Firefighters would greatly benefit from a technology based on predictive fire simulations, able to assist their decision making process. For those simulations to be useful, they need a certain degree of precision and resolution that can only be provided by CFD type fire models. But CFD simulations typically take large periods of time to complete, and their results would thus not be available in time to be of use during an emergency. Due to the high complexity of fire spread dynamics that arises from the interaction between solid and gas phase and the corresponding physical-chemical processes (e.g. pyrolysis), the spread of the fire cannot be predicted from first principles in real-time using contemporary computers, and has to be given as parameters to the model. Data can be incorporated into the model to characterise the fire, but only a limited range of measurements are recorded in current buildings. While it might be possible that buildings of the future incorporate a higher density of sensors than contemporary buildings, it is likely that emergency response systems will have access only to conventional data such as smoke detectors and sprinkler activation time for the foreseeable future. In this study the use of conventional detection and suppression devices for the estimation of fire characteristics by means of an inverse modelling framework is explored. Additionally to the growth rate of the fire, the location of the fire origin is successfully estimated. Inverse CFD modelling and tangent linearisation is used to assimilate the data. The nature of the incoming data is consistent with current detection and suppression devices, in such that only a time of activation is recorded and fed into the model. It is shown that the growth rate of the fire and the location of its origin can be correctly and efficiently estimated using sprinkler and smoke detector activation time only. It is further shown that the estimated spread rate is not sensitive to fire origin location.