For me, I find that it depends on what I'm researching. If the feature, functionality etc... has been tried and tested and proves to work, I try not to reinvent the wheel and move into assimilation mode into my wire-frame, low-high fidelity prototype and screenshots. This process usually lasts half a day to a full day (depending on the complexity of the puzzle).

If it's something new and there's very little out there that addresses this particular puzzle I'm trying to solve, then I research not more than 2 days and start solving the puzzle. User testing will confirm the hypotheses used.

Hope this helps.

These guys do good work: https://dribbble.com/Creativedashhttps://dribbble.com/JulienRenvoyehttps://dribbble.com/ormanclarkhttps://dribbble.com/kolagehttps://dribbble.com/haraldurthorleifsson