Distribution and modeling of TV ad views

An advertiser can show his ad in several different time windows on TV. For each time window the number of viewers who have their TV turned on (=viewers) varies. In addition it is assumed that the probability a viewer actually recognizes a broadcasted ad (=recognition rate) is not only different for each time window (e.g. during prime time more viewers might stay in front of their running TV than during lunch time), but it is also dependent on the slot number within a time window. Slot number 1 (means the ad is shown first) has a higher recognition rate than slot number 2 and so on (e.g. because the later an ad is shown the more viewers have left their living room).

Using past data provided by the TV channels I want to create two models, one for predicting the number of viewers for each time window, and another one for predicting the recognition rate in each time window depending on the slot number.

My findings so far (at least what I believe):

a) The number of viewers in each time window on a given day is Poisson distributed and varies over time (trend, season, weekday). For modeling it I will use time series analysis methods, but I am not sure if the insight that it's Poisson distributed affects the forecasting model.

b) The recognition rate is binomially distributed. Apparently the following log-lin model explains the dependence on the slot number (=s) quite well: ln(recognition rate)=a+(b*s). I want to use least-squares to estimate the parameters, but since the underlying process (viewers) is Poisson distributed, does that mean I have to use maximum-likelihood instead?

An help regarding my questions and the general approach would be greatly appreciated. Thank you very much.