Survey research encounters ever more problems related to declining response rates and growing costs, which seriously deteriorates overall quality and optimization of survey data collection. Web surveys offered a promising potential to cope with the cost issues, but at the same time further aggravated the nonresponse problem. In addition, the coverage issues still prevent successful applications of web surveys on probability samples of general populations. These problems present one of the greatest challenges of contemporary survey methodology. Methodological literature already considered the response-related problems of web surveys and suggested some promising mixed-mode designs that include the web mode. However, the cost issues remain highly underexplored. This severely limits the reach of survey methodology in identifying the most optimal survey design for specific purposes. The paper addresses this problem by discussing the possibilities of optimal integration of the  web into probability-based sample surveys. It is grounded on a study conducted in 2008 among Slovenian citizens. An official Eurostat survey on ICT use in households was performedusing an experimental design that manipulated different  combinations of survey modes (web, mail, telephone and face-to-face) and types of incentives (no incentives, a small gift and monetary incentive).

The problem of optimization is addressed through the relations between response rates and research costs, using the Mean Square Error (MSE) concept. These results, supported by the findings on interaction between factors influencing response rates, present the foundation of a pilot model for optimization of a survey process. Such model allows identification of the most appropriate selection of solicitation, reminders, incentives and and survey mode for individual respondents. The paper evaluates this optimization model and outlines some possibilities of future development that will enable automatic detection of the optimal survey strategy at different phases of a survey project for different groups of respondents.