I've just been told about a data quality issue which may not have cost millions to resolve, but illustrates very well the effect poor data quality (and lack of information quality) can have at every level.
Call 1: A café reports to its head office that its bank card payment terminal has stopped working. Head office assumes (assumption 1) that it is a technical issue and instructs the café (call 2) to call in the technical support from the equipment provider (call 3).
The equipment provider sends a technician (visit 1), finds no technical faults and assumes (assumption 2) that the problem lies with the telecommunications provider. They leave behind a hefty bill for the visit. Café calls (call 4) the telecommunications company, who check their systems and find no fault, and therefore assume (assumption 3) that the equipment manufacturer is at fault. Call 5 to the equipment manufacturer, who point the finger back to the telecommunications provider; call 6 to the telecommunications provider who .... well, you're getting the picture.
For 6 months this situation remained unresolved. Calls were made, technicians sent, invoices sent. More importantly, the café was losing customers that didn't have the cash to pay and wanted to use their bank cards. Trying to resolve the situation was costing their staff time and money and their good humour.
A new round of calls and eventually somebody at head office actually checked - and found that the bill had not been paid. The equipment was working again within 24 hours, but the costs of this situation were crippling for the café concerned.
Not only do we see how dangerous it is to make assumptions, but here also there's a clear information quality problem in the support department of the equipment manufacturers. When their technical service was called they could check on technical aspects of the equipment installed, but there was no link with the financial system, so nobody could tell the café that the switch had been thrown because the bill went unpaid.
Sometimes it doesn't matter how good your data is - you have to actually go and look at it and make sure you use it properly.