Miele's diagnosis model helped technicians fix it on the first visit

For an appliance maker, service cost and customer satisfaction both hinge on the same thing: diagnosing the fault correctly before the technician leaves the depot. A wrong diagnosis means a wasted visit and a second one.

Miele, with IBM Research, built a system that learns from historical technician-visit data — structured records plus unstructured textual notes — and combines them in a probabilistic model. A semantic model of the domain shapes both the analysis pipelines and the probability model.

The result was a significant improvement in service efficiency together with an increase in an already high first-fix rate: technicians carried the right parts and completed repairs in less time.

The work was published as a Wagner Prize special-issue paper in the INFORMS Journal on Applied Analytics (2019).