Buy influencers with your budget (bigger audiences cost more), run the campaign and watch awareness spread round by round through the network. Then compare strategies over many random runs.
Running check...
The science
Network. The audience is a scale-free graph built by preferential attachment (Barabasi and Albert 1999, m = 2): a few accounts have very many followers (mega-influencers), most have few (micro-influencers), as observed on real social platforms.
Campaign. An influencer with degree k costs 1 + 0.2·k budget units. When a seed posts, each follower becomes aware with probability q (platform reach). Every newly aware person then shares to each neighbour once with probability p (organic virality): the independent cascade model of Kempe, Kleinberg and Tardos (2003). Awareness A is the share of aware nodes; conversions = A·N·conv; ROI = (conversions·revenue − spend)/spend.
Percolation check. For transmissibility T a cascade from one random node is the component of a bond-percolation graph. The giant cascade appears when T·(⟨k²⟩−⟨k⟩)/⟨k⟩ > 1, so heavy-tailed networks have a very low threshold. Below it, the mean cascade size from a random seed is (Newman 2002) 1 + T⟨k⟩/(1 − T(⟨k²⟩−⟨k⟩)/⟨k⟩). The page measures ⟨k⟩ and ⟨k²⟩ of the generated graph and compares thousands of simulated cascades at T = half the threshold with that formula.
Simplifications. One product, one platform, static graph, no fatigue, no negative sentiment, no paid-versus-organic distinction, and no cost data from any real market; all costs and rates are illustrative. Results from the strategy comparison are Monte Carlo averages with sampling noise.