Service markets allow machine capacity to be matched to many small plots, and in China they have grown to the point where combine teams follow the harvest across provinces (Yang et al., 2013). The same literature records failures of supply-led models. In Ghana, state-supported mechanisation centres equipped with large tractors struggled to achieve the utilisation needed to cover their costs, while private owner-operators serving neighbours expanded (Diao et al., 2014). Hiring also has an inherent weakness for time-critical work. Every farmer in a locality wants the same operation in the same fortnight, and a farmer who hires bears the cost of waiting, which is one reason farmers who can afford it prefer to own (Allen and Lueck, 2002). Rental frictions of this kind help explain why machine capacity and farm size remain poorly matched (Foster and Rosenzweig, 2022).

The dominant rotation of the plains places a wet-season rice crop before a winter wheat crop, with a turnaround that is short everywhere and shortest where long-duration rice is grown (Timsina and Connor, 2001). Wheat sown after its optimum date loses yield as grain filling is pushed into rising spring temperatures, a relationship established experimentally for north-west India (Ortiz-Monasterio et al., 1994). Zero tillage, in which wheat is drilled directly into rice stubble, shortens the turnaround and lowers cultivation costs, and its spread in the region has depended on service providers who own the drill (Erenstein and Laxmi, 2008; Keil et al., 2016). Timeliness is therefore the main product that a hiring centre sells in this system, and capacity planned on annual averages will fail at the moment it is needed.

Jointly owned machinery is a shared resource with familiar hazards. Members have an incentive to use a machine hard and maintain it little, to demand priority at the peak, and to resist charges high enough to replace the fleet. Cook (1995) classified these as free-rider, horizon and control problems in cooperatives generally. Ostrom (1990) found that groups managing shared resources successfully had clear membership boundaries, rules matched to local conditions, monitoring by the users themselves and graduated sanctions. A machinery centre needs each of these in operational form.

The centre should begin from a register of the members it will serve, with the area each cultivates, the crops grown and the location and access of plots. Travel between scattered plots consumes machine time that earns nothing, so a compact area served well is preferable to a wide one served late. The register also settles in advance who is entitled to member rates and priority, which is the boundary condition on which the later rules depend.

For each crop the operations to be offered are listed, such as primary tillage, puddling, sowing, spraying, harvesting, threshing and residue handling, together with the calendar window in which each must be completed. Multiplying the members' crop areas by the operations gives the hectares of work falling in each window. The number of days within a window on which fieldwork is actually possible is smaller than the calendar length, because of rain, wet soil and breakdowns, and should be estimated conservatively from local experience (Witney, 1988).

Machinery management texts give the effective field capacity of an implement, in hectares per hour, as forward speed in kilometres per hour multiplied by working width in metres and by field efficiency, divided by ten (Hunt, 2001; ASABE, 2006). Typical ranges of speed and field efficiency for each class of machine are tabulated in the engineering standards (ASABE, 2011). The number of units required for an operation is the hectares to be covered in its window divided by the product of field capacity, working hours per day and workable days. The calculation is made for the tightest window, which in this system is usually wheat sowing, and tractors are then matched to the implements they must pull. Machines that serve several operations or seasons deserve preference, because annual use is what spreads fixed cost.

Ownership costs comprise depreciation, interest on the capital invested, insurance and housing. They accrue whether or not the machine works. Operating costs comprise fuel, lubricants, repairs and maintenance, and the operator's wage. Standard procedures estimate each from purchase price, expected life and annual hours (ASABE, 2006; Hunt, 2001). Costing should be done twice, at the subsidised price the cooperative pays and at the full price that replacement will eventually require. Dividing annual ownership cost by the margin between the hire charge and the operating cost per hour gives the break-even hours of use, the single most informative number for judging whether a machine belongs in the fleet.

Charges should recover full cost at the utilisation expected from step 2, and should be compared with local private rates. A charge far below the private rate invites excess demand and resale of the service, and one above it loses members. Rules for the peak matter as much as the tariff. Options include serving bookings in order of receipt with a deposit, rotating priority among hamlets from year to year, or capping the area any one member can book in the first days of a window. Service to non-members at a higher rate can raise utilisation outside the peak without displacing members during it.

Each machine needs a named operator, a logbook recording hours, area, fuel and the member served, and a schedule of routine maintenance. Operators paid partly by the area completed and held responsible for avoidable damage have reason to work quickly and carefully. A fixed share of every hire charge should go to a maintenance and replacement fund that the committee cannot divert to other uses. Logbooks and accounts should be open to members and examined annually by someone outside the committee, which gives practical effect to the monitoring that Ostrom (1990) identified as a condition of durable collective management.

At the end of each season the hours worked by each machine are compared with its break-even hours, and requests that could not be met on time are counted alongside them. A machine that falls well short of break-even is a candidate for sale or for wider hiring out, and unmet requests in a window indicate where capacity should be added. The plan for the following year is revised accordingly.