Amlodipine Besylate Tablets - Process Validation

Dear members,
We have problem when trying to validate our manufacturing process of lisinopril tablets. The process is direct compression. We use stainless steel thief sampler to take samples for powder blend uniformity. After the test results were reported we found low drug content in powder and normal drug content in tablets.
When we developed the assay method, we found that if the tablets were finely powdered the assay recovery would be low. By this reason, we choose to put several tablets directly in a volumetric flask and then shake with the diluent, instead of grinding the tablets, in the sample solution preparation. So, we did not homogenize the powder blend samples using mortar and pestle. The result was still low.
Does anyone have the same experience (using the similar or different API)? What is the root cause of this phenomenon and how to solve it? Thanks in advance.