I write all my tools myself, so I know exactly what I can do without any actual coding, and I know what additional features are easy to implement. I try to avoid projects which require a lot of additional new code. Then I estimate the hours and multiply by two. The risk of additional work is on me, and I was once off by a factor of around 3, but I try to only accept projects where any newly developed code will either be useful for a long-term contract with the client (which then basically consists of me running a cronjob for money), or where I know that the internal tools I have to develop will be useful for future jobs. Examples of such tools are data converters, data analysis tools, etc.

To paraphrase a previous employer's strategy: fixed fee projects are for ones you plan to do over and over where it makes sense to invest at getting good at them.

The first one you lose a bunch of money, the second you might break even if you are lucky, and the tenth onward you make a bunch of money.