Based on the current set of bid units in portfolio and the breadth of data available for their cost and revenue models, the optimizer may not be able to find a set of bids that exactly hits the budget target and simultaneously maximizes the revenue. As a result, the optimizer may set bids to spend more on some days of the week and less on the remaining days, so that the average weekly spend and revenue are in line with the portfolio target.
This won't happen as much if the portfolio's bid units have adequate historical data to build accurate models across a wide range of bid points. To build a wider range of historical data for the models to use, look for the bid units that have the greatest shift between high and low bids. If the bid units don't already have a wide range of estimates for clicks and impressions, then set constraints on those bid units for a few days to accumulate data at various bid points.