By comparing the results from two language versions of a computer based intervention program (Magical garden, MG) we aim to learn how to best increase early mathematic skills in 4-5-year olds with a first language (L1, Arabic) different from the official language used in school (L2, Swedish). Research on acquisition of novel concepts favors learning in L1 followed by L2, to learning in L2 only (Perozzi & Sanhez, 1992). However, this has not been researched in mathematics even though often suggested (Clements, Sarama, Wolfe, & Spitler, 2013). The game is based on results by Griffin and colleagues (Griffin, Case, & Siegler, 1994) and focuses on promoting an understanding of early numeracy. Children learn by teaching a panda how to play the game, which is adaptive to their success rate. The game is socially inclusive a) everyone plays the same scenarios, though at different levels and b) the garden grows with amount played independent of level. By virtue of the experimental design, half of the children play the version hypothesized to be less effective. However, all ECCs will have free access to both versions after the intervention period. Previous studies show monolingual children using MG developing their number sense (Gulz, 2018). Here we will discuss the preliminary findings with bilingual children. Importantly, we will discuss our challenges when implementing the study at challenged ECCs. Interventions based on educational software can easily be scaled-up and teachers can deploy them even with little own knowledge and interest in math (Praet & Desoete, 2014).