I am forecasting revenue/sales which have an upward trend. The dataset includes monthly values of revenue(>60 months), hr cost(60>months), and marketing cost(24 months). HR with different departments and marketing cost with different channels. The main purpose is to find the effect of each marketing channel and each department to revenue (1 unit spent on cost, how much gonna earn in revenue and how long it takes) For some reason, I have a problem with making the data stationary at 1 level; some are at 2, and some are even higher level (ADF, KPSS test). 
For the causality effect, I did use ARDL and OLS to do regression on all exogenous variables, and their time lags up to t=5, but the result did not show good significance and also some negative coef. I used regression on each single variable and its lag to find the significant coef, which I hope 'affects' to revenue, but it faces omitted variables. I am wondering if including fixed effect into single pannel ols will make sense to see the effect of each independent variable without fear of the effect of other independent variables?
About the forecasting model, I am thinking about VAR model, but the stationary part always annoys me. Is it because of the marketing cost has less data than the target variables or are there any things I could do to make to whole data set stationary at the same level?
