How to Model ROI for an Interior-Cleaning Robot

Hardware Is Only ~38% of What You Pay

An honest TCO has seven line items, not one. Hardware (the robot, arm, sensors, compute) is the headline number on the quote, but across a 5-year service life it is typically only one-third to two-fifths of what you actually spend.

The seven categories. (1) Hardware, amortized straight-line over expected service life. (2) Install and commissioning — site survey, electrical, mounting, mapping, safety review — typically 8–12% of CapEx for indoor mobile manipulators [3][7]. (3) Service contract or maintenance-as-a-service (MaaS), generally priced at 10–15% of CapEx per year for industrial-grade platforms [5][7]. (4) Consumables and energy — pads, filters, cleaning solution, and the ~200–600 W average draw of a typical mobile platform with manipulation. (5) Operator supervision — the share of a human FTE allocated across the fleet (Section 04). (6) Software, cloud, and OTA updates — fleet manager subscription, telemetry storage, security patching. (7) Downtime and spares reserve — a planning buffer for the inevitable MTBF events.

The number that surprises CFOs: the recurring lines (service + supervision + software + consumables) often sum to 45–55% of 5-year TCO, even when supervision is shared efficiently. That is why a $20k discount on hardware moves payback far less than a 10% improvement in operator-to-robot ratio. The next two sections show why.

Cycles Per Bay Per Shift — The Right Denominator

Vendors quote cycle time (minutes per car). Operators care about cars completed end-to-end per bay per shift. The two are not the same number. The bridge between them is three multipliers, each of which compounds.

Cycle time is the nominal robot-only time on a clean, in-spec vehicle. Bay turnover time adds the human work of pulling a vehicle in, plugging keys in, and pulling it out — typically 90–180 seconds per car in a high-volume site. Exception rate is the percentage of vehicles that escalate to a human for any reason: a child seat that needs lifting, a stain that exceeds the soft-tool's spec, a pet-hair load the vacuum cannot finish in budget. Even a 10% escalation rate, if each escalation takes 6 minutes of human attention, is the equivalent of pulling 36 minutes of operator labor out of every hour of fleet operation.

Why this matters for ROI math. Two robots with identical 12-minute cycle times can have a 25% gap in cars per shift purely from differences in exception rate. The exception rate is also the slowest-improving variable — it depends on perception robustness, gripper coverage of weird objects, and dispatch logic, all of which are software-improvement curves measured in quarters, not weeks. A model that does not separate cycle time from exception rate will systematically over-promise.

The right benchmark to ask your vendor for is not minutes-per-vehicle but autonomous completion rate at a stated cycle-time budget: what percentage of cars finish without human intervention if the robot has, say, 14 minutes to do them. That single number, plotted against the labor cost displaced per autonomous completion, is your real productivity gauge.

How Many Robots Can One Operator Watch?

This is the single most leveraged number in the whole model and the one most likely to be guessed at. The answer is not a constant — it is a function of exception rate, exception duration, and how clean the handoff UI is.

The literature gives a band, not a point. Human-factors research on supervised autonomy (Eriksson & Birrell at Coventry studying automated vehicles [6]; Cai et al. on multi-robot teleoperation scheduling, RSS 2022 [8]; Chen et al. on human–agent teaming, US Army Research Lab [10]) converges on a few stable findings. With continuous teleoperation, ratio is 1:1 by definition. With assistive autonomy where the human approves moves and handles exceptions, sustainable ratios sit around 1:4 to 1:8 for cognitively demanding tasks. With mature supervisory autonomy — the operator only sees alerts that the robot cannot self-resolve — warehouse AMR fleet managers report sustained ratios of 1:20 to 1:50+ [9].

Where interior cleaning lands. Honest answer for the current generation: 1:4 to 1:8 in the first year of a deployment, climbing to 1:10 to 1:20 as exception handling matures and the long-tail object library grows. A vendor promising 1:50 on a service-robot platform on day one is either selling a different product or has not run the math on exception rate × duration.

Why this dominates the unit economics. At 1:1, the human cost per robot-hour equals the human labor you were trying to replace — you have automated nothing. At 1:8 the supervision cost is one-eighth of a wage, often a rounding error against the throughput gain. The transition between those two regimes is where your business case lives or dies.

The Bay Only Pays for Itself When It Is Running

Utilization is the variable that bends the payback curve fastest, and it is the one franchise owners control most directly. A robot that runs 6 productive hours a day pays back at one rate; the same robot running 14 hours pays back roughly 2.3× faster, because the fixed monthly costs (service, supervision share, software) are amortized across more value-producing hours.

The 24/7 lever is unique to mobility services. Most labor-replacing automation competes with a wage that is the same at 3 a.m. as at 3 p.m. Interior cleaning competes with a wage that does not exist at 3 a.m. — overnight detailing is operationally impractical at most car-wash sites and economically impossible at most rental hubs. A robot that can run an unsupervised graveyard shift on the cars returned at end-of-day is unlocking hours that have a zero-labor alternative. That is a structurally better dollar than the daytime hour you are replacing.

The honest caveat. 24/7 only counts if the robot can actually run untended. That requires a real charging strategy, a contained workspace where a stuck robot does not block the bay until 7 a.m., and a remote-supervisor tier (Section 04) that can intervene without driving to the site. Many operators model 24/7 utilization on the spec sheet and discover, post-deployment, that the practical number is closer to 14–16 productive hours. Build the model both ways.

Three Curves, Three Different Businesses

Plotted as cumulative cash flow from day zero, the same robot tells three stories under three operating regimes, and these are the slides that should drive the procurement decision rather than a single payback number.

1-shift / underutilized. Common in pilots and small franchises. Payback can stretch to 4+ years; the robot is profitable but barely. The risk is that any deterioration — exception rate creep, a quarter of weak demand — pushes the curve past the platform's expected service life.

2-shift operation. The bread-and-butter case for an established car wash or detailing chain. Payback typically lands in the 18–28 month band, depending on labor rate and supervision ratio. This is the regime where the math works without heroics.

24/7 / fleet overnight. Achievable for rental hubs, robotaxi depots, and large car-wash sites with overnight inventory. Payback compresses below 18 months, and the post-payback margin contribution is structurally higher because you are converting fixed assets into revenue during hours human labor cannot economically reach.

Look at the curves below before you look at any single payback number. The slope after break-even matters as much as where the curve crosses zero — that slope is your future cash flow once the platform is paid off.

When to Lease and When to Own

Robotics-as-a-Service (RaaS) has become the default procurement model for mobile manipulators in service industries, and for good reason: it removes the upfront capital hurdle, bundles maintenance, and shifts platform-obsolescence risk to the vendor [2][7]. It is not, however, always the cheaper option in the long run.

The crossover math. Indicative bands from Locus, 6 River Systems, Geek+, and OPSdesign place RaaS pricing for mid-complexity mobile manipulators in the $3,500–$6,000/month range, all-in (hardware lease + service + software + a base supervision-tooling tier) [2][7]. An equivalent outright purchase clusters around $80k–$150k upfront plus a service contract at ~10–15% of CapEx/year. With those bands, the cumulative-cost lines cross somewhere in months 30–42.

The decision rule that matters. If your utilization plan is uncertain, your operations team is small, or you expect to swap platforms within 3 years (likely for first-generation deployments in any new category), RaaS is the right answer almost regardless of the long-run math — the option value of being able to walk away exceeds the cost premium. If utilization is proven, the platform is mature, and you have an in-house service capability, ownership pulls ahead in year four and the gap widens from there.

A defensible framing for the board: RaaS is venture capital you pay the vendor for absorbing your deployment risk. Once that risk is gone, you should refinance.

Which Assumption Actually Moves Payback

An ROI model is only as honest as its sensitivities. Run the spreadsheet at ±25% on every input and rank the inputs by how many months they move payback. The ranking is consistent across the deployments we have seen and across the published industrial-robotics surveys [4][7]: utilization first, labor rate second, supervision ratio third, exception rate fourth, then everything else.

Utilization is dominant because it acts as a multiplier on every productive hour, while leaving fixed OpEx unchanged. Labor rate is the headline driver of revenue per autonomous completion and has the additional virtue of trending up in nearly every metro [1]. Supervision ratio acts on the largest controllable OpEx line and benefits from compounding: as exception handling improves, both the ratio and the throughput improve together. Exception rate is slower-moving but matters because it drives both throughput loss and supervision load.

If your spreadsheet's payback is dominated by a single input, that is the input to negotiate hardest on — service contract terms, an SLA on supervision tooling, a utilization-tied lease structure, or a contractual commitment from the vendor on the autonomous completion rate at a stated cycle time.

What Operators Consistently Underestimate

Five line items show up as unwelcome surprises in nearly every first-year post-mortem. Build them into the model from day one.

Integration with site systems. Tying the robot's 'available' state to the wash POS or the rental return system is unglamorous middleware work that is rarely scoped in the original quote. Budget 40–120 hours of integration engineering per site for anything beyond a standalone deployment.

Site preparation. Wet floors are not standard mobile-robot terrain. IP-rated platforms, drainage routing, and slip-resistant flooring upgrades are real costs. An anchor point for safe charging, network coverage in the bay, and a ventilated cleaning-fluid storage area are easy to forget at procurement time and expensive to retrofit.

Operator training and turnover. Your supervisor is now operating a robot fleet console, not a pressure washer. Training time is real (2–4 weeks to competency in our deployments), and detailer turnover in the industry is high — budget annual retraining as a recurring line.

Software lifecycle. Five-year TCO models routinely assume the platform's software is static. It will not be. Major version upgrades may require recommissioning, schema migrations of your fleet data, or limited downtime windows. A small annual reserve (2–3% of CapEx) covers most of this.

Insurance and compliance. A mobile robot operating in a public-adjacent environment (rental returns, car-wash bays with customers nearby) sits in a different liability bucket than fixed industrial automation. Confirm coverage before the first install, not after the first incident.

Two Operators, Two Models

Small franchise (single bay, 2-shift). $110k CapEx, $1.4k/mo service, supervision ratio 1:4 (one detailer doubles as supervisor and exception handler), 12h/day productive utilization, $22/hr fully-loaded labor displaced, 9-minute net cycle time, 12% exception rate. Result: ≈ $4.2k/mo net contribution, payback ≈ 24 months, 5-year NPV positive at 10% discount.

Depot operator (5 bays, 24/7). Same hardware, but RaaS at $4.5k/mo per robot, supervision ratio 1:8 (one remote supervisor per shift, three shifts), 18h/day productive utilization, exception rate down to 7% (better dispatch, cleaner vehicle mix). Result: ≈ $7k/mo net per robot, payback ≈ 17 months on each unit, and the post-payback margin per robot-hour is roughly 2.4× the franchise case because supervision cost is amortized across the fleet.

Same robot, completely different business. The ROI is not a property of the platform — it is a property of how you run it.

How to Pressure-Test Your Own Model

Before you commit, run this checklist against any vendor-supplied ROI deck.

1. Demand the autonomous completion rate. Not cycle time. Not 'cars per hour'. The percentage of cars finished without human intervention at a stated time budget, on a vehicle mix representative of your fleet.

2. Model supervision explicitly. Insist on a defended ratio with the underlying exception rate × duration math. If the vendor cannot produce that breakdown, assume 1:4 for year one and revise upward only after a 90-day pilot.

3. Run the sensitivity tornado. ±25% on every input. If a single input owns the result, that input is your negotiation lever.

4. Build both the buy and the lease curves. Even if you intend to lease, knowing the crossover month tells you when to refinance.

5. Subtract the integration line. If the quote does not have a site-prep and integration line item, the project is going to learn what those numbers are the hard way.

6. Pressure-test against an independent benchmark. The Robotomated TCO calculator [3], McKinsey's industrial robotics ROI bands [4], and the IFR World Robotics service-robot section [5] are all reasonable triangulation points. If your model is twice as optimistic as all three, the model is wrong.

ROI math for mobile manipulators is harder than for fixed automation, but it is not mysterious. Honest math, run on honest inputs, with utilization and supervision modeled explicitly, gives a defensible answer. The vendors who can produce that math without flinching are the ones to take seriously.