If we consider the more general case of $$S_n(x)=\sum_{k=1}^n \frac{k}{k+1}{x^k}=\sum_{k=1}^n x^k\biggl(1-\frac{1}{k+1}\biggl)=\sum_{k=1}^n x^k-\sum_{k=1}^n \frac{x^k}{(k+1)}$$ the problem looks quite simple when the summation is done up to $\infty$.
For finite $n$, at least to me, the problem seems to be difficult since we have $$\sum_{k=1}^n x^k=\frac{x \left(1-x^n\right)}{1-x}$$
$$\sum_{k=1}^n \frac{x^k}{(k+1)}=-1-\frac{\log (1-x)}{x}-x^{n+1} \Phi (x,1,n+2)$$ where appears the Lerch transcendent function. This makes
$$S_n(x)=1+\frac{x \left(1-x^n\right)}{1-x}+\frac{\log (1-x)}{x}+x^{n+1} \Phi (x,1,n+2)$$
$$S_n\left(\frac 12\right)=2-2\log(2)+\frac{\Phi \left(\frac 12,1,n+2\right)}{2^{n+1}}$$ As shown below, the last term decreases very fast
$$\left(
\begin{array}{cc}
n & \frac{\Phi \left(\frac 12,1,n+2\right)}{2^{n+1}}\\
1 & 0.136294 \\
2 & 0.052961 \\
3 & 0.021711 \\
4 & 0.00921103 \\
5 & 0.00400269 \\
6 & 0.00177055 \\
7 & 0.000793989 \\
8 & 0.000359961 \\
9 & 0.000164649 \\
10 & 0.0000758704
\end{array}
\right)$$ and a quick and dirty nonlinear regression (done for $1\leq n \leq 20$) seems to show a quite good approximation by $$\frac{\Phi \left(\frac 12,1,n+2\right)}{2^{n+1}}\approx 0.414102\, e^{-1.11129\, n^{0.888026}}$$