Industries / Industrial equipment and services / Equipment rental
Reservations and delivery
This is the work of taking an order for a machine, finding one that is free, and getting it to the site and back again on a truck or trailer.
5
Milestones achieved, from 1982 to 2017.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1982
A rental store's own software
Point of Rental Software, now based in Fort Worth, Texas, says it began as a rental store in 1982 and wrote its own counter system because the software of the day did not understand rental. By its own count it now serves more than 5,000 rental businesses in over 80 countries.
Credit
Point of Rental Software
1998
One system, many branches
Wynne Systems was founded in Long Beach, California, in January 1991 to write RentalMan, software for running rental companies with many branches. United Rentals bought it on 24 September 1998, and in its report on 2000 said that seven of the ten largest equipment rental companies, itself included, used RentalMan.
Credit
Wynne Systems; United Rentals
2000
Branches share one fleet
United Rentals' annual report for 2000 described systems that let branch staff search for a machine across a whole region, find the closest one and arrange delivery to the site. It said about 10.7% of its 2000 rental revenue came from machines shared between branches, and that it tied branch pay to each branch's return on assets to encourage that sharing. Customers could also rent or buy online at any hour through its E-Rental Store.
Credit
United Rentals
2002
Four tiedowns per machine
On 27 September 2002 the US Federal Motor Carrier Safety Administration (FMCSA) published cargo securement rules based on a North American model standard written with Canadian regulators. For machines on wheels or tracks weighing 4,536 kg (10,000 lb) or more, such as loaders and bulldozers, the rule requires at least four tiedowns placed near the front and rear, buckets and other attachments lowered and secured, and articulated machines held so they cannot bend in transit. It took effect on 26 December 2002, with compliance required by 1 January 2004.
Credit
FMCSA, with the Canadian Council of Motor Transport Administrators and Transport Canada
2017
Peer-to-peer turns to software
Yard Club was founded in San Francisco in 2013 so that contractors could rent their idle machines to one another through an app and website. Caterpillar invested in 2015 and bought the company in May 2017. By then Yard Club had moved from taking a cut of each rental to selling software for dispatch, scheduling and tracking of owned and rented fleets; it said it handled $120 million of rentals across 2,500 contractors and rental companies in 2016. Rivals named at the time included EquipmentShare, Getable and Dozr.
LessonMatching idle machines between contractors did not remain the business; managing the fleet did.
Credit
Yard Club (chief executive Colin Evran); Caterpillar
Watch for
A chat agent picks machines
United Rentals launched the Equipment Agent, a chat tool on its website: a customer describes the job in plain words and gets suggested machines, specifications such as capacity and reach, and links to the product pages from which they can reserve. The company says customers using it see "a 70-percent improvement in finding the right equipment"; it did not say how that was measured.
Achieved whenan outside party tests its suggestions against experienced counter staff on a published set of job requests.
Claimed by
United Rentals (Tony Leopold, chief technology and strategy officer), March 2026
Phone agent for the counter
Point of Rental's Intelligent Phone Agent answers calls at any hour and, using the rental firm's live price, stock and contract data, answers price and availability questions, writes quotes, takes off-hire calls (a renter's notice that a machine is finished with) and books collections.
Achieved whena rental firm or outside party publishes measured results, such as the share of calls completed without staff and the error rate in quotes, checked against call recordings.
Claimed by
Point of Rental Software, March 2026
Open
01
Test set for booking agents
A public set of 300 rental requests, written and spoken, modelled on real counter calls ("trench 1.2 m deep for a water line, gate is 1.5 m wide, soft ground"), each with an answer key written by two experienced counter staff: machine class and size, attachments, how it travels, and the questions a good counter clerk would ask first. Then score chat and phone agents against it, using only their public question-and-answer functions and placing no bookings.
The briefCredit · unclaimed02
Delivery access checked first
A planning check that takes a delivery address, the drop point on the site and the planned vehicle (a rigid truck with a tail-lift, a beavertail trailer or a low loader) and flags likely problems before dispatch: narrow streets, tight turns and weight or height limits on the last stretch from open map data, and gate width, slopes and overhead lines from photos the customer sends. It only advises the planner; drivers still check on arrival.
The briefCredit · unclaimed
Open 01 · the brief
Test set for booking agents
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of 300 rental requests, written and spoken, modelled on real counter calls ("trench 1.2 m deep for a water line, gate is 1.5 m wide, soft ground"), each with an answer key written by two experienced counter staff: machine class and size, attachments, how it travels, and the questions a good counter clerk would ask first. Then score chat and phone agents against it, using only their public question-and-answer functions and placing no bookings.
Why it is still open
United Rentals and Point of Rental launched booking agents in March 2026 and published only their own figures. The nearest public benchmarks test something else: OutfitterBench (November 2026) has AI agents run a simulated outdoor-gear rental business for 200 days, testing stock and commitment decisions, and AECBench (2025) asks language models 4,800 questions of construction knowledge, not which machine fits a job. A search of arXiv, Hugging Face and GitHub found no set of construction rental requests with expert answers.
Built with what exists
Manufacturers' public specification sheets for the answer key; ElevenLabs Scribe v3 or another speech-to-text model to turn recorded requests into text (see the Note "Fewer word errors"); any large language model API as a baseline; a spreadsheet for scoring.
First step
Write 50 requests from common job types (trenching, slab pour, ceiling work at 6 m) and have two counter staff key them independently.
Done looks like
300 requests and answer keys published under an open licence, agreement between the two keyers reported (Cohen's kappa, a standard measure of agreement beyond chance), and a scored table for at least three agents: share of requests with the correct machine class and size, count of suggestions that could not do the stated job, and share where the agent asked the needed questions.
Open 02 · the brief
Delivery access checked first
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A planning check that takes a delivery address, the drop point on the site and the planned vehicle (a rigid truck with a tail-lift, a beavertail trailer or a low loader) and flags likely problems before dispatch: narrow streets, tight turns and weight or height limits on the last stretch from open map data, and gate width, slopes and overhead lines from photos the customer sends. It only advises the planner; drivers still check on arrival.
Why it is still open
The International Powered Access Federation's guidance on unloading lifts on public roads (2023) asks firms to plan where the delivery vehicle will park, to allow for narrow roads, overhead lines and level crossings, and to request smaller delivery vehicles when placing the order. It is a checklist, not a tool, and a search found no published test of predicting whether a rental delivery vehicle can reach a site's drop point.
Built with what exists
OpenStreetMap tags for road width, maximum height and maximum weight; an open routing engine with a truck profile, such as GraphHopper or Valhalla; a swept-path calculation for each vehicle, the space its wheels and body sweep through a turn; a vision-language model to read gate widths and overhead lines from customers' photos, checked by a person.
First step
With a depot's permission, pull 50 delivery tickets that carry drivers' notes on access and code each by vehicle type and outcome.
Done looks like
200 past deliveries from at least two depots, each predicted without looking at the outcome and then scored against the driver's record (delivered as planned, delivered with another vehicle, or not delivered), with the share of real problems caught and the number of false alarms per 100 deliveries.