleet without the vendor on site?
06
Safety, standards & regulation
Safety is assessed on the complete application (robot, environment, task and people), not just the product. These are the main references for Europe and the US.
| Reference | Scope | Status & dates |
|---|---|---|
| ISO 10218-1 / -2:2025 | Industrial robots (part 1) and their integration into systems and applications (part 2), including collaborative operation. | 2025 editions; replace the 2011 ones and absorb the former ISO/TS 15066.[ANSI Blog, 2025] |
| ISO 13482 | Personal care robots: mobile servant, physical assistant and person carrier robots. | 2014 edition, under revision. Excludes industrial, flying and medical robots.[ANSI Webstore, 2014] |
| ISO 3691-4 | Driverless industrial trucks (AGVs and AMRs) and their operating environment. | 2023 edition current; a 3rd edition is in development.[ISO] |
| ANSI/A3 R15.08 | Industrial mobile robots (US): robot, system integration and continued use. | Part 1 (2020, reaffirmed 2026), part 2 (2023) and part 3 (2026).[Association for Advancing…, 2026] |
| ISO 13849-1:2023 · IEC 61508 | Functional safety of control systems: performance levels (PL) and safety integrity levels (SIL). | ISO 13849-1, 2023 edition; IEC 61508, edition 2.0 (2010).[Estonian Centre for…, 2023][IEC Webstore, 2010] |
| Machinery Regulation (EU) 2023/1230 | Safety requirements for machinery placed on the EU market, with new requirements for autonomous mobile machinery and systems with evolving behavior. | Applies from 20 January 2027 and repeals Directive 2006/42/EC.[EUR-Lex — Publications Office…, 2023] |
| AI Act — Regulation (EU) 2024/1689 | Obligations by AI-system risk. AI acting as a safety component of machinery is high-risk when the machinery requires third-party conformity assessment (Art. 6(1)). | In force since August 2024. After Regulation (EU) 2026/1744, Annex III high-risk applies from 2 December 2027 and AI in Annex I products from 2 August 2028.[EUR-Lex — Publications Office…, 2024][Garrigues, 2026] |
| Cyber Resilience Act — Regulation (EU) 2024/2847 | Cybersecurity of products with digital elements, including connected robots. | Vulnerability and incident reporting from 11 September 2026; full application on 11 December 2027.[Jones Day, 2026] |
| GDPR · EDPB Guidelines 3/2019 | Processing of personal data captured by cameras and other video devices. | Version 2.0, adopted January 2020.[European Data Protection Board, 2020] |
| Regulations (EU) 2019/947 & 2019/945 (EASA) | Civil drone operation in the open, specific and certified categories. | Current framework; the specific category requires an operational authorisation or a declared standard scenario.[European Union Aviation…] |
General information as of the review date. Not legal or technical advice: consult the official text and a specialist for your case.
07
People & change management
Adoption fails because of people more often than technology. EU-OSHA recommends human-centred design and active measures to support worker acceptance.[European Agency for Safety…, 2022][European Agency for Safety…]
Involve those who do the work today, early
Operators know the exceptions the robot will meet. Bring them into defining the ODD and what counts as valid data.
Inform and consult worker representatives
Where labor law requires it, introducing new technology must go through information and consultation. Doing it early reduces resistance.
Redesign jobs, not just tasks
New roles appear: fleet supervisor, first-level technician, inspection data analyst. Define the career path and training for each.
Assess psychosocial and ergonomic risks
Working alongside robots changes pace, perceived monitoring and mental workload. Include them in the risk assessment.
08
From pilot to fleet
Scaling is not buying more robots: it is turning a hand-crafted pilot into a repeatable process across sites.
01
Standardize
Reusable site-assessment template, maps, operating procedures and acceptance criteria.
02
Integrate
Connect the fleet to maintenance (CMMS/EAM), warehouse (WMS) or control (SCADA) systems so data triggers action.
03
Govern the fleet
Centralized management of missions, software versions, maps and permissions. An ungoverned fleet multiplies incidents.
04
Secure it
Network segmentation, identity management and a vulnerability process aligned with the Cyber Resilience Act.
05
Plan maintenance
Spare-part stock, battery rotation and trained technicians per region, based on the MTBF observed in the pilot.
06
Measure as a portfolio
The same KPIs at every site to compare, prioritize and renegotiate with vendors.
Next step
Compare real platforms →
Filter by environment and compare up to three models.
Review documented deployments →
What worked, with which metrics and what was learned.
Calculate the return →
36-month payback with your own assumptions.
Configure the sensor suite →
Achievable autonomy, power, cost and resilience.