It may help to use a membrane made of a different material.

See this document for example:http://www.sartorius.com/fileadmin/fm-dam/sartorius_media/Bioprocess-Solutions/Lab_Technologies/Lab_Purification/Ultrafiltration/Catalogues/Cata_Ultrafiltration-Products_SLU0001-e.pdf

To add to the above recommendations, keep in mind that every protein behaves differently some are more hydrophobic and some are more hydrophilic. This is why there are so many different material types that amicon sells. For example, you might want to try PVDF instead of cellulose.  Also, while absorption does happen, more commonly the problem is precipitation or the protein passing through the membrane. Are you sure your protein isn't precipitating or passing through the membrane? This is very simple to test, spec your input and output at 280, or run a gel to make sure that your protein isn't going through nor precipitating on the membrane.   

Try a different membrane material, as mentioned above. You also may want to try a smaller cutoff . . . roughly 1/3 kDa was what I was always taught, but proteins are shaped differently. If it's long and skinny (tube-shaped, etc.), it could be slipping through the membrane that way and, therefore, you'd want to use an even smaller cutoff. Have you tested the flow through on the Amicon to see if your protein is getting through?

When I had a sample situation like yours at the start, however, have had a very dilute protein solution I would block my device (15 mL Amicon centriprep - looks like a 50 mL centrifuge tube)  with an irrelevant protein first.  This blocking involves 3 washes of the device with buffer (PBS).  My 4th wash I would run the device like package insert then take the concentrate and measure A280 (background absorbance - blocker is bound).

I recover more than 90% of my material from a dilute solution or from a small sample - usually buffer exchange at same time,

Hopes this helps you - this has literally saved me (the day) a few times.

Best regards

Dan