So i am performing k-means clustering on RFM variables (Recency, Frequency, Monetary). The RFM variables are in the form of quantiles (1-4). I used PCA and found the PCA components. I then used the elbow method to find the optimal number of clusters and then I use it in the k-means algorithm. Could anyone guide me if this is a correct method? Further, the clusters I get range on the graph, their axis ranges from -3 to 3 and I am not entirely sure why it ranges from that way.
Judging from the plot, there are no clusters.
K-means requires continuous variables to work well. The data you have has discrete steps (which causes the grid pattern in your plot).
There is no benefit of using PCA here. Use it only for visualization. The scale -3:3 that you don't understand is from PCA. So you probably have not understood PCA enough either.
Recency, Frequency, and Monetary Value (RFM) are measured on a Likert scale from 1-5. Likert scale variables are not amenable to Principal Component Analysis (PCA) or k-means clustering because those methods assume continuous measurement.
It would be more appropriate to use multiple correspondence analysis (MCA) and k-modes which assume nominal level measurement.