I totally agree with the parent.
I'm talking from a first hand experience, since I've been working on a Java POS solution for past 18 months. I'm a part of a new team of 6 developers dedicated to this project. The company has been (continuously and successfully) in PC POS business for the past 15 years. We are replacing their legacy DOS application. Here are some lesson we learned (some are POS specific, some are valid for enterprise SW development in general):
1. We couldn't have done it, if the company didn't already have extensive experience in this market. Programming experience is not enough, one has to know the market and customer expectations. PC POS is a tough market with a lot of competition that has been around for ages (lately including Microsoft). You can not get by with an amateur product.
2. Forget about the low-end off-the-shelf PC hardware. POS needs reliability (every hour of down-time means real money lost + getting a bad name with customers). Go for all-in-one specialized POS systems. They start at about $1500 (+options).
3. POS market needs support, even if some inexperienced first-time merchants think otherwise. Prepare some support plans and people to do it.
4. Store owners/managers put A LOT of emphasis on controlling their own staff. All serious POS systems are concurrent multi-user enabled (multiple users logged-in to application at the same time with fast GUI switching). RFID dongle for every user is what really works in the end.
5. You will not be able to customize the application for every small customer and their POS process. A lot of times you will have to teach them best POS practices and have this supported in the application. However you will have to bow to the bigger customers and customize the application for them. Result: you will end up with a lot of different sub-versions of the application. Bug-fixing and upgrade process is going to be a pain. Plan for it before all this starts.
6. Sales people will always sell features that are not there yet or are impossible to produce (due to technical or cost constraint). Train sales people to check with you first.
7. Have people that are not developers test the application before release. Testing your own code virtually means nothing. Once you release it to the customers you'll be amazed at how many bugs they can discover. As if all the unit testing and hand testing was in vain (it is not, continue with unit testing etc)
8. Write comments and API documentation (javadoc or equivalent). I REALLY MEAN IT. I often find myself wondering what a piece of code is doing, just to discover that I wrote it 6 months ago.
9. Do not write UML or any other developer documentation that is outside of code. It's just additional work that is always out of sync with code. And it only makes PHBs happy. Only exception to this rule is a short architecture overview of max 10 pages - this will save your time when new developers arrive on the project.
10. Choose a good IDE and be willing to pay money for it. Good IDE is much more important than a good car! Some developers I know are you willing to spend obscene amounts on a car but nothing on an IDE. You can always get a cab if your car lets you down, but IDE.. Again - this is no place to save money.
11. You should already know this: you need a versioning control system and a continuous integration system. Go with open source ones - they just work.
12. Automatic update system for your application is A GOOD THING. Selective (forceable, with consent) update per customer is even better. Try to stage updates - update first customers that are close to you (in miles and mentality). If all works well, roll out to everybody.
13. In case of troubleshooting a remote connection (VPN, RD, SSH, whatever..) to customers computer is worth a lot. Try to plan this upfront.
That's about it. Good luck writing your own POS application, just to save $2k.
And next time you need a new car, just build one.
