I would be willing to help even if I am not selected as your preferred bidder. Basically I will chip in for free.
It would be hard to predict the complexity of the job without seeing the data but I will say a few things.
First, it is certainly possible. We can use the experiential data to predict what will be encountered. Then use the same data to predict what to send.
So like if under these conditions, units a,b,c,d, are used most frequently, send the units that have the highest success rates vs them under the same conditions.
Second, unless you've been planning things for a while, it's unlikely you've gathered the right data. A bigger pool is better, for sure, so it's not totally wasted, but some data has to be gathered meticulously.
Finally, reverse engineering game mechanics and user behavior is complicated. This will likely be a living project. It will likely require you to be able to make changes to it. If I was to accept this job it wouldn't be me working on it, it would be us working on it. I would need to teach you to do it yourself in the process.