Confusing confidence interval question from old textbook 
Lets say an investigator reports a 95 percent confidence interval of 1 to 23 dollars per month in reduced utility bills for a randomly selected group of 50 customers who underwent training in energy conservation.
a) If this interval is too wide, what, if anything, can be done with the existing data to obtain a narrower confidence interval?
b) What can be done to obtain a narrower 95 percent confidence interval if another similar investigation is being planned?

Note: this is not an an homework question. I'm solving random questions from a textbook and I find it confusing. It's more of an applied question.
Attempt:

*

*I think the answer to the a part is by increasing the sample size of customers.


*I also think the answer to the b part is also the same but I'm not sure.
 A: It's about the first part of the question, isn't it?
Well, you can't go back and increase the number of subjects. You are done collecting data. You now have a standardized sample statistic, from which you want to estimate where the population parameter lies.
The larger the sample size the more certain you can be that your estimate will contain the population parameter , and the narrower your CI will be. Contrarily, the higher your confidence level (or smaller your significance level), the broader the CI will become.
Now, you want to reduce the CI, and there'll be a price to pay in terms of confidence level.


A confidence interval for a population parameter, $\theta$, is a random interval, calculated from the sample, that contains $\theta$ with some specified probability. For example, a $95\%$ CI for $\mu$ is a random interval that contains $\mu$ with probability $0.95$; if we were to take many random samples and form a confidence interval from each one, about $95\%$ of these intervals would contain $\mu$.
                                            Mathematical Statistics and Data Analysis
                                                                         John A. Rice


A: The second part of the question is indeed about increasing the sample size.  
But the first can't be because you only have the sample you have.  Now, the question as you reported it is a bit unclear on the actual experiment.  Let's assume that 50 people got the training and some other number of 'control' customers did not, that the first 50 were a random sample from all customers, and that the comparison that lead to the confidence interval was based on a comparison of utility bills between these two groups.  
If that's the situation then you might reason as follows: there are many causes of higher or lower utility bills, of which the training program is only one. 
Consequently you can often (but not always) do better, in the sense of being more precise, by conditioning in your analysis on attributes of the customers that you think a) are not a consequence of the training program, but b) predict a higher (or lower) utility bill - things such as the age of the house, the number of teenagers living in it, etc. These are refered to as 'covariates' and one would normally 'adjust for' them using a regression model.  
Loosely speaking, for reasonable sample sizes and predictive covariates your confidence interval for utility bill reduction should then be smaller.  Some discussion is here.
So, one possible answer to the first question would be: find some good pre-treatment covariates for the folk in the original experiment and condition on them.  For the same reasons, finding some good covariates is also a pretty good idea for the second question.
