First, DW/BI isn't so risky anymore. The industry data that many vendors/consultants use to frighten customers into parting with big bucks is old news. Today, fewer than 10% of DW projects are considered failures. True, hardly anybody gets everything they want to out of the project, but usually that's due to unreasonable expectations and a lack of appreciation for how poor data quality *really* is.
In addition to the basic blocking-and-tackling that every organization should perform (e.g. solid project approach, experienced people, rigorous source system/data analysis, executive & user commitment), a midsize retailer should consider whether it's just sales you want to examine, or is it also customer behavior, financials, inventories, suppliers, economic conditions, etc? Add 2-3 months for every subject area you plan to integrate. The more disparate data you integrate, the more data quality (especially inconsistencies) will bite you, and the lesser chance you'll find packaged analytic solutions to accommodate you -- meaning "more custom development." Keep in mind that the ultimate DW "user" is a business process, not a pair of eyeballs. Focus on analytics that can be assimilated into business decision-making to affect business performance...not on analytics that the users think might be cool.
Spending might approach $1M annually for an enterprise DW, but maybe less than $250K for a single subject area data mart, e.g. customer behavior. Typically, hardware, software (data integration & analytics) and services (internal & external) are 1/3 of the cost each. Remember Stats 101 and the value of data sampling (instead of building monsterous lower-performing databases) for strategic analysis. A stable data mart is a bad thing, but requires no more than 3 full-time people to keep it alive and well.
Retek, Stratum and JD Edwards are strong players in mid-tier retail analytic solutions. SAP is moving down the chain to compete there too.