The time-consuming nature of current methods for detecting antimicrobial resistance (AMR) to guide mastitis treatment and for surveillance, drives innovation towards faster, easier, and more portable technology. Rapid on-farm testing could guide antibiotic selection, reducing misuse that contributes to resistance. We identify challenges that arise when developing miniaturized antibiotic susceptibility tests (AST) for rapid on-farm use directly in milk. We experimentally studied three factors: sample matrix (specifically milk or spoiled milk); the commensal bacteria found in fresh bovine milk; and result time on the performance of miniaturised AST. Microfluidic "dip-and-test" devices made from microcapillary film (MCF) were able to monitor Gram-negative bacterial growth colourimetrically even in the presence of milk and yoghurt (used to simulate spoiled milk samples), as long as this sample matrix was diluted 1:5 or more in growth medium. Growth detection kinetics using resazurin was not changed by milk at final concentrations of 20% or lower, but a significant delay was seen with yoghurt above 10%. The minimum inhibitory concentration (MIC) for ciprofloxacin and gentamicin was increased in the presence of higher concentrations of milk and yoghurt. When diluted to 1% all observed MIC were within range, indicating dilution may be sufficient to avoid milk matrix interfering with microfluidic AST. We found a median commensal cell count of 6 × 105 CFU/mL across 40 healthy milk samples and tested if these bacteria could alter microfluidic AST. We found that false susceptibility may be observed at early endpoint times if testing some pathogen and commensal mixtures. However, such errors are only expected to occur when a susceptible commensal organism is present at higher cell density relative to the resistant pathogen, and this can be avoided by reading at later endpoints, leading to a trade-off between accuracy and time-to-result. We conclude that with further optimisation, and additional studies of Gram-positive organisms, it should be possible to obtain rapid results for microfluidic AST, but a trade-off is needed between time-to-result, sample dilution, and accuracy.