Industries / Technology / Software and SaaS / Field service software
Parts and inventory
Having the right spare part in the right van or depot, so the technician can finish the job on the first visit.
4
Milestones achieved, from 1966 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1966
A model for spare parts
Craig Sherbrooke of the RAND Corporation published METRIC, a mathematical model for deciding how many spares of each kind to hold at bases and at a central depot. RAND calls it "probably the first multi-echelon, multi-item model ever proposed for Air Force implementation"; multi-echelon means stock held at more than one level, such as a central warehouse and local sites. The parts it covered were of a type that made up 78% of the Air Force's investment in spares. It appeared in the journal Operations Research in 1968.
Credit
Craig C. Sherbrooke, RAND
1982
What goes in the kit
John Mamer of UCLA and Stephen Smith of Xerox's Palo Alto Research Center set out how to choose "the optimal kit of parts and tools for on-site equipment repairs", judged by how many jobs it lets a technician complete, and showed that the problem can be solved with a standard network method called max-flow, min-cut.
Credit
John W. Mamer and Stephen A. Smith
1990
IBM's parts network
IBM put in Optimizer, a system for "flexible and optimal control of service levels and spare parts inventory" across its US service network. Morris Cohen of Wharton, one of its developers, later said it let IBM cut its inventory investment by $250 million and its annual operating costs by 10%, across hundreds of thousands of parts in thousands of locations; these are the developers' figures.
Credit
Morris Cohen, Pasumarti Kamesam, Paul Kleindorfer, Hau Lee and Armen Tekerian
2024
Parts predicted from the fault
A team in Italy built a system that reads a fault description and the records of past repairs and ranks the spare parts the job is likely to need. It combines case-based reasoning, which finds similar past cases, with language processing. The team reports in IEEE Access that in real use on repairs of electrical appliances, including industrial ones, it reduced multiple visits "up to 50%"; the paper's summary gives no baseline.
Credit
Schiavo, Marcelloni, Ducange, Renda, Ruffini and Dani
Watch for
Parts named before the visit
Fisher & Paykel, an appliance maker, uses Salesforce's Agentforce to give its technicians a pre-work brief "with the most likely fault and recommended spare parts" before they reach the customer. No figures for first-visit fixes are given.
Achieved whenFisher & Paykel or an outside party publishes first-visit fix rates before and after, for a stated period and number of jobs.
Claimed by
Salesforce and Fisher & Paykel, September 2026
Open
01
Parts prediction, tested in public
An anonymized public dataset of field service jobs, from several firms and trades, that links each fault description and equipment model to the parts actually used. With it, a baseline model that ranks the likely parts, and a scoreboard for others to beat.
The briefCredit · unclaimed02
The kit, tried in vans
An open-source tool that reads a firm's job history and works out, by Mamer and Smith's method or a modern solver, which parts each van should carry to finish the most jobs on the first visit. Then a trial: half of a firm's vans restocked to the tool's list for three months, half left as they are.
The briefCredit · unclaimed
Open 01 · the brief
Parts prediction, tested in public
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An anonymized public dataset of field service jobs, from several firms and trades, that links each fault description and equipment model to the parts actually used. With it, a baseline model that ranks the likely parts, and a scoreboard for others to beat.
Why it is still open
The 2024 Italian system was built and tested on private repair records. This research found no public dataset that links fault reports to the parts later used.
Built with what exists
exports from field service software, used with the firms' permission; Python to strip personal data; retrieval of similar past jobs, as in the Italian system; small language models fine-tuned on one graphics chip (see the Note "One GPU to fine-tune").
First step
Take a year of jobs from one firm and count how often the parts used were already on the van when the job was sent.
Done looks like
At least 20,000 jobs from at least three firms released under an open license; the baseline's top-three hit rate, meaning how often the parts used are among its first three guesses; a second team reproduces it.
Open 02 · the brief
The kit, tried in vans
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-source tool that reads a firm's job history and works out, by Mamer and Smith's method or a modern solver, which parts each van should carry to finish the most jobs on the first visit. Then a trial: half of a firm's vans restocked to the tool's list for three months, half left as they are.
Why it is still open
The arithmetic for choosing a repair kit was published in 1982. A 2022 study of a forklift service business found that "parts on hand" were among the main factors in first-time completion. This research found no open tool for small trade firms and no published trial in vans.
Built with what exists
Google OR-Tools or PuLP for the optimization; job and parts exports from field service software; a spreadsheet of each van's current stock.
First step
Compute the kit from one firm's last year of jobs and compare it, line by line, with what is on its vans today.
Done looks like
10 vans on the tool's lists and 10 left as they are for three months, with first-visit completion rates, the value of parts carried and return trips published for both groups.