An aggregation method adapted to life insurance portfolios is presented. We aim at optimizing the computation time when using Monte Carlo simulations for best estimate liability calculation. The method is a two-step procedure. The first step consists in using statistical partitioning methods in order to gather insurance policies. The second step is the construction of a representative policy for each aforementioned groups. The efficiency of the aggregation method is illustrated on a real saving contracts portfolio within the frame of a cash flow projection model used for best estimate liabilities and solvency capital requirements computations. The procedure is already part of AXA France valuation process.