PointToPoint Transfers are usually point-to-point interfaces between 2 systems.
 
The files tend to be created for specific purposes and it is unusual for the data in the file to be used by more than one receiving system. 

The format of the file is critical to the applications but could be column based, delimited (usually with commas) or XML format. 

Data transferred in this way can be either transactional or reference. It is most common for the data in a transactional transfer to be for a group of transactions rather than separate files for each transaction. 

The source application runs a report or routine to extract data to a text file with a specified format. It is then sent to the receiving application and an import routine is run to import and process the data.

This is not EnterpriseApplicationIntegration in the true sense but is what has been developed over time by many organizations and IT WORKS! There are vendors would have you believe that this is wrong and that you shouldn't integrate your applications in this way. If they said anything else they couldn't justify their products!

''Usual problem with the aforementioned products are shoddy, bloated, overpriced monstrosities. One of them I had the "honor" of trying out came on about 10 CDs), most of them being non-optional. Often they contain some "programming-for-nonprogrammers" tool for mapping and message routing which you must us e unless you want to reverse-engineer the underyling "language". And due to the complexity, you need highly trained experts (or vendor's consultants...) to work with the system.''

----

Anyone who has done this before knows that the answer always lies somewhere between the two extremes. Point-to-point integration works well for small numbers of integration endpoints. As that number grows, the number of channels grows exponentially requiring you to consider alternatives like EnterpriseApplicationIntegration and MessagingHub. --PeterProvost

Yes - and the problem is that the number of endpoints inevitably does grow (DecisionsAlwaysChange). --DavidAllsopp

----
PointToPointIntegration breaks if there are a lot of data transfers between different applications (each application linked to say 2 other) AND if number of applications is large (say 5 or more). Theoretically, you have to program O(n^2) different data transfers. Of course, this is the worst case (note the big-Oh). If applications are serialy linked or if they are already broken up into separate "islands", then you don't have a problem.

But then there are modules that are common to all applications in almost any enterprise. Obvious example is customer-related functionality. Just count the number of products on the market that promise "a single view of the customer"...

-- BostjanDolenc