# The Universe of Discourse

Sun, 04 Dec 2022

This complaint is a little stale, but maybe it will still be interesting. A while back I was traveling to California on business several times a year, and the company I worked for required that I use SAP Concur expense management software to submit receipts for reimbursement.

At one time I would have had many, many complaints about Concur. But today I will make only one. Here I am trying to explain to the Concur phone app where my expense occurred, maybe it was a cab ride from the airport or something.

I had to interact with this control every time there was another expense to report, so this is part of the app's core functionality.

There are a lot of good choices about how to order this list. The best ones require some work. The app might use the phone's location feature to figure out where it is and make an educated guess about how to order the place names. (“I'm in California, so I'll put those first.”) It could keep a count of how often this user has chosen each location before, and put most commonly chosen ones first. It could store a list of the locations the user has selected before and put the previously-selected ones before the ones that had never been selected. It could have asked, when the expense report was first created, if there was an associated location, say “California”, and then and then used that to put California places first, then United States places, then the rest. It could have a hardwired list of the importance of each place (or some proxy for that, like population) and put the most important places at the top.

The actual authors of SAP Concur's phone app did none of these things. I understand. Budgets are small, deadlines are tight, product managers can be pigheaded. Sometimes the programmer doesn't have the resources to do the best solution.

But this list isn't even alphabetized.

There are two places named Los Alamos; they are not adjacent. There are two places in Spain; they are also not adjacent. This is inexcusable. There is no resource constraint that is so stringent that it would prevent the programmers from replacing

    displaySelectionList(matches)


with

    displaySelectionList(matches.sorted())


They just didn't.

And then whoever reviewed the code, if there was a code review, didn't say “hey, why didn't you use displaySortedSelectionList here?”

And then the product manager didn't point at the screen and say “wouldn't it be better to alphabetize these?”

And the UX person, if there was one, didn't raise any red flag, or if they did nothing was done.

I don't know what Concur's software development and release process is like, but somehow it had a complete top-to-bottom failure of quality control and let this shit out the door.

I would love to know how this happened. I said a while back:

Assume that bad technical decisions are made rationally, for reasons that are not apparent.

I think this might be a useful counterexample. And if it isn't, if the individual decision-makers all made choices that were locally rational, it might be an instructive example on how an organization can be so dysfunctional and so filled with perverse incentives that it produces a stack of separately rational decisions that somehow add up to a failure to alphabetize a pick list.

### Addendum : A possible explanation

Dennis Felsing, a former employee of SAP working on their HANA database, has suggested how this might have come about. Suppose that the app originally used a database that produced the results already sorted, so that no sorting in the client was necessary, or at least any omitted sorting wouldn't have been noticed. Then later, the backend database was changed or upgraded to one that didn't have the autosorting feature. (This might have happened when Concur was acquired by SAP, if SAP insisted on converting the app to use HANA instead of whatever it had been using.)

This change could have broken many similar picklists in the same way. Perhaps there was large and complex project to replace the database backend, and the unsorted picklist were discovered relatively late and were among the less severe problems that had to be overcome. I said “there is no resource constraint that is so stringent that it would prevent the programmers from (sorting the list)”. But if fifty picklists broke all at the same time for the same reason? And you weren't sure where they all were in the code? At the tail end of a large, difficult project? It might have made good sense to put off the minor problems like unsorted picklists for a future development cycle. This seems quite plausible, and if it's true, then this is not a counterexample of “bad technical decisions are made rationally for reasons that are not apparent”. (I should add, though, that the sorting issue was not fixed in the next few years.)

In the earlier article I said “until I got the correct explanation, the only explanation I could think of was unlimited incompetence.” That happened this time also! I could not imagine a plausible explanation, but M. Felsing provided one that was so plausible I could imagine making the decision the same way myself. I wish I were better at thinking of this kind of explanation.