Manufacturing Analytics is a vast subject matter and the discussions around the same can be really detailed and absorbing. My mind goes back to an engagement revolving around warranty cost, for a global energy management specialist. This particular engagement concerned reduction of warranty claims costs by 6-10% for a yearly $400M warranty spend.
Talking about the engagement, It is one of the most complex analytics scenarios as reducing warranty cost requires numerous data points and advanced correlation in data. For example, consider the aforesaid engagement, the company has 30000+ repair centers across 80+ countries, 20+ factories across globe, 100,000+ SKU, 50-100 Million units over 3-5 years of warranty support and a very complex organization structure. To reduce warranty cost you should first integrate 30000+ rows of repair data every day and reverse engineer faulty products back to their manufacturing process history, manufacturing analytics test result history with BOM (Bill of Material) and identify the major root cause for failure.
It is simply very large and complex data coming from repair stations, manual inspections, and corporate master data, manufacturing machine logs, manufacturing test station, ERP, CRM and few other sources. The data volume could be in tune of 10+ TB with 30-50GB data coming in every day.