By limiting the heat spread during a fire hazard, intumescent coatings are important components of passive protection systems. They swell due to heat induced reactions of micro constituents and are transformed into carbonaceous porous-like media, known as intumescent chars. Their multiscale inner structures, key elements of performance, are costly to predict by recurrent and large scale fire testing while numerical simulations are challenging due to complex kinetics. Hence, we propose a novel approach using the fractal theory and the random nature of events to conceptualize the coating expansion. Experimental specimens were obtained from fire protective coatings exposed to bench scale hydrocarbon fire. Mass fractals were evidenced in the slices of 3D sample volumes reconstructed from X-ray microtomography. Consequently, geometrical building blocks were simulated by random walk, active walk, aggregation-like and site percolation: physical-chemical modes of action were inherent in the attribution of the randomness. It is a first demonstration to conceptualize different types of intumescent actions by a generalized approach with dimensionless parameters at multiscale, thus eliminating the simulation of complex kinetics to obtain a realistic morphology. Also, fractal results brought new evidence to former chemical analyses on fire test residues trying to explain the kinetics of expansion. Expected outcomes are to predict virtually the reaction of fire protective systems hence to speed-up the assessment of fire performance through computed properties of virtual volumes.