GiveDirectly (https://www.givedirectly.org/) gives 88% of the money straight to recipients and offers a nice baseline for charity work.
There's also better metrics than "how much money does the charity take off the top" which you can learn about in Will McAskill's book "Doing Good Better," namely you can try to track things like "how much good are we doing for the amount of money we get." The main idea is that if a charity can use more of its donation money in order to have a better impact, then that's what it should do.
The point is that the money can be used only once, so your money gets 88% of its value. A lender uses the money more than once, so they can provide more value from a donation.
The money is used more than once as it stays in that community rather than being paid back with interest.
Often money earned, lent or given to the least well to do has the highest velocity in changing hands, resulting in much more positive economic impact than if the money were given to people who were better off.
A costly loan provides only a tiny fraction of value, compared to an actual transfer. You can't sustainably "loan" money to people in extreme poverty, they can't possibly pay you back. And once these people have bought assets with their transfer they can even access traditional loans.
We're talking about people in truly extreme poverty here, which is trivial to verify. Sure, you can and should make some effort to direct your aid to the very poorest, etc. but at this level, fraud just doesn't matter the way you think it does.
There's also better metrics than "how much money does the charity take off the top" which you can learn about in Will McAskill's book "Doing Good Better," namely you can try to track things like "how much good are we doing for the amount of money we get." The main idea is that if a charity can use more of its donation money in order to have a better impact, then that's what it should do.