Five things we check before replacing a legacy application
An internal application that has run for fifteen years has usually outlived its documentation and sometimes its authors. Replacing it looks like a programming job. Most of the risk is somewhere else, in the things nobody wrote down. This is the list we work through before we estimate anything.
1. What people actually do with it
We sit with the people who use the application and watch a normal day. The screens show what the system was designed to do. The staff show what it is used for now, including the workarounds: the field that is used for something other than its label, the report that is exported and corrected by hand every month.
Those workarounds are requirements. If the new system does not account for them, staff will rebuild them in a spreadsheet within a month.
2. What the data really looks like
We ask for a copy of the data early, under whatever confidentiality terms the client needs. Old databases contain duplicate records, dates in three formats, and codes whose meaning changed in 2011. The effort of migration depends far more on this than on the number of tables.
3. Who is allowed to do what
Older applications often have one shared login, or permissions that exist only as office convention. A replacement is the moment to write down the roles properly: who can view, who can change, who can approve, and who must never see a particular record. This takes several conversations and should not be left to the developers to guess.
4. What depends on it
Something else usually reads from the old system: a finance export, a public web page, a nightly file sent to another department. These connections are rarely documented and tend to be discovered when they stop working. We look for them in scheduled tasks, shared folders and the memory of whoever has been there longest.
5. How the switch will happen
We plan the cutover before we write code. Will both systems run side by side for a period? Who checks that the migrated records match? What happens if a problem is found in the second week? The old system should stay available, read-only, until the new one has been formally accepted.
What comes out of this
The result is a short written scope: what the new application must do, what data moves and how it will be verified, the roles, the connections to other systems, and the cutover plan. Only then do we give an estimate.
This step takes time, and it is tempting to skip it. In our experience it is the cheapest part of the project, and the part that decides whether the rest goes well.