For asset owners, service providers and robotics vendors who need the next pilot to end in a work order.
Book a dive briefingThe real problem
Why industrial robotics pilots fail is almost never a hardware question. The robot flies the stack, the data is clean, and the program still stops after the demo.
We are 8scale, the market intelligence layer between industry, technology and innovation, and the publisher of Scalebook 2025, Inspection & Maintenance Robotics Market Report. We work with asset owners, industrial service providers and robotics companies across worldwide. 8scale research on inspection and maintenance robotics starts in the operating context, not in the demo hangar.
McKinsey's 2024 Manufacturing Survey reports that 87 % of robotics initiatives never scale beyond pilot, and puts a failed automation project at USD 2.3 million. Both figures are McKinsey's, as cited in Scalebook 2025. They are not our data, and they describe the symptom rather than the cause.
The cause is the Industrial Translation Gap. Three parties describe the same deployment in three vocabularies that never reconcile.
The vendor speaks in detection accuracy, autonomy level and flight time. Operations speaks in permits, isolation, turnaround windows and man-hours per shutdown. Finance speaks in payback period, CapEx approval thresholds and operating budget lines.
A pilot that produces a beautiful point cloud, but no man-hour number, no permit precedent and no budget owner, has answered nobody's question.
Before the next pilot starts, write the closing sentence each of the three sides needs to hear. If you cannot write all three, the pilot is not ready to run.
Most robots that fail in industry were validated under conditions that no brown-field plant offers. The gap between the test site and the live unit is where schedules and business cases die.
ATEX comes first. In a Zone 1 or Zone 2 area, an uncertified platform is a hard stop. Certification and Ex-rated variants carry lead times measured in quarters, and the pilot plan rarely budgets for them. In the United States, the equivalent question is the classified-area rating and the OSHA-driven permit path, not the CE mark.
Work rules come second. Confined-space entry still needs a permit, an attendant and a rescue plan, even when a crawler goes in instead of a technician. Night-shift restrictions, radio protocols, lockout-tagout and, in Germany, works council agreements on data capture all shape when a robot may operate at all.
Integration comes third and is usually underestimated. Decades-old PLCs speak Modbus RTU, not OPC UA. Historians are air-gapped, the tank farm has no LTE coverage, and GPS is unavailable inside steel structures. The IT and OT security review of a vendor cloud can take longer than the entire pilot.
None of this is exotic. It is the standard operating context of energy, chemicals, maritime, manufacturing and infrastructure assets, and it is the ground truth we test in our work on industrial robotics implementation.
Run a site readiness check before you shortlist a platform: Ex zone map, permit path, network path, data landing zone, and the named person who signs each one.
Brown-field reality
Shared definition
Beyond pilot is a testable state, not a feeling in the room. Most programs stall because nobody wrote down what would count as deployment before the money was spent.
We use five conditions, and all five must be checkable by the operator, the service provider and the technology company.
One: the inspection exists as a recurring work order in the CMMS, scheduled like any other maintenance task. Two: it sits in an operating budget with a named owner, not in an innovation fund. Three: there is a procurement route, a framework agreement or a service contract, that survives the champion leaving.
Four: the output is accepted by whoever signs the inspection. That is the internal inspection authority, the class society or the notified body, depending on the asset. A report that engineering likes but the certifying party will not accept has no path to production.
Five: on the third campaign, the vendor's application engineer is not on site. If the crew cannot run it, you have bought a service disguised as a product.
Add the counterfactual explicitly. Name what the robot replaces or defers: rope-access shifts, scaffolding meters, confined-space entries, shutdown hours. Without that line, savings stay rhetorical.
This definition is deliberately unflattering to everyone. It stops vendors from selling autonomy that is really teleoperation, and it stops operators from calling a repeated trial a rollout.
Write the five conditions into the pilot scope, signed by all three sides, and set the review date on which you test them.
Capital committees approve baselines, not market forecasts. A growing category tells you the technology is real; it has never funded a single work order.
For context: the inspection and maintenance robotics market was USD 2.89 billion in 2024 and we forecast 15.8 % CAGR to 2030 (cite as: 8scale Scalebook 2025). That number belongs in a strategy deck, not in your approval paper.
What moves an approval paper is your own operating baseline. Power and utilities operators spend roughly US $400 billion a year on inspection and maintenance globally, base year 2024 (cite as: 8scale Scalebook 2025), see inspection and maintenance spend by industry for the other seven industries. Your share of that spend, per asset class, is the denominator the robot has to move.
Measure before, not after. Useful evidence is asset-level: confined-space entries avoided per year, permits raised, scaffolding cost per campaign, wall-thickness readings per shift, false-call rate, re-inspection rate, and turnaround days on the critical path.
Then apply the threshold honestly. A platform that is 20 % faster loses to procurement friction, training cost and change resistance. Adoption in inspection generally needs a step change, which is the argument behind the 10X improvement threshold.
State residual uncertainty in the same paper. Name what the pilot could not measure, seasonal access, corrosion progression over two cycles, insurer acceptance, instead of smoothing it into a single ROI figure. Committees discount hidden risk far harder than declared risk.
Build the baseline before the pilot begins. A pilot without a pre-measured baseline cannot produce evidence, only impressions.
Evidence for approval
Pilot to production
Production is a different transaction, run by different people, on a different budget. Teams that expect a smooth continuation of the pilot are the ones that stall.
Ownership moves first. The pilot belongs to innovation, digital or engineering; production belongs to maintenance planning and the plant operating budget. If the maintenance planner will not take the work order, the program ends when the pilot ends, whatever the results said.
The commercial model changes next. Pilots are paid from project money; fleets need either a CapEx case with a depreciation schedule or a service contract with a per-inspection or per-asset rate. Which of the two you choose decides who carries utilization risk, spares and obsolescence, the argument we set out in CapEx versus Robotics-as-a-Service.
The operating burden becomes visible. Spares and MTBF, calibration, certified internal pilots and their license renewals, battery logistics, and a service level agreement with a response time that matches your outage plan. None of this appears in a four-week trial.
Data is the quiet blocker. Reports must land in the CMMS and the historian in a format the certifying party accepts, with a clear answer on who owns the raw data and where it is stored.
One caution on platform choice: legged robots are the fastest-growing of the eight segments we track, at 25 % CAGR (cite as: 8scale Scalebook 2025). Fast growth is not deployment readiness in a Zone 1 area, and segment momentum is a poor substitute for Ex certification.
Name the production owner, the contract model and the data path before the pilot closes out.
COMMON QUESTIONS
They answer a technology question and skip the deployment question. The robot performs, but nobody agreed which inspection it replaces, who owns the budget line afterward, whether the certifying party accepts the output, and how it clears ATEX, permits and legacy control system integration.
No. The 87 % of robotics initiatives that never scale beyond pilot, and the USD 2.3 million per failed automation project, are McKinsey’s 2024 Manufacturing Survey, as cited in Scalebook 2025. Our own figures come from Scalebook 2025 and are marked as such.
Long enough to cover the real access conditions, which usually means at least two inspection cycles or one full turnaround window. Four-week trials in fair weather prove flight time, not fitness. Set the review date and the five deployment conditions before day one.
Asset-level KPIs against a pre-measured baseline: confined-space entries avoided, permits raised, scaffolding cost, readings per shift, false-call and re-inspection rate, and turnaround days removed from the critical path. Modeled savings without a baseline do not survive a capital committee.
It solves the approval threshold, not the deployment gap. RaaS moves spend from CapEx to OpEx and lowers the sign-off barrier, but permits, Ex certification, data acceptance and work-order ownership remain unchanged. Read our market sizing and research methodology for how we test such claims.
RELATED READING
01
Product-market fit, sales-cycle length and missing validation evidence sink industrial tech ventures. Read why, then book a 30-minute market briefing.
02
A staged implementation strategy for industrial robotics, with realistic timelines and outcome measures for plant teams. Book a 30-minute briefing.
03
Where AI and automation help inspection and maintenance service work, and where they do not. Read the analysis and book a 30-minute market briefing.
GET STARTED
Thirty minutes, one insight about your market, no pitch. Bring the pilot that stalled or the deployment you are scoping, and we will tell you what the evidence in inspection and maintenance robotics currently supports.
Book a dive briefing