This ends up acting like you the "seller" of hashing power getting paid in bitcoin to hash whatever altcoin is most profitable at the moment.
They have an auto-switching GUI client for windows, and allow you to point basically any miner that they support the algorithm at their service and get paid in bitcoin at a roughly comparable rate to mining the actual coin.
Suppose there's altcoin called ALT which can be mined very quickly and is worth let's say $0.00001 per coin. Mining a block takes 5 minutes and the payout is 10000 ALT coins or whatever. The 10000 coins are distributed across the pool, and you determine that on your setup you should be able to mine $3 USD in terms of BTC a day on the pool. You get wind that ALT is about to get pumped hard, which means your potential earnings could reach $20 per day (or whatever).
What's the problem (ignoring exchange fees) with going "all in"?
Firstly, ALT coin may have a confirmation time of several hours or even days. You can't sell unconfirmed coins.
Secondly, the exchange has to sell these coins elsewhere (usually another exchange). Because of the low value, this other exchange may have a rule that you can only sell in batches of 1 million. So even though you've mined X amount of coins, you still have to wait for the batch to be fulfilled.
Thirdly once the batch is fulfilled, you still have to wait for it to be sold on the other exchange.
And finally, the BTC <-> ALT exchange rate may fluctuate independently of known factors.
By the time this happens, your ALT coin will almost certainly be worth a different value to what they were when you first mined them.
It could be worth more or less, but the point is you can't efficiently forecast on most alt-coins when you use exchanges like Zpool and NiceHash. This is why they usually have several values for your "earnings" - Total, Unconfirmed, Confirmed, and Sold. It's also why "profit switching" algorithms and techniques are kind of a silly idea for most alt-coins - you cannot calculate profit based on current market values.
You as a "seller" get paid in bitcoin per mhash/s (or whatever the unit is for that algorithm). so if you are hashing "DaggerHashimoto" at 26Mhash/s for 24 hours, you are getting paid x btc per mhash/s
The "buyers" are basically then paying NiceHash to say "I want x ghash/s of this algorithm and I want it going toward this mining pool and I'll pay you X bitcoin for it." So then NiceHash plays the part of the broker and matches that X bitcoin payment up with as many miners as is needed to reach the threshold of ghash/s and takes a little off the top as a fee.
It's not perfect, you may end up taking a sub-optimal algorithm and making less than you could have if you did it perfectly, but it works well enough that I make profit in bitcoin.
https://www.nicehash.com/?p=calc
That's usually fine for the bigger, more established alts (since the known alts use the known algos - DaggerHashimoto, Equihash etc), but not for the alts which can potentially generate the most profit in shorter periods of time. So if a coin that uses some obscure algo gets pumped for a few days it's going to skew the chart.
The real benefit of Nicehash is that it's more accessible for people who are new to crypto mining. If you work out the numbers, mining ETH (for example) directly is almost always more profitable than using Nicehash. Nicehash is just simpler and more convenient.
But I don't want ETH, I want BTC. And while it's not a knock against ETH (It seems like a really cool technology!), I just don't want to hold my money in it and deal with it separately than I do my BTC. And I'm more than willing to pay that 4% fee to be able to hold my BTC securely with the ecosystem I have setup to sell and convert it to USD if/when I need to.