Another reminder: Things That Have Caught, Etc. is going paid from Monday 23 September.
Another reminder that we live in late-capitalism market economy where people are compelled to build a “personal brand”, perform for attention and how much subscriptions will cost: $7/month or $150/year (they’re exactly the same, you didn’t miss anything, it’s just a way for some people who want to pay more to be able to pay more)
A red button:
The actual content after, as they say, the “jump”:
Now, I don’t want to alarm anyone, but I am sitting on a Boeing 737-800 (not one of the 737 MAX 8 ones that has been grounded, obviously), and the captain has just told us that our flight will be delayed because, and I paraphrase:
“Well, we’ve got two computers here that control our cabin pressure and things like whether the oxygen masks drop down for you, and they back each other up. Like all computers, one of them has stopped working which means we’ll be a little delayed while we fill out some paperwork so we can get going with just one of them. These days this is pretty much just a computer with wings, so give us 10-15 minutes and we’ll be on our way...”
Reader, let me tell you: I know enough to be dangerous with computers and boy do I have opinions about how Boeing went about the 737 MAX 8.
Seeing as I’m sitting here with nothing to do but to be slightly confused about how you can have a system designed with a failover/backup, but if one of them stops working it’s totally okay to... not have the backup working and fill in some paperwork and be on your merry way, I’ve been googling (brb: googling how cabin pressure computers work) and because I am stuck here and you are stuck here with this email, then you get to read about what I’ve learned.
(Planes! They’re like buses! Only presumably now they’re computers with wheels.)
There are two digital pressure cabin controllers on the 737-800 that I’m on, they take turns and alternate between being the main controller on a flight, failing over to the other one. This makes sense, right? It’s not like the Shuttle or anything where you’ve got three computers that help control your brick falling at terminal velocity from low earth orbit, and if any of those three computers have got an idiosyncratic opinion about what’s happening with your brick they have a chat with the other two and if the other two are in agreement they get to lock that other computer away and never talk to it again.
So, anyway. There’s a great Stack Exchange question about why there’s a manual cabin pressure controller mode which at first glance seems like it should have an obvious answer (for when your computer stops working, duh), but the reason behind the not-dumb question is because do you remember that one time there was a big plane crash because the cabin pressure controller on the 737 was in manual mode and the crew didn’t notice until they passed out and the plane crashed? I mean, the obvious answer is because yes, your computers will occasionally stop working because: computers ¯\_(ツ)_/¯, and clearly Boeing has been in for some criticism about being in manual mode without some sort of reminder that you are indeed in manual mode and if you don’t pay attention you might all pass out due to hypoxia.
In the meantime, we’ve had an update. Of the two pressurization controllers we have, one of them isn’t working and a mechanic is on board to disable it, and then there’s a bunch of paperwork to get sorted out before we get on our way.
It is now 40 minutes since we were told the problem would take 10-15 minutes to resolve and we are indeed on our way. I am, in a way, quite happy that we learned a mechanic had come on board to disable the Not Working Very Well Pressure Controller because if our Only Working Pressure Controller stopped controlling pressure we wouldn’t want it to switch over to the Not Working Very Well Pressure Controller, would we? I am also very grateful that paperwork has been filed so the next people on this plane will benefit from a meticulous paper trail in the event of either my fellow passengers or our plane’s subsequent passengers encountering a sudden cabin depressurization event.
Planes. Just big computers with wings.
It’s 7:54am PDT, on September 12, 2019.
One short one and one long one today. Let’s go.
The driver on the way home from Portland airport was a gigging musician; we’d been talking about missing the snow that had been forecast and what had happened last time (known locally, of course, as the snowpocalypse). Turned out they’d missed that one, after having been called up for a last-minute job playing on a cruise ship. They’d done cruise ships before - signing up for a 3-6 month contract that would feel like jail, but this one, the one that meant missing the snowpocalypse, was an impromptu three week gig that meant going down south to Mexico.
Three weeks was fine, they said. They wouldn’t do too much longer than that now, feels too much like jail.
“Huh,” I said. “So for gigs like that, do they fly you down there to meet the ship, or do you have to make your own-“
“Nah, all the transpo’s covered. They always cover to and from.”
“Oh, so I guess it is a bit like jail, then.”
Some of you may know that my day job involves helping the State of California replace and modernize its Child Welfare IT system. I had a meeting in Sacramento the other week where there was me, a colleague, and a big whiteboard where I’d written the words DATA STRATEGY and let me tell you that although at first it was disappointing that we were not there to talk about a) the Soong-type positronic android character DATA, from acclaimed and culturally relevant television show Star Trek: The Next Generation; b) the aforementioned DATA and any of their STRATEGIES as shown during seven seasons of television, at least three of which in aggregate were probably quite good and not completely embarrassing, we did indeed have a really interesting meeting.
It was the kind of meeting that left me thinking I’d either uncovered something really quite interesting and relevant or I’d uncovered something unbelievably trite and obvious. I would like to believe it’s more the former than the latter, so I’m going to type about it now.
So. Imagine you have a big government computer system. Ignore the work of child welfare, imagine it’s just any big computer system that works with hundreds of thousands of people. It is used to carry out, to administer, some sort of policy. Normally it’s a social policy, unless it’s tax related, because what else is going to have that many users.
Imagine you have this system and it’s the year 2019 and someone has pointed out, say, that the system you have is getting a bit old and maybe it’s also not compliant with whatever upstream regulations or requirements you have, and there’s also the matter of it not really doing what its government users need it to do, and even when it does do those things, nobody’s really that happy about it. Imagine that the only reason they’re happy about it is some sort of combination of “because we have to use it” and “we tried using paper and that’s just even worse, but not so much worse, because we’re also using a lot of paper too. In fact, it’s a toss-up between paper and this thing, to be honest”.
None of this, of course, is any particular person’s fault. This was an amazing computer system when it was released (which had nothing to do with you, the imaginer), state of the art and all that stuff. Not so much now, though.
Anyway, you have this big computer system and you’re really excited because you’re a data scientist or you’re a Product Person, or you’re a Service Designer, and it’s exciting because you’re imagining all this Data that you’re going to have and how you’re going to improve outcomes and isn’t that what anyone really wants to do, at the end of the day, to improve outcomes for vulnerable people?
So you take a look at the “data” (ie someone lets you go spelunking in DB2 or whatever and run some reports) and you put your hand over your mouth and you gasp a little bit, maybe you even emit a sort of astonished squeak, because it turns out you can’t really do anything much useful with the data.
It’s all over the place. For some of the reports you’re required to produce, maybe you’re not even collecting all of the required elements. Okay, maybe for all of them. And when there are multiple reports that cover similar concepts, it turns out you’ve got duplicate entities. And none of them really match up!
Okay, so your upstream reports aren’t great, so you decide to take a look at process or outcome data. Are we actually doing a good job? Do we know?
Reader, you may wish to remove your hand from your mouth and replace it with your other hand, and emit the same squeaking noise, for you may not be able to answer that question! You may not know! You may be able to say, at a coarse-grained level, that it took someone x days from filing an application to being granted a license. But if I were to ask you, hey, person sitting in a meeting with Dan, do you know what the bottlenecks are in this process, you would look down sorrowfully at the floor and say No Dan, I do not, because our system does not record anything like the status or duration of subsidiary tasks in the context of a larger process, or if it did, it does so unreliable and inconsistently.
“Oh,” says I, who by now has totally gotten over this not being a meeting about a Star Trek: The Next Generation character. “Hmm”.
(For the record, I am condensing this meeting and straight-up making up a lot of it, especially the Star Trek parts, to make it entertaining and readable, and I only want feedback on the latter, please).
There’s a bit of conversation and it turns out that of the upstream-required reporting, only two kinds of data are being recorded. They can basically be broken down into i) The Kind Of Data We Need Because Otherwise People Won’t Give Us Money, and ii) The Kind Of Data We Need Because Otherwise I Literally Can’t Get To The Next Screen Of The Thing I Want To Do.
At this point, the epiphany strikes:
“The problem,” he reckons, with the confidence of a man, “is that you’re only collecting administrative and statistical reporting data.”
“Okay, let me try again. This is a system that was built to administer child welfare. It is a bureaucratic system. It records structured data only to the extent required to process and proceed in a bureaucracy.”
A slight nod.
“This isn’t to say that there isn’t outcome oriented data in the system - but it’s unstructured and, I’m guessing, hidden away long-hand in things like meeting notes and transcripts.”
A slightly more enthusiastic nod.
“This isn’t anyone’s fault. The system is working exactly as designed.”
What we’ve got here is an entirely typical administrative computer system, one borne out of an era where the promise of computing was to do what you did on paper, but faster, and... more-r.
The joke, amongst people who work on what is laughably called digital transformation (which is about computers and not, say, “plastic surgery, but for fingers”) is that bad digital transformation (still the computer kind) takes real life existing right now paper forms and puts them on computers so people can fill them in marginally more easily (you don’t need paper!) and maybe you need fewer people to reject social services applications because computers can do that by reading not-paper-forms badly.
But the problem is a fundamental one, and a bit like a category error. There is the business of administering child welfare (are you allowed to have this thing, did we find out whether we can give someone money so they can give you this thing, did you indeed receive the thing, who do we have to tell that you got the thing) and actually delivering child welfare, which is to say: keeping children safe, improving their wellbeing, and supporting permanency with their familial and cultural ties.
One of those things is not like the other.
This is not to say that social workers and caseworkers have not been concentrating on the latter, on actually delivering child welfare! But, at a high level, the computer systems they historically work with, because of how government was conceived at the time, are administrative in nature. Fill in this form so that you can do something about a child’s safety.
Another way of looking at this, and this is going to sound really obvious and trite to the people who work in this area, and hats off to you, is the difference between process and outcome.
The computer systems we’re busy modernizing and replacing and the ones that you encounter when, say, an airline has prematurely announced a strike and cancelled a bunch of flights and you’re trying to call them to get a refund, that system is a system designed to enforce an industrial process. It is not a system, necessarily, designed around any particular outcome, other than any outcome strictly defined as the process outcome.
Let me try again: the only reason why we’re able to run a report that says it took this many days for Bucky to get licensed as a foster parent is because administratively, we put a RECEIVED ON stamp when we got his application and put a LICENSE VALID FROM date when his application is approved. There wasn’t a goal of “let’s make sure we know how long it takes to get approved because if it ends up taking over 230 days for Bucky to get approved as a foster parent and yes I know we had to look into that whole episode when he went on holiday in Russia but he’s back now and he’s much better, then maybe we need to improve our foster licensing process”. There just wasn’t. The goal was: license foster parents.
But what if we weren’t so process driven? What if we were outcome driven?
I remember a conversation I had many years ago with a friend at the UK’s Government Digital Service (TWO! TWO EPISODES IN AND I’VE ALREADY MENTIONED GDS) who was working on driver licensing. It was the kind of conversation I imagine people might get scolded for having in an earlier time, because it was about whether people are looking to get a driving license, or whether they just want to drive. There’s a difference, you see, because if you want to get a driving license there’s all the stuff about the procedure of getting a driving license. But if you want to drive, well, that’s a whole different framing.
In other words: “as a user I want a driving license so that” versus “as a user I want to drive so that”.
In America it’s normally called the Department of Motor Vehicles, in the UK it’s the DVLA - the driver vehicle licensing agency. But there’s a way of looking at why these entities exist and what they exist for where the answer is: what about a Department for Road Safety? Or a Department for Transport Safety?
Because down one route, the job of the department is to license drivers or vehicles. Make sure that they are safe for the road. Follow this process and get a piece of paper, or a flag in a system, that says that this person or this internal combustion engine/electric motor with wheels meets some definition of safe, as expressed by this set of rules and regulations.
But the thing is, if you live in a bureaucracy that’s process driven, it can be incredibly hard to remember that the bureaucracy exists for the outcome. The outcome is not the conclusion of bureaucracy, it just feels like it and looks like it because you’re part of a big machine that on a very real level, decides whether or not people get a piece of plastic and then issues that piece of plastic.
So a particular problem I get to see, and I have the utmost sympathy here, is the occasional look of terror when we ask someone: what is it that we want to get out of this? Not just a faster process, but what’s the process supposed to achieve? Does it do that? How would we even know?
How would we even know if our current processes are working in the way we’d like them to? And the answer right now be: we have no real idea, because we don’t collect any of the data.
And from there we get into a whole psychological safety trap because what you don’t know can’t hurt you and as soon as you start measuring, you’re going to know much more than you did before about whether you’re doing a good job or not.
(This is a placeholder and a reminder that I really should have a catchup chat with my friend Dave Guarino, who was instrumental in incrementally building a new way to apply for food stamps in the State of California, because he has a great story about how you progressively insert instrumentation into an existing process).
Anyway, that’s the epiphany. If you’re trying to do service design in a big organization, consider the fact that you work in a bureaucracy (I mean, you know this, but do you really know it?) and that the bureaucracy may explain the huge goddamn lacunae and blind spot where No Data Exists.
For the work that I’m doing, I feel this is really valuable in explaining to some people - leadership, managers, executives, people with the word Director in their job title - in a non-confrontational way, why there are problems with moving to a performance-based, outcome focused culture (sorry). Too much processing, not enough understanding. And then with that understanding, being able to build a compelling case for the DATA STRATEGY on that whiteboard and making the data scientist super happy because finally they’ll be able to work with a FISTFUL OF DATAS and ha see I did it, I fucking did it and you can’t stop me.
It’s 8:39am PDT, on September 12, 2019. iA Writer reckons this draft is clocking in at over 3,000 words and my inner evil business guy is shouting WHY ARE YOU WRITING SO MANY WORDS BEFORE THE SUBSCRIPTIONS START and I am saying BECAUSE I AM A HUMAN BEING, GO AWAY EVIL BUSINESS GUY.
Anyway, a reminder: Things That Have Caught, etc. is going paid from Monday 23 September.
$7/month, or $150/year depending on how much you feel like paying: a little, or a bit more. Now, I don’t want to pressure you or anything but I can’t stop looking at my Stripe dashboard because several very smart, really cool and interesting people have already subscribed and you definitely want to be like them, but if you don’t that’s totally cool and you’re still smart, really cool and interesting. (You can see why I don’t work in advertising anymore).
All proceeds go directly to me, or if they’re behaving, the kids’ college funds. (Ha, what are colleges even).
Or! $0.00/month and you get one or two newsletters a month.
I may add another plan! Who knows, it all depends on how things go!
And here’s a red subscribe button that I think should be a nicer color (you know, like a nice blue, or a green?) and not an angry warning color. I suppose it could be worse and be like that yellow/black police tape.
Plane’s still in the air and no oxygen masks have dropped, plus my ears still work. So there’s that.
How are you?
PS. I LIKE REPLIES! Reply to this email and I think it goes straight to me? I reply to people who reply to me, even if you just say “hi”. Hello to all the people who already sent me replies!