A single active US equity can produce millions of trade prints a day and far more quote updates. Storing a year of full-depth data for a broad universe runs into terabytes, which is why most research uses bars and only drops to ticks when microstructure matters.
It matters when your holding period is short, when you need to model queue position, or when you want honest fill assumptions. A backtest that assumes a fill at the bar close is guessing; one that replays the actual quote stream can check whether size was available.
Tick data is also the dirtiest data. Expect out-of-sequence timestamps, cancelled and corrected prints, condition codes that mark trades ineligible for high-low calculation, and venue-specific quirks. Cleaning rules are part of the strategy and should be version-controlled with it.
Related: quote-data, bid-ask-bounce, ohlcv-bar, bad-tick