Next you get to do a lot of datamining and analysis to figure out which reports are outliers. You'll need a control of people that do not use wattvision.
Once you've done that you can compare the two and figure out how much of the drop is attributable to the device and how much to other influences (season and so on).
That should give you an idea based on the current price per KWh how long on average it takes to earn back the value of the device.
You'll have to add a safety margin. In order to qualify for a refund people would have to send you their actual before and after stats for a 12 or 24 month period, so scans or PDFs of their electrical bills.
I'm sure there are lots of details to be worked out but that's the approach that I would take and before making any public claims you'd have to make sure the whole process can't be gamed in any way.
No point in becoming another cue-cat ;)
Really, keep it up, super marketing! Have you considered a 'the device will pay for itself within a year or your money back' guarantee?