What do you like best?
The best feature of AppVeyor is they allow *any* artefacts to be captured from the build and downloaded easily from Appveyor. This is a big feature over Travis, which has a recipe for using Amazon S3 to store artefacts, but that turns out to be quite expensive for very active projects.
What do you dislike?
No support for Github organisation authorisation.
Slow builds for free accounts.
Protection of environment variables is sub-optimal. Several times a command has failed and Appveyor has dumped out all variables, including passwords, to the log. As the log is on the net, the log then needs to be deleted.
Recommendations to others considering the product
Be *very* careful with putting passwords in environment variables. Appveyor likes dumping them to the log if there is a problem with a command.
Avoid cmd scripts like the plague.
Create build logic as .ps1 scripts.
For Python, use the demo project's appveyor.yml: https://github.com/ogrisel/python-appveyor-demo
For coverage, use https://codecov.io/ instead of https://coveralls.io/ , works very well with Appveyor, but also integrated with Travis - it merges coverage data from Appveyor and Travis together, allowing cross-platform coverage data. coveralls.io fails badly in this scenario.
What business problems are you solving with the product? What benefits have you realized?
I primarily work on Pywikibot, a framework for automated maintenance of wikis.
Automatic regression testing of Python packages using Appveyor ensures Windows users do not suffer regressions caused by our primarily Linux/OSX development team.