Industry Guide 📅 September 25, 2026 ⏱ 11 min read

What Telematics Leaves Untagged: RFID for Attachments, Scaffolding and Small Plant

A rental depot knows exactly where its excavators are. It has known for years, because a powered machine carries a battery, and a battery carries a SIM. The question a depot answers at 6 a.m. on a Monday is how many 600 mm digging buckets it owns, which of them went out on contract 44817, and whether the breaker that came back on Friday is the breaker that left on Tuesday.

That population — attachments, scaffolding, props, formwork panels, trench boxes, leads and the small-tool kit — is unpowered, so its identity comes from the reader rather than from a SIM. It is also the population a depot moves most often. This is how we tag it, what the tags cost against the value at risk, and what the arithmetic at the gate has to look like before anyone signs off a portal.

Rental over ownership is the driver; theft is the second act

The reason this population is growing is the shift from owning equipment to hiring it, and the association data puts numbers on it.

The American Rental Association’s updated forecast, compiled by S&P Global — the article names “Scott Hazelton, managing director at S&P Global, the international forecasting firm that compiles data and analysis for the ARA forecast” — was reported on 21 May 2026. It puts combined US construction and industrial equipment plus general tool rental revenue at “$83.5 billion” in 2026, up “3.6 percent”, followed by “3.8 percent” in 2027 and “4.4 percent” in 2028. Canada’s combined figure is “$6.3 billion” in 2026, up “5 percent”; US event rental is “$6.1 billion”, up “8 percent”.

ARA’s own attribution matters more than the totals, because it explains the physical consequence. Tom Doyle, ARA vice president, is quoted: “Rental tailwinds include project uncertainty, market volatility, sustainability, financial flexibility for the rental user and the high cost of owning.”

For a European anchor we pin a single edition deliberately. The ERA Market Report 2023, dated October 2023, covering 16 European countries, states that “The European rental market is expected to slow down from a strong growth of 7.5% in 2022 to 2.7% in 2023”, with a rebound “with rental growing by 4.6%” in 2025. Its mechanism is the same one: “high machinery purchase prices and uncertainty about the economic environment is shifting demand for usership over ownership”. On turnover, that edition puts the UK, its largest single market, at €6,570 million for 2021 — the most recent year it classes as actual data rather than estimate.

One sourcing note, because the editions are revised and ERA says so. The 2023 edition puts the 2025 rebound at “4.6%”; ERA’s own announcement of the 2024 edition projects that “the European rental market is projected to grow by 0.9% in 2024, 2.8% in 2025, and 3.6% in 2026”. The report carries the instruction itself: figures “cannot be directly compared with the figures of previous reports”. So quote one edition, name it, and state the year of every figure.

What the shift means physically is simple: more assets, moving between more sites, more often, with shorter intervals between handovers. That is the condition RFID exists for.

Split the fleet in two

Every useful conversation about rental tracking starts by drawing one line through the fleet.

The telematics-covered half. Powered machines report over a fleet data exchange schema. The Association of Equipment Manufacturers publishes the schema on its standards page, where it is called the ISO Fleet Data Exchange and describes it as specifying communication for “mobile machinery status data from a telematics provider’s server to customer applications via the Internet”, with server responses carrying “specified data elements” for analysing machine performance and management status. Take the designation from that page when you write it into a specification, and list the data elements your own provider actually returns rather than a generic set.

The telematics-uncovered half, itemised, because precision here is what carries a project: digging and grading buckets, hydraulic breakers, quick-hitches, compaction plates, augers, scaffold tube and fittings, props, formwork panels, trench boxes, lighting-tower and generator leads, and the small-tool kit.

The defining property of that list is that its identity is passive for the life of the steel: energised by the reader at the moment it is interrogated, inert the rest of the time, nothing to fit and nothing to charge, with a service life measured by the asset rather than by a battery.

This half is also the half that moves most, comes back on the wrong contract, and sits off-hired-but-unreturned in a corner of a customer’s compound for three weeks. It is high-count, low-unit-value, and almost always reconciled on paper. The addressable scope is precise: on-metal hard tags, read at the depot gate and at check-out, against the identity the hire system already holds.

A worked count for a mid-size depot

Numbers first, because the ratio is the entire argument. The depot below is an illustrative model, not a survey — every figure is an assumption you should replace with your own asset register before costing anything.

Population	Items	Powered	Identity method	Tag, indicative USD each	Tag spend
Powered plant (excavators, dumpers, rollers, telehandlers, towers, gensets)	220	Yes	Onboard telematics	—	—
Attachments: buckets, breakers, quick-hitches, plates, augers	360	No	On-metal hard tag	2.50	$900
Scaffold tube and fittings, props, formwork panels, trench boxes	4,200	No	Small on-metal tag	0.45	$1,890
Leads, hoses, ancillaries	180	No	On-metal or cable-flag tag	0.40	$72
Small tools and kit boxes	900	No	Tag the box; inlay on tool bodies	0.35	$315
Non-powered total	5,640	—	—	—	$3,177

Two ratios fall out. Non-powered to powered is 5,640 ÷ 220 = 25.6 : 1. And telematics, which absorbs most of the fleet-visibility budget, covers 220 of 5,860 items, or 3.75% of the things the depot has to hand over and get back.

Put the tag spend against the value at risk rather than against a saving you cannot yet prove. The insurers’ report quoted in the next section gives “The average estimated value of a stolen piece of equipment is $17,400.” Tagging the entire non-powered population of this depot costs $3,177, about 18% of one average stolen machine. That is the comparison to take to a finance director, because both numbers have a source.

Which event to instrument. Out-on-hire and back-on-return carry the value; the yard cycle count is a by-product you get free once the tags exist. If 70% or more of outbound lines leave through one or two controlled exits, a gate portal is the primary event — indicatively $3,800–$6,500 per lane for a four-port reader, four antennas, trigger, enclosure and commissioning. Where loads are assembled loose in the yard with no choke point, rugged handhelds at the point of loading do the same job, indicatively $700–$1,400 each. Most depots need both eventually; the ratio decides which is first.

What the insurers say the problem actually is

Here is the primary document, quoted exactly, with its date stated up front.

The NER / NICB 2013 Equipment Theft Report states: “The NICB compiled 11,486 reports of stolen machines in 2013” against “2,465 recoveries of equipment listed in the NCIC active theft file”, and headlines it “Only 21 percent of stolen equipment was recovered in 2013.”

On cost: “Annual estimates of the cost of equipment theft vary from about $300 million to $1 billion, with most estimates in the range of $400 million.” The report is explicit that these estimates cover the machines themselves, and that tool and material theft, damage caused during a theft, and business-interruption losses — rentals, project-delay penalties and wasted workforce time — sit outside the range. For a hire company, that surrounding cost is the expensive part.

Two more figures worth carrying: “$27,123 — Average value of machines recovered by police with NICB and NER assistance”, and the summary line “Most thefts are from work sites with little or no security.”

The part that matters most for a tagging decision is the report’s own list of factors behind the low recovery rate:

“Delays in discovery and reporting of theft”
“Inaccurate or nonexistent owner records”
“Lack of pre-purchase screening of used equipment”
“Limited law enforcement resources dedicated to equipment investigations”
“Complexities in equipment numbering systems”

And its own recommendation: “At a minimum, equipment owners should keep accurate lists of equipment with PIN/serial numbers and submit them to law enforcement, their insurers, and NER as soon they discover a theft.”

State the provenance plainly. The figures above are the 2013 edition, and every figure carries its year. We quote it as what it is: a 2013 baseline, with every figure carrying its year.

What a serialised tag register fixes on that list

Read the insurers’ factor list again and notice how much of it is a records problem rather than a security problem. That is the part a serialised tag register addresses directly.

Inaccurate or nonexistent owner records. The tag carries the identity the depot system already uses. A bucket stops being “600 mm GP bucket, one of nine” and becomes a unique number that reads the same at the gate, on the handheld and in the hire contract.
Complexities in equipment numbering systems. A serialised tag is one canonical number per physical item, readable without line of sight, without a clean paint surface, and without someone kneeling in mud to find a stamped plate.
Delays in discovery. Discovery moves to the next gate or yard event instead of the next audit. On a depot doing a weekly cycle count, that is days rather than months.
Pre-purchase screening. A verifiable, machine-read identity is something a buyer or an investigator can check against a register, which is precisely what the report asks owners to maintain.

Be precise about what each technology answers, and specify each for the job it does. A passive tag register answers which asset, on which contract, at which gate, at which minute — identity and custody, captured as an event. Continuous presence while an asset sits inside a live site is a different question with different hardware: an active 2.4 GHz zone read, which reports a tag’s presence in a zone within seconds of it entering, at up to 100 m typical and 400 m line-of-sight, confirmed by site survey.

In our active line the IDTS-F529 reader puts four antennas in one enclosure — omni and directional, 12 dBi dual-polarised, IP65 — with TCP/IP and 4G backhaul, carrying FCC and CE. For plant, the AT507 tag beacons at 1 s intervals with a 3-year battery, IP67, rated −20 to +45 °C, and mounts by screw eyelets or a 3M VHB pad. What the register adds to a compound is that every asset inside it has a name a machine can read.

Depots already running our RFID asset management system or tool tracking system are working from the same serialised register, which is why attachments and tool kits tend to be the same project rather than two.

Tagging steel, and what that costs in range

An attachment is a lump of hardened steel that gets loaded, dropped, painted over and pressure-washed. Two things decide whether a tag works on it: the mount, and the detuning.

Mounting, in order of survivability. A bolt through a drilled lug or a weld-on carrier plate survives a bucket; a rivet survives a compaction plate deck; VHB suits a clean flat formwork face indoors or under cover. The single most effective trick is geometric: put the tag in a recessed pocket or behind a shoulder so impacts and the wash lance pass over it rather than into it. On a bucket, that means the back of the ear or inside the lifting lug, not the wear face.

Now the physics, because your range is the one the link budget gives you. A UHF tag designed for free space and laid flat on steel detunes: the metal loads the antenna and the gain it actually realises drops. Received power falls with the square of distance, so for a fixed turn-on threshold the reachable range scales with the square root of the power available at the tag — in dB terms, every 6 dB of loss halves the read distance (10^−6/20 = 0.50).

Derive the number rather than quoting one. Worked forward-link budget at the lower-band legal ceiling, which India and Europe share:

Ceiling: 2 W e.r.p. = 33 dBm e.r.p. ETSI EN 302 208 V3.3.1 (2020-08) sets “33 dBm e.r.p. specified in a bandwidth of 200 kHz” on each lower-band high-power channel. India is the same ceiling under G.S.R. 853(E) of 10 December 2021.
Convert to EIRP: 33 + 2.15 = 35.15 dBm EIRP.
Budget back to the port: with a 6 dBd antenna and 1.5 dB of cable, the reader runs at 33 + 1.5 − 6 = 28.5 dBm conducted. Set the reader to that figure so the antenna port lands exactly on the legal ceiling.
Free-space path loss at 866 MHz (λ = 0.3462 m) over 4 m: 20 log₁₀(4π × 4 ÷ 0.3462) = 43.23 dB.
Turn-on threshold, sourced rather than assumed: the Impinj M780 / M781 tag chip datasheet v. 1.0 specifies “Read sensitivity of up to −23.5 dBm with a dipole antenna” (and “Write sensitivity of up to −20.5 dBm”, which is why an encoding station sits closer than a read portal). Use the chip figure as the floor and carry the finished tag’s realised gain on your asset as a separate, named input.
Power reaching the chip at 4 m, with realised tag gain G: 35.15 − 43.23 + G = −8.08 + G dBm.

Now run that one input across its plausible span, which is the honest way to present it, because realised gain is exactly the number your tag choice and your mounting buy:

Realised tag gain on the asset (design input)	Power at the chip at 4 m	Margin over −23.5 dBm	Free-space range at threshold	Range with our 6 dB engineering margin
0 dBi	−8.08 dBm	15.42 dB	23.59 m	11.82 m
−4 dBi	−12.08 dBm	11.42 dB	14.88 m	7.46 m
−8 dBi	−16.08 dBm	7.42 dB	9.39 m	4.71 m
−12 dBi	−20.08 dBm	3.42 dB	5.92 m	2.97 m

Each 4 dB step costs about 37% of the range (10^−4/20 = 0.63), and 12 dB of it takes the free-space figure from 23.59 m to 5.92 m at the same legal power. That is the whole argument for spending money on the tag and the standoff rather than on the reader: the reader is already at the ceiling, so the remaining decibels live in the mount. Establish your own figure by measurement — read your chosen tag on your own steel at a fixed distance, read the same tag in free air at the same distance, and record the difference in dB. That single measurement, per asset class, is what turns this table into your numbers.

Assumptions: free space, no multipath, forward-link limited, single antenna, chip matched to its antenna. Steel yards are not free space and the reverse link can bind first, so the commissioned figure comes from the site survey; the 6 dB column is our own design allowance for orientation, polarisation mismatch and multipath. The point of the budget is that it tells you which changes are worth making before you ever power a reader.

Asset class	Tag form	Mount	Primary read event
Buckets, breakers, quick-hitches	Rugged on-metal hard tag	M5/M6 bolt through drilled lug, or weld-on plate	Gate portal plus handheld at check-out
Compaction plates, augers	Smaller-footprint on-metal tag	Bolt or rivet into a recessed pocket	Gate portal
Scaffold tube, fittings, props	Low-cost small on-metal tag	Banded or riveted at the collar end	Bulk read at the stillage
Formwork panels, trench boxes	On-metal hard tag	Bolt, or VHB on a flat unpainted face	Gate portal
Kit boxes and small tools	On-metal tag on the box; inlay on tool bodies	VHB or rivet on box; adhesive inlay on tools	Handheld at issue and return

Scaffolding, props and formwork are a different problem from a breaker: thousands of near-identical items, low unit value, repetitive geometry that stacks into a dense metal lattice. Read them as a stillage rather than item by item at the gate — the container is the unit of hire, and per-item identity exists for audit and loss attribution rather than for the daily event. The same logic our returnable crate and cylinder tracking customers apply to circulating packaging applies exactly here.

Kit boxes: tag the box, the contents, or both? The arithmetic is short. Tagging 900 boxes at $0.35 costs $315. Tagging the contents as well, at roughly 12 items per box and $0.30 an inlay, adds about $3,240 — more than the entire rest of the programme. So tag every box now, and tag contents for the classes whose replacement cost or calibration record justifies it.

The depot gate

Assume a 4 m roller-shutter exit and steel assets going out on a low-loader. Four antennas, cross-polarised, two per side at roughly 1.2 m and 2.4 m above finished floor, angled inward so the beams cross ahead of the door line rather than at it. That gives a read zone roughly 3.0 m deep along the direction of travel — and zone depth is the number the whole design turns on.

Gate speed against zone depth. Dwell time is depth divided by speed. Assume the reader cycles four antennas at 200 ms each, so one full cycle is 800 ms:

Movement	Speed	Speed, m/s	Dwell over a 3.0 m zone	Full 4-antenna cycles per pass
Operator walking an item out	4 km/h	1.11	2.70 s	3.4
Low-loader crawling off the yard	5 km/h	1.39	2.16 s	2.7
Yard truck at pace	8 km/h	2.22	1.35 s	1.7
Forklift crossing the door	12 km/h	3.33	0.90 s	1.1

The answer to “by how much is a crawling low-loader easier than a forklift?” is 2.16 ÷ 0.90 = 2.4× the dwell. At 5 km/h every tag gets two to three presentations from every antenna. At 12 km/h it gets about one, which makes orientation decisive. Where traffic speed is fixed, deepen the zone instead — depth and speed are interchangeable in the arithmetic.

Metal-on-metal shadowing is the other thing to design for. A tag on the offside of a machine body sitting on a low-loader deck can be shadowed from both gate uprights. Two things fix it: mount tags on the same nominated face across the fleet as a tagging standard, and add a second read point — the handheld at the moment of loading, or a low-level antenna reading under the deck line. A second independent event is worth more than a fifth antenna at the same point.

Trigger the portal from a loop or a photo-eye. For the lower band this is a standards requirement, not a preference. EN 302 208 V3.3.1 clause 4.2.1 states that “For the lower band interrogators shall support trigger techniques that indicate the presence or arrival of objects that may be tagged”, and that such an interrogator “shall stop transmitting after it has ceased to read any further tags”. Clause 4.3.7.3 names the means: “triggers for motion, light beam or by applying a duty cycles for polling.” An inductive loop or a photo-eye is therefore the correct design, and it also happens to cut noise, cross-reads from the adjacent lane, and stray reads of the yard stock behind the door.

The same clause caps a continuous transmission: “the on-duration of A shall not exceed 4 s” with “the off-duration of B shall be not less than 100 ms”. Every pass in the table above completes inside a single 4 s transmission, so the legal timing sits comfortably around this application — and it is the reason your reader is configured for the band it ships into. India mirrors the ETSI lower band exactly: four channels centred at 865.7, 866.3, 866.9 and 867.5 MHz, channels of 200 kHz or less, at 2 W e.r.p. FCC-band variants for 902–928 MHz are a different configuration, where 47 CFR 15.247 allows 1 W conducted on the basis of “antennas with directional gains that do not exceed 6 dBi”, reduced dB-for-dB above that. We configure per destination market at order.

Automate two events first: out on hire, which writes a confirmed line-by-line dispatch against the contract, and back on return, which closes it. Everything else can follow once those two are trusted.

Check-out, check-in, and the reconciliation that pays for it

The financial case does not live in theft. It lives in the gap between what the contract says went out and what actually went out.

Lines against reads. At the moment the truck clears the gate, the system holds two lists: the hire contract lines, and the tags actually read. Where they agree, the dispatch is confirmed without a signature. Where they differ, an exception is raised while the vehicle is still on site — which is the moment the discrepancy is cheapest to fix.

Off-hire disputes. These are almost always arguments about dates, and they are settled by evidence rather than recollection. A timestamped gate read attributable to a contract and a vehicle is the evidence. The same read closes the rental clock, which is the difference between crediting a disputed week and invoicing it.

Damage attribution. Condition is a human judgement, but which item, on which contract, on which date is a data question. Serialised identity means a damaged quick-hitch is traceable to the hire it came back from, rather than absorbed into a pool of nine identical ones.

Utilisation from movements. Once out and back are both captured, utilisation is computed from real events. That number comes from your own movement data: the first three months of reads give you a baseline measured on your fleet rather than borrowed from a category average, and it keeps improving as the register fills.

Integration points are the part to scope early, because they are where these projects stall. The hire management system owns contracts and rates; depot stock owns location and condition; the RFID layer owns events. Three interfaces are enough: a contract-lines lookup, an event post-back, and an exception queue a named human works daily. Name that person on day one — that single assignment is what keeps the data clean. Depots moving assets between branches usually wire the same events into their logistics and supply chain tracking so an inter-depot transfer is one movement rather than a return plus a new hire.

Pilot scope and acceptance test

Scope a pilot you can judge cleanly. One depot, one gate, one asset class — attachments first, because they are the highest-value-per-loss population with the lowest tagging cost per item, and they are the class where the on-metal mount is hardest and therefore most worth proving early.

Measure before you tag. Four baselines, taken over four weeks, with no new hardware involved: count of lines reconciled on paper versus electronically; number of off-hire date disputes and their credited value; items recorded as missing at cycle count; and average time to complete a yard count. With those four numbers in hand, every later improvement is measurable.

Acceptance criteria should be numeric and agreed in writing before installation:

First-pass read rate on a loaded low-loader pass at a stated speed — state the speed, because as the table above shows it changes the physics. We commission against a nominated speed with the trigger live, on tags mounted to the agreed fleet standard.
Reconciliation accuracy against the hire contract, measured as lines correctly matched without human intervention.
Exception rate, with a stated ceiling for how many passes a human adjudicates per day. Set that ceiling low enough that a named person clears the queue inside a shift, because a queue that clears is what keeps a system in daily use.

Tag survivability test that actually matters. Beyond the IP rating on the datasheet, five real insults, on real assets, for a real period: impact against steel; pressure wash at the depot’s actual lance distance; painted over during refurbishment; heat cycling through a full summer in the yard; and six months in live hire circulation. Read every tagged item at the start and at the end, and report the survival rate per mount type. This is the test that tells you whether to bolt, rivet, weld or bond across the rest of the fleet, and it is the reason the pilot runs six months rather than six weeks.

What scales next, in order: the second gate, which turns a single choke point into a network; then the yard cycle count, which is nearly free once tags exist; then delivery confirmation at the customer’s site, which is where the off-hire argument actually starts.

We write the software and configure the readers per destination market, so a pilot can be scoped against your hire system rather than around it. If you want the tag and mount selection done against your own asset register, send the attachment list with material, mounting face and wash regime, and we will come back with tag classes, a link budget for your gate geometry, and a costed bill of materials in USD, EXW or delivered.

Frequently asked questions

How do you track rental equipment that has no power for telematics?

With passive UHF RFID. A passive tag is energised by the reader at the moment it is interrogated, so it carries an identity on a bucket, a scaffold tube or a formwork panel for the service life of the steel. Telematics answers where a powered machine is, continuously. A passive tag answers which item this is, on which contract, at which gate, at which minute — captured as an event at check-out, at the gate and at cycle count. On a typical depot model the non-powered population outnumbers the powered fleet by around 25 to 1, so this is where most of the unreconciled items sit.

Can an RFID tag survive on a steel bucket or a breaker?

Yes, with the right tag and the right mount. Use a purpose-built on-metal hard tag: bolt it through a drilled lug or onto a weld-on carrier plate, and recess it behind a shoulder so impacts and the wash lance pass over it rather than into it. On a bucket that means the back of the ear or inside the lifting lug rather than the wear face. Prove it rather than assume it: run impact, pressure wash, paint-over, summer heat cycling and six months of live circulation, then report survival rate by mount type before rolling the standard out across the fleet.

Does telematics already cover rental attachments?

Telematics covers the powered machine. The fleet data exchange schema behind it — called the ISO Fleet Data Exchange on the Association of Equipment Manufacturers’ standards page — moves “mobile machinery status data from a telematics provider’s server to customer applications via the Internet”, returning the data elements that provider specifies. Attachments, scaffolding, props, formwork, trench boxes, leads and tool kits are unpowered, so their identity comes from a tag and a reader instead. In the illustrative depot model in this article, telematics reaches 220 of 5,860 items, or 3.75% of what the depot hands over and takes back.

How do you tag scaffolding and formwork economically?

Change the read event rather than hunting for a cheaper tag. Scaffolding and props are high-count, low-value and stack into a dense metal lattice, so they read best as a stillage. Tag each item with a low-cost small on-metal tag, banded or riveted at the collar end, then read the stillage as the unit of hire — the container is what goes out on the contract. Per-item identity then exists for audit, loss attribution and reconciliation rather than for the daily movement. In the model above, 4,200 scaffolding and formwork items tag for about $1,890 at $0.45 each.

Will RFID help recover stolen plant?

A serialised tag register attacks the records half of the problem, which the insurers’ own analysis says is substantial. The NER/NICB 2013 Equipment Theft Report lists among the factors behind low recovery “inaccurate or nonexistent owner records”, “complexities in equipment numbering systems”, “delays in discovery and reporting of theft” and “lack of pre-purchase screening of used equipment”, and recommends owners “keep accurate lists of equipment with PIN/serial numbers”. A tag supplies exactly that: one canonical, machine-readable identity per item, discovery at the next gate or yard event rather than the next audit, and an identity a buyer or investigator can check against a register. The figures are the 2013 edition, and every figure carries its year.

What should a plant-hire depot's RFID gate actually read?

Two events, before anything else: out on hire, which confirms dispatch line by line against the contract, and back on return, which closes it. Design against dwell time. A 3.0 m read zone gives 2.16 s of dwell at 5 km/h and 0.90 s at 12 km/h — 2.4 times less read opportunity — so either slow the traffic or deepen the zone. Trigger the portal from a loop or photo-eye: for the lower band, EN 302 208 V3.3.1 requires interrogators to support trigger techniques indicating the arrival of objects that may be tagged, and to stop transmitting once no further tags are read.

What read range should I expect from a tag on steel?

Work it out rather than take it from a datasheet. At the 2 W e.r.p. lower-band ceiling India and Europe share, a reader delivers 35.15 dBm EIRP, free-space path loss at 866 MHz over 4 m is 43.23 dB, and the Impinj M780 / M781 chip datasheet gives “Read sensitivity of up to −23.5 dBm with a dipole antenna”. The one variable left is the gain your finished tag realises on your asset. At 0 dBi that budget reaches 23.59 m in free space; at −8 dBi, 9.39 m; at −12 dBi, 5.92 m — and applying our 6 dB engineering margin for orientation, polarisation and multipath gives 11.82 m, 4.71 m and 2.97 m respectively. Every 6 dB halves the distance, which is why the tag and the standoff earn more range than the reader setting does. Measure your own tag on your own steel against the same tag in free air, and the table becomes your numbers.

How accurate is an RFID check-out against a hire contract?

Accurate enough to be the record, provided you design for it and then verify it. Agree numeric acceptance criteria in writing before installation: first-pass read rate on a loaded pass at a nominated speed, reconciliation accuracy measured as contract lines matched without human intervention, and a ceiling on the exception rate a person adjudicates each day. Two design choices carry most of the accuracy: a single nominated tagging face across the fleet, so orientation is predictable, and a second independent read point — a handheld at loading or a low-level antenna — to resolve items shadowed behind a machine body on a low-loader deck. Baseline your current paper reconciliation for four weeks first, so the improvement is provable.