Which from an outsider perspective could have been companies recouping on bets made (discounts offered) in 2016 to grow their brand recognition, etc. It's hard to draw too strong a conclusion from data for for so few years, where costs change so much, and the outlier is the most recent year (which is also incomplete). It's also possible that the actual business costs from year-over-year operation are finally becoming well known, and the installation costs are finally reflecting items that weren't correctly priced in earlier (botched installation fixed, warranty repairs, etc).