Milk drawn from a healthy udder contains few bacteria. Contamination comes from teat skin, hands, utensils and containers, and what happens next depends on temperature. At ambient tropical temperatures mesophilic bacteria multiply quickly once the short-lived natural inhibitory systems of fresh milk are exhausted, and lactic acid producers sour the milk within hours. Prompt cooling to refrigeration temperature slows growth sharply, although it does not stop psychrotrophic organisms, whose heat-stable enzymes become a problem when cold milk is stored for days (Cousin, 1982; Walstra et al., 2006). For a collection chain measured in hours, the practical variables are the initial load of bacteria and the length of time the milk stays warm.

The Codex code of hygienic practice for milk requires that time and temperature between milking and processing be controlled so as to minimise microbial growth, and leaves the specific limits to national authorities (Codex Alimentarius Commission, 2004). The joint guide of the Food and Agriculture Organization and the International Dairy Federation gives the corresponding advice for the farm: clean equipment, and milk cooled or delivered within a specified time (FAO and IDF, 2011). The harmonised East African standard for raw cow milk sets the grades against which processors accept or reject deliveries (EAC, 2006). Where refrigeration is not feasible, Codex guidelines permit activation of the lactoperoxidase system to extend keeping time during collection, as a stopgap and not a substitute for hygiene or for cooling (Codex Alimentarius Commission, 1991).

Research in Kenya showed that the quality of marketed milk declines along the chain and that handling practice, container type and time to market explain much of the decline (Omore et al., 2005). Training material written for small-scale traders in Eastern Africa accordingly concentrates on containers that can be cleaned, on speed, and on the simple platform tests by which milk is accepted or refused: appearance and smell, the alcohol test, clot-on-boiling, the lactometer and the resazurin test (Kurwijila, 2006). Risk assessment of unpasteurised milk in the region has added a public health dimension to what might otherwise be seen as a commercial problem (Grace et al., 2008). The policy context matters too, since milk that fails formal channels often finds its way into informal ones (Kaitibie et al., 2010).

The first task is descriptive. The assessor walks or rides each collection route and records where milk is produced, who carries it, in what, how far and at what hours, where it changes hands and where it waits. Evening milk needs particular attention, because a cooperative that receives only in the morning leaves the evening milking on the farm overnight. The result is a map on which every later measurement can be located.

Small battery-powered temperature loggers are placed in a sample of cans at the farm and travel with the milk to the cooling tank; further loggers record tank temperature and, where possible, the tanker's journey to the processor. The sample should cover the longest and shortest routes, morning and evening milk, and the wet and dry seasons, since road conditions and milk volumes change through the year. The traces show how long milk stays warm before it reaches the tank and how fast the tank brings each addition down, which is the direct measure of cold chain performance that rejection figures only imply.

Temperature histories need to be tied to quality. At fixed points, namely the farm, arrival at the collection point, the bulk tank before dispatch and delivery to the processor, milk is examined with the platform tests the cooperative already uses and, for a subsample, by total bacterial count in a competent laboratory. Results are read against the grades of the regional raw milk standard (EAC, 2006). The same milk should be followed through the points where feasible, so that a deterioration can be attributed to a segment of the chain.

A cooling tank is designed for a stated volume added at a stated rate. The audit compares that specification with actual use: how full the tank runs, how warm milk is blended with cold, whether the agitator and thermostat work, how the tank is cleaned. Power is recorded separately, by logging interruptions over several weeks and noting whether a generator exists, starts and has fuel. Maintenance history and the availability of spare parts and a technician complete the picture.

Cooperatives keep intake and payment records and receive quality reports from their buyers. A review covering at least one full year shows when rejections cluster, by season, by route or by day of the week. Observation at milking, on the road and on the reception platform, together with structured interviews with farmers, transporters, graders and the manager, explains the patterns: why evening milk is or is not delivered, what transporters are paid for, how graders handle borderline milk. Interviews follow ordinary ethical practice, with consent and without recording names against answers.

The evidence is then organised with two standard quality tools. A cause-and-effect diagram sorts contributing factors under headings such as people, equipment, methods, materials and environment (Ishikawa, 1976). A failure mode and effects analysis lists each way the chain can fail and scores it for severity, frequency of occurrence and likelihood of going undetected, so that the failures can be ranked (Stamatis, 2003; IEC, 2018). The scores are judgements, and the protocol requires that the evidence behind each be written beside it. The ranked list is presented to management and members, who know constraints that an outside assessor does not, and the meeting agrees which failures to address first and who is responsible.