Unless you start with covering something completely new, I think it makes sense to start as analytics provider for insurance. Becoming a carrier is expensive, and MGA can only sell existing policies already created by a carrier.
If you sell existing policy you are at severe data disadvantage comparing to existing insurance companies, even with superior tech.
Shameless plug: At https://tensorflight.com we are working on P&C insurance and we focus on analytics for commercial properties mentioned in the article. Please get in touch at ( kozikow [at] tensorflight.com ) if you are interested in the subject.
This is not really related to your post, but you mentioned MGAs, which reminded me of something -
If you are getting into this business and are looking to write in most/all 50 states, one of the more cost-effective ways to acquire the licensing and company structure is to purchase a struggling/defunct carrier or MGA (depending on what your needs are). I didn't say cheap, but it's often cheaper than structuring and licensing yourself from nothing.
Lemonade probably should ramp up slowly until they have more experience data to look at (my opinion only of course), so it's probably not too disadvantageous.
Plus, if you're going the route of getting licenses via acquisition, you need the money to make that investment. It can be cost effective to go that way, but not cheap, so if you're not trying to blow all of your cash at once and can benefit from a slow roll to 50 states, the traditional route will be fine.
Shameless plug: At https://tensorflight.com we are working on P&C insurance and we focus on analytics for commercial properties mentioned in the article. Please get in touch at ( kozikow [at] tensorflight.com ) if you are interested in the subject.