How An Expedite Request Sunk the Titanic - Daniel S. Vacanti at FlowCon France 2019

[00:00:00] Ladies and gentlemen, good evening. I guess it's it's no it's no dark. Thank you for being here.
[00:00:08] I think a lot of you are stressed out about the about the strike and getting home, yeah? Uh I was going to do my best to get you out early, but unfortunately we started a little bit late. Um I've realized that the organizers of FloCon obviously don't like me very much, um because they put me on as as a last time slot. I don't I don't know what I did to you, I don't know what what I did. I think they got me confused with Carl Scotland. Has it has anyone ever did anyone see Carl's talk before? No? Um we're going to be talking today about uh about expedites. Uh this is a conference about flow, so we're going to be talking about expedites.
[00:00:43] I will do my best to get you out as quickly as possible. Um but that uh that doesn't mean don't ask questions. Um I'll probably pause several times throughout the presentation to ask for questions. I'd like to keep this as interactive as possible. I know it's it's late in the day for you, it's it's late in the day for me, um as well because I I think my jet lag may be kicking in, I'm not even sure what time zone it is. Um so maybe we could all take a nap together by the end of this thing. I don't know. Uh but but let's uh let let's let's let's jump jump right into it. Did everybody everybody see the movie Titanic? Did everybody? Is anyone not seen the movie Titanic? Yeah? That's what I really wanted to do. I wanted to start off by playing playing so, you know, a clip or or some music from from the movie. But uh we're going to have to skip that for a little bit. Um but everybody everybody remember this guy from uh from Titanic? Remember him? Does anybody remember his name? No? Nobody nobody remembers this guy? He was the the bad guy, no? Bruce Ismay? Bruce Ismay, so we're not we're not actually not going to be talking about this guy. Uh we're going to be talking about this guy. This is the real Bruce Ismay. Um Bruce Ismay was the uh at the time, uh in in the early 1900s, was the chairman of a uh shipbuilding company or ship operating company called the White Star Line. Um and about the turn of the century, early 1900s, uh the White Star Line was actually dropping a little bit in terms of overall competition in the world. Uh and Bruce Ismay wanted to get White the White Star Line back on top. And so he had this vision that he was going to go out and he was going to build the biggest ships that had ever sailed, ever. That was that was his vision for uh for for for turning the White Star Line around. And he had an idea, he he called them Olympic-class ships, and there were three of them. He had a vision for three of them. The Olympic, the Titanic, and the Britannic. Right? Now, the Olympic was built first, um construction started in 1908 in Belfast. Um everybody knows that all the best ships are built are are built in Scotland, but they decided to build this stuff in in Northern Ireland. Um and so they placed it, are there any Scottish people, any Northern Irish people in the audience? No, okay, good, so I can say these things. Um so the Olympic was built first, it was placed in into service in 1911. Right? So as of 1911, the Olympics out there and it's sailing around and it's it's doing its thing. Now Titanic was uh was built second, construction began in 1909 also in Belfast, same shipyard. Um and when it started when Titanic was started to be built, it was given a tentative maiden voyage date of sometime in early 1912. So about a year after Olympic would be finished, okay? Everybody following these dates, these dates become crucial here in just a second, as you can imagine. Right? As as construction was nearing completion on Titanic, Um Olympic, remember Olympic's out there and it's sailing around and it's doing its thing. So as construction nears completion on Titanic, uh the Olympic gets damaged. It throws a propeller. Right? Um and uh um and uh that's this why was it? Um so this is February 1912, Olympic throws a propeller. Right? So it has to go back to the shipyard, that shipyard in Belfast that I was talking about, um for repairs. The same shipyard that Titanic is being built. So Harland and Wolff are the shipbuilders, uh and now these guys have a decision to make. Um because they've already built Olympic, they've got that out, they've got all of their people focused on building Titanic, uh but now now Olympic has come back into port. Um and they need to figure out, well, um I've got Olympic sitting here, um and it needs to be repaired, right? It's it's losing a whole bunch of money, but I've got Titanic that needs to be finished, and I don't have enough people for both of these things, to handle both of these things. So the question is, how does Harland and Wolff make this decision? So some of the some of the things they're up against, right? So Olympic, it's already in service, it's already out there, it's already making money. It's already doing its thing. Right? Each day that the Olympic spends in port is lost money, right? Because it's supposed to be out there shipping. They've already got bookings, they've already got all these things, right? So each day it sits in port, it's losing money. Um and of course, allocating workers, if we were to take workers off of Titanic and put them on Olympic, that would of course delay the delivery of Titanic. Now on the other hand, we've got we've got Titanic, it's not finished yet. But however, each day that it spends in port, we're losing money because it should be out there sailing around the world doing its thing. Uh and of course, if we were to keep workers allocated to Titanic, then that would of course delay our opportunity to get Olympic back out. Right? So what do you do?
[00:05:36] What do you guys think? What do you do? You got this situation.
[00:05:40] Anybody? Any thoughts? Anyone who who doesn't know the answer?
[00:05:45] Or anybody who does?
[00:05:46] Repair it?
[00:05:47] So repair the Olympic?
[00:05:50] Anyone else?
[00:05:51] Why can't we do both?
[00:05:52] Why can't we do both? Yeah, right on, yes. Let's just go out and magically hire a whole bunch of people and or or better yet, um well because this is France, this doesn't really happen in France. But in America what would happened is is they would fire half of the people and then say still get both of those things done. That's that's what would normally happen. Um so uh So I'm hoping all of this sounds familiar. My guess is you've come across a situation like this once or twice before in your life. Um and and how did you, how did you handle this? How did you handle this particular situation? So let's say we've got we've got uh a sample con bond board like this. We've got um an analysis dev and test column. Don't let's not talk about the quality of this con bond board. But um and then all of a sudden this what is normally called an expedite shows up, this super valuable, super important, super whatever it is thing. Right? That we have to start right away, right? We absolutely have to start this thing right away. And so now it's in analysis and we've got three things sitting here in this analysis column, and now we need to figure out, well how do we decide which one of these things to work on? Because our board is signaling to us that we've only got capacity to work on two of these things, but now we got three things in there. So how do we decide? Right? Well, as you can imagine, that's what the rest of this talk is going to be about. The decisions around how um the sorry, the policies around um the decisions how the how we make these decisions? We're going to call these pull policies, and by the way, I'm going to I'm going to make it really hard for the camera guy to follow me. Um the um the decisions that we're we're making, these are going to be called these are going to be called pull policies. Uh pull policies have much more impact on predictability than most people think. In fact, usually, if you're running your process, probably the biggest determiner of um of how predictable your your process is, is not necessarily work in progress. It's more about our decisions around the order in which we're pulling things through our through our our system. So let's let's talk about that in a second. I just gave you the answer. Um once a customer or a manager gives you work, gives you a hands you something to work on, uh and you start working on that thing, what's the very first question that your manager or customer is going to ask you? Very first question, that I just showed you.
[00:08:10] When will it be done?
[00:08:11] Right? That's what that's what that's what that's what our customers care about. When will it be done? And of course, it's worse than that because our customers care a very about a very, very specific form of this answer. They're asking when will it be done, but they they're expecting an answer in return. What is the form of the answer that they are expecting? A date, a date or a number of days. It's even worse than that. Because let's say, just for example, this is my favorite example, oh yeah, what is what is the date today? December 12th? Is it December 12th? 13th? 12th? Okay. Good. Let's say it's December 12th, um and we tell our customers that uh we started this item and it's going to be done in 18 days. Remember, they want a number of days. So we started this item and we tell them it's going to be done in 18 days. When is our when are our customers expecting that item to be done?
[00:09:03] 30th of December.
[00:09:04] Our customers are not sitting there thinking, well, you know, uh the Christmas holidays are coming up, and uh nobody really works over Christmas. So um and and there's some weekends in there, nobody really works over the weekends, and of course, that means that now we're probably into the New Year, and they're New Year holidays too. So when they say 18 days, what they really mean is sometime in the middle of January. That's exactly not how our customers are thinking. They think about the world in elapsed time.
[00:09:31] So when we say, when they ask the question, when will it be done, they're looking for an answer in terms of a a date or a number of days and they're looking, they're their expectation is that number of days is always in terms of elapsed time. Well, why is that important? Well, let's run a little little thought experiment here. How long does it take you to get to work in the morning? Think about it, I don't know if you well maybe this week is not Actually, this week actually actually this week is a very, very, very good example. Um how many people take take the metro or take any train to get to work in the morning? Right? Normally, how long does it take you to get to work every day? Does anybody, anybody throw throw out a time?
[00:10:06] One hour?
[00:10:07] One one hour, 20 minutes. Right? Does it always take one hour or 20 minutes?
[00:10:13] No.
[00:10:16] No, right? The real answer to how long it takes you to get to work in the morning is it depends. What are some of the things that it depends on, whether you whether you take the train, whether you take a car, whether whatever, whether you walk even? What are some of the things that the amount of time it takes you to commute to work, what are some of the things that it depends on how long it takes you to get to work? What are some things? What's the what's the obvious answer going on right now?
[00:10:41] Strike.
[00:10:42] Yeah, whether there's a strike going on or not is a big it depends. What else? Even if there's not a strike going on, what else what else does it depend on? What sorry?
[00:10:50] Coffee machine? Delay?
[00:10:52] Coffee machine time, yeah. How many Starbucks we stop at on the way? Yeah. Um what else?
[00:10:59] The weather.
[00:11:00] The weather, right? If it's raining or not, if we're in a car, whether there's construction, whether we're dropping off our kids, what time we leave? Right? They're all kinds of things that go into that make up it depends. So try an experiment for me, my guess is because you guys are all experienced flow practitioners, you're already all doing this, I'm sure. But um uh and just for the record, I normally hate histograms, but for the purposes of purposes of this uh of this talk we are going to talk a lot about histograms, but just just so you know, I normally hate them. But what I want you to do is I want you to to build a histogram for me for uh for your commute time. And across the bottom is going to be the amount of time that it takes you to get to work, say in minutes. Right? So each each one of these things across the model uh bottom is in minutes, so it could be 11 minutes, 19 minutes, 28 minutes, whatever. Right? And then up the side is the frequency. How many days did it take me exactly 11 minutes? How many days did it take me exactly 19 minutes? How many days did it take me exactly 28 minutes? What what whatever that is, right? If you do that for every day that you go to work, right, um you know, track it over a couple weeks, couple months, whatever. If you do that, you will get a a histogram that looks like this. You get a shape of a histogram that looks something like this. Right? Trust me, I know these things. This is exactly what your what your commute time histogram will will uh will will look like. Right? And here's the deal, try the same thing for your process. Again, I'm sure you're all doing this. But try the same thing for your process. Right? We've got we've got a workflow, we've got hopefully, we've got a well-defined point at which we consider work to have started, we've got a well-defined point at which we consider work to have finished. Right? So just take a timer, right? Every time every time something crosses that start line, start your timer, every time something crosses that finish line, stop your timer for everything that goes through your process. Track that time, build a histogram and you'll get a histogram that looks something like this. What does this histogram look like?
[00:12:54] It looks yeah, pretty much the same. It pretty much looks the same as as as our as our commute histogram. So um what we've just done, you guys know this, what we've just done is calculate something that I'm going to refer to as cycle time, let's not get into any philosophical needless philosophical debates around cycle time, lead time, flow time, whatever, I don't care. For the for the purposes of this conversation, I'm going to I'm going to call these things cycle time. Cycle time is just again, for the purposes of this conversation, is really, really a simple definition, is just the amount of elapsed time, um it takes for a given amount of work to complete. So we have that well-defined point at which we start work, well-defined point at which we consider work to have finished, the total amount of elapsed time between those two points, we're going to call cycle time. Right? When will it be done for a single item is best answered by this flow metric cycle time. So the question is, why does why am I so confident that our cycle time histograms look like this? Right? That's a rhetorical question, you don't have to answer that because I'm going to answer that for you right now. Here's the deal, when an item is sitting in the backlog, let's say, let's say this item number three, when it's sitting in the backlog, do we know, before we've started it, before we've started it at all, do we know exactly how long it's going to take? Remember, we haven't started it yet, do we know how long it's going to take?
[00:14:17] That wasn't me, I don't think.
[00:14:21] Um we don't, we don't know how long it's going to take at all, right? We don't know if item number three is going to end up exactly there, exactly there, or exactly there. We have no idea. The reason we have no idea is because of all that all that it depends we just talked about. Remember? So, here's the deal, we can't think deterministically because we don't know how long it's going to take, um because because we're talking about predicting the future and the future's full of uncertainty. We can't think deterministically, which means now we have to start thinking probabilistically.
[00:14:54] Probabilistically is problematic for humans, we're just not very, very good at thinking probabilistically at all. But what I can do, in fact, I'm not even sure, I'm not even sure I know what it means to think probabilistically. Uh but what I can do is give you a very, very, very simple definition, I hope it's a very, very simple definition of thinking probabilistically. And all that means, all thinking probabilistic means is acknowledging that there is more than one possible future outcome. Right? Think about rolling a fair six-sided die. Right? Before we roll it, we don't know, we don't know what that outcome's going to be. There's more than one possible outcome, in fact, there are six possible outcomes. Right? This is what thinking probabilistic means, we just need to acknowledge, we don't know the future, there's more than one possible outcome, and each one of those each one of those possible outcomes has a probability associated with it. Right? So what that means then is when somebody says cycle time to you, right, when somebody says, hey, what's the cycle time of your process, what's the cycle time of your commute, what's the cycle time of whatever? You can't think of cycle time as a number. You can't. Because remember, we can't think deterministically, we have to think probabilistically, so back to our friend, our good old friend the histogram. You can't think of cycle time as a number, you have to think of it as a shape. And we talk about driving predictability into our process, when it comes to predictability, what we really want to focus on is that shape of the histogram, why does the shape of that histogram look like that and what are some of the things that we can do to influence that shape? So that's the question. What are some factors that influence the shape of the histogram? What are some factors that roll into the it depends? Remember what I said before, this is spoiler alert, I already gave you the answer here. The number one thing that is probably going to affect the shape of that histogram are your pull policies. Right? So let's talk about that why. Right? Let's try let's let's uh try another experiment. Right? And and for this experiment, we are going to be magically transported um to this this fantasy land, and in this fantasy land, we can control exactly how long it takes active time for items to complete. Right? Wouldn't this be a wonderful world to live in? But you know, hands-on active time, we are going to be able to control how long it takes for things to complete. Right? That's this is the land that we're going to live in. And this is the process that we're going to use to work on the items in this magical fantasy land. Right? We've got an analysis column split active and done, dev split active and done and in test, and we've got work in progress limits. What we're going to do is we're going to put 55 items into a backlog, and we're going to work those those 55 items through our process. And each one of those 55 items are going to are going to take exactly 10 days of active time to complete in each column. Right? We're going to have no other blocking events or no other added items that come to us. Right? Isn't this a wonderful place to be, but this is this is where we're going to live for the next 30 seconds or so. Right? So for example, um on day one of this um of this magical mythical process, you can see. We will have pulled two items into uh analysis active and there we still have items on our backlog, right? At the end of day 10, those two items in analysis active, remember, everything's taking exactly 10 days to complete. So those two items in analysis active now move to to dev active, because we have capacity for that, we've pulled in two items behind them. Right? At the end of day 20, this is what our board looks like. The things in dev active are done and the things in analysis active are done. But now we got a problem. What's our problem?
[00:18:38] How much capacity do we have in test?
[00:18:41] One. How many things do we have done? Two. We've got two that are ready to pull, but we've only got capacity to pull one. How do we choose? How do we choose? What are some normal things that people consider when they're trying to determine which one of these we should choose? What are some normal things that you hear all the time now?
[00:19:01] Priority?
[00:19:02] Priority. Yep. Priority. What else? Value is another good one. What else?
[00:19:09] People.
[00:19:10] People, if we have people. Yeah. Uh what people we maybe we have expertise required, cost of delay. Right, all of those things. Um all of those, what haven't you heard? What thing haven't you heard?
[00:19:26] And of course you can't tell me because you haven't heard it. So I'm going to tell you. Right? Um in this first example, let's let's see if we can't derive from a predictability perspective, let's see if we can't derive what maybe the best option is. So in this first example, we are we're going to pull items through in a strict strict FIFO order, strict FIFO, first in first out. Completely um, you know, so we're going to work them through one, two, three, four, five, six. Okay? Um so that means that at the beginning of day 21, we're going to pull item one through. Right? The next time we have capacity, we're going to pull item through, right? So very, very, very strict FIFO queuing order. We're going to do this over and over and over and we're going to track all of the outcomes, and after 50 simulations, does anybody want to guess, can anybody make a quick guess what you think our cycle time might be at the 85th percentile? Remember we're talking probabilistically, so whenever we're we we communicate a cycle time, we need to communicate an associated probability. Can anybody guess? So remember it's 10 days in each active column.
[00:20:28] We're purposely not accounting for wait time just yet.
[00:20:32] Anybody want to guess? 85% of the items, how long is it going to take them to get through the through the process? you know, so we're going to work them through. 1, 2, 3, 4, 5, 6. Okay? Um so that means that at the beginning of day 21, we're going to pull item one through. Right? The next time we have capacity, we're going to pull item through, right? So very, very, very strict FIFO queuing order. We're going to do this over and over and over and we're going to track all of the outcomes. And after 50 simulations, does anybody want to guess, can anybody make a quick guess what you think our cycle time might be at the 85th percentile. Remember we're talking probabilistically, so whenever we're we we communicate a cycle time, we need to communicate an associated probability. Can anybody guess? So remember, it's 10 days in each active column. We're purposely not accounting for wait time just yet. Anybody want to guess, 85% of the items, how long it's going to take them to get through the through the process. What would you, what's the absolute best they can be?
[00:20:43] 30 days. The absolute best it can be is 30 days. Do you think all of them are going to be 30 days? No. So what do you think? 85% of items, how long do you think?
[00:20:56] Around 210. Yeah, that might be a little bit high.
[00:21:01] Sorry, I think I heard somebody say something else?
[00:21:05] No, it's not, it's not as bad as maybe maybe you might think, right? So in in FIFO queuing, this is our results histogram, right? 85% of items get done in 50 days or less. So you can see we have, we have one item that got done in exactly, that's that first item, oops, got done in exactly 30 days. We had a, I think another two items that got done in 40 days, but then everything else takes exactly 50 days. Okay?
[00:21:29] So now let's try something else.
[00:21:32] Now we're going to extend this case, right? It's it's the same setup. Only now, what we're going to do is when we have these two items here, we're going to completely choose at random. Literally, flip a coin. And whatever one comes up, right? If if heads comes up, we're going to pick one, if tails comes up, we're going to pick two. We'll literally just just just going to pick everything at random. Okay? Everybody makes sense? Everything makes sense? Same, same simulation, right? Only now, instead of FIFO, we're just picking things at random. Okay?
[00:22:03] What do you think? What do you guys think this does to to our cycle time at this point? Remember before, in strict FIFO, 85% of our time 85% of the items was, uh, uh, 50 days or less. What do you think now? In this. Do you think, do you think overall our cycle times will be longer or shorter?
[00:22:20] Longer.
[00:22:21] Longer or shorter? This our only other option is the same, by the way. So what what do you guys think? In terms of numbers, can anybody shout out numbers, what do you think? Is this going to be 85% of items, what do you think?
[00:22:35] 40? 40 is a good guess from 40. Somebody thinks it's it's going to be a little bit shorter, yeah, okay, good.
[00:22:41] Anyone else?
[00:22:44] 60? So maybe so just a little bit longer, yeah?
[00:22:50] Good. Yeah, I'm you can imagine I'm I'm keeping you from from from fighting all the commuting stuff, right? So you, you know, we can we can have this conversation. Well, let's um, let's see. This is what the histogram looks like now. What does this histogram starting to look like? We'll come back to that. The 85th percentile of this histogram is now 60 days, so whoever said 60 days gets a prize. Um, I don't have a prize on me, so. Maybe see me tomorrow. I don't know, I'll figure out some. The person sitting next to him, give him a prize, right? So this is random queuing, no expedites.
[00:23:23] All right, we have yet another experiment to run. So this time, we're going to go back to FIFO strict FIFO queuing, only now, now we're going to introduce the concept of expedites. Right? And there is always going to be at least one expedite on our board, always going to be, um, one and I should say one and only one expedite on our board. Um, and whenever an expedite shows up, that expedite is going to be allowed to violate work in progress limits. And whenever that expedite is in one of those columns and it's finished, we're always going to choose the expedite instead of the other ones. Okay? Does that make sense? So always always an expedite, so let me go back here. So here's an example, we can say that number one is an expedite. Um, when we have the situation here where we've we're trying to choose between several items, we're always going to pick that number one. We're always going to pick that expedite, sorry, we're always going to pick that expedite. What do you think this does to our our, uh, our cycle time now?
[00:24:26] What do what do you think the shape of that histogram's going to look like?
[00:24:32] Do you think it's, do you think our, do you think our cycle times, cycle time is going to get better or worse at the 85th percentile?
[00:24:39] Worse.
[00:24:40] Worse. Anybody think it's going to get better?
[00:24:43] No, nobody thinks it's going to get better.
[00:24:46] Um, so what do you think? So before it was so strict FIFO was 50, random was 60. What do you think this will be?
[00:24:54] 100? That's a good guess.
[00:24:58] 80? All all good guesses. This is what our our histogram looks like now. A little bit different. Um honestly honestly a little bit more predictable. We'll talk about why it's potentially a little bit more predictable now. Uh but now the 85th percentile is 65 days. 65 days, okay? Last case, you guys are probably guess what the last case is going to be. We're going to throw everything together. We've always got expedites and we're and we're just going to randomly pick stuff. Right? Whenever there's an expedite there, we'll always pick the expedite. But whenever there's not an expedite there, we're just going to randomly choose what to do. Right?
[00:25:37] What do you guys, oops, what do you guys think now? You saw, I mean, I can show you that, right? I mean, I guess I can show you that. What do you guys think now in terms of cycle time?
[00:25:48] Can you see?
[00:25:51] Oops, went the wrong way.
[00:25:54] 100 days. The 85th percentile is 100 days. It's double, it's double um essentially doubled that that first uh that first case. So if we if we look at all these kind of compared, right? So remember all we're doing is playing around with poll policies. That's all we're doing here. We're just playing around with with with poll policies and that's that's kind of the evolution of that um of that histogram. By the way, that bottom histogram, what does that kind of look like? What does that bottom histogram look like?
[00:26:25] Yeah, our you know, our remember the commute time histogram and our process time histogram. Kind of looks like that. Do you see that big hump on the on the left and kind of a long, long tail out to the right? That's really kind of what it looks like. Um, so if we if we look at kind of a very, very, very, very, very crude measure of variability, um, you can see starting with the last case, the max height of an um of any one of these bars, um, was seven items, only seven items, and the max width from the from the, um, fastest time to the longest time was 130 days. So what we had is a really, really kind of long histogram and a really, really so short, short, fat, long histogram.
[00:27:10] Um, right? FIFO queuing with uh with always one expedite, 40 days and 27 items. Right? FIFO queuing with no expedites, 20 days, 47 items. So you can see as we kind of work backward that histogram gets shorter and taller. When we talk about kind of driving more a little bit more predictability in our process, that's really kind of what we're looking at. You might think, well, you know, Dan, this is great for fantasy land, what about in the real world? Well, this is this is actually, this is a real example. Um, a team that I worked with, no work in progress limits and no pull policies whatsoever. Right? We came in and and we changed that so that we did, um, we all we did was implement some some decent pull policies and, uh, limited work in progress. And you can see kind of if I if I stack this on top of each other, kind of the difference that that we get. Um,
[00:28:05] I want to be, somebody remind me, I want to be clear about something maybe a little bit later, but now's not not the right time. Um, question, maybe I should ask, maybe I should pause. Questions about any of this? Does this does this make sense? I've gone really, really, really fast through it. But I hope it makes sense.
[00:28:20] Right?
[00:28:22] Most people think, most people think it's all of these other things that really are really affecting our cycle time. Um, but the bad things that happen to us are really things that we're doing to ourselves. For the most part, we're just not paying attention to the order in which we're pulling things through. For the most part, that's that's really what's happening. Or if we are if we are paying attention, um, I hope this doesn't sound wrong. But if we are paying attention, we're choosing incorrectly. Right? From a predictability perspective, I want to make sure from a predictability perspective. For the most part, we're choosing we're choosing the work that we work on incorrectly. Right? Um, I maybe I should say it now. By the way, I'm not advocating, um, a strict FIFO, this might sound like I'm I'm advocating a strict FIFO pull policy. I'm absolutely not doing that because in the real world, it's actually really, really hard to achieve. Um, but from a predictability perspective, what do you think should be our default pull policy? I'm not saying it should always be this. But when we have when we have to choose between two items, from a predictability perspective, what do you think is kind of the the best default algorithm? Or choice, what do you think?
[00:29:26] Max smoking.
[00:29:29] Maximum value with minimum effort, yes. Um, I've got shameless plug. Total shameless plug. I've got a a talk tomorrow that I'm going to do where I'm going to talk about the pitfalls of thinking about maximum value and minimum effort. Um, but so from a so I'm going to say no, from a predictability perspective, that's probably not our best choice.
[00:29:54] Just based on this, can anybody think?
[00:29:59] No, not FIFO.
[00:30:02] It's a different it's a slight it's a slight twist on it. Um, just just very, very basically, your default pull policy should be whichever one's been there the longest. Right? If all you're all you're doing is looking at the item age. And all of the things being equal, assuming we don't have any better information, which usually we don't, by the way, we think we do, but we don't. Um, all other things being equal, you should always choose the one that's just been there the longest. Right? From a predictability perspective, I want to be clear from a predictability. That's not strictly speaking, that's not FIFO queuing. A lot of people think that's FIFO queuing, but it's it's not necessarily. But anyway. When we last left Titanic, so remember I said,
[00:30:41] we should we should choose the one that's been there the longest, from a predictability perspective, we should choose the one that's been there the longest. So when we last left Titanic, we're in the situation, Titanic's been there a while, Olympic has just come back in. Right? Which one, from a predictability perspective, what do you think Harland and Wolf should choose? Finish Titanic. Which one do you think they chose?
[00:31:06] Worked on Olympic. Right? So they pulled people off of Titanic and they went and worked on Olympic. Now, interestingly enough, they pulled people off of Titanic to go work on Olympic, but Titanic still finished first. Right? So Titanic, remember I said, uh, um, Titanic had that, uh, maiden, our our our tentatively scheduled maiden voyage for March 20th. They pulled people off a Titanic to go work on Olympic, which meant that its maiden voyage got pushed up by about three weeks to April 10th, 1912.
[00:31:38] Now, a three-week delay, generally speaking, might not seem like that big a deal. Right? Um, until you start looking at a little something called iceberg activity in the North Atlantic. So this is a chart of iceberg activity in the North Atlantic. Especially in 1912. Can anybody tell me what you see here in terms of the difference between iceberg activity in March versus April? Which one's more active?
[00:32:06] April. Yes, I mean, it goes up, um, you know, icebergs in the Northern Atlantic go up significantly and especially in 1912, for whatever reason in 1912, there were a ton of icebergs out there, right? Um, furthermore, because there was a delay to Olympic, and remember I said Titanic got finished first, now the captain of Titanic had his pick of crew. He had he had officers slated for Titanic, but he also had the Olympic crew sitting on the sidelines here. And remember, the Olympic had been out sailing around, so these people had um had a lot of experience sailing these these big ships. So what do you think the captain did? He's got he's got his pick of anybody he can he can he can take. What do you think he did?
[00:32:49] He yeah, he picked up some of the Olympic crew. So specifically what he did was for Titanic, he took the chief officer from the Olympic. Right? So follow that. Hopefully this is hopefully this makes sense. So for the new crew for the Titanic, he took the chief officer from the Olympic, which meant the chief officer of Titanic now became the first officer. Right? And the old first officer of Titanic now became the second officer. And what do you think they told the told the second officer at this point, the old second officer for Titanic?
[00:33:23] Bye.
[00:33:25] It's been nice knowing you. Thank you very much. Um, see you later. Right? They they asked him to go home. So the old the old second officer of Titanic goes home. Again, not that big a deal until you realize that one of the jobs of the second officer is to keep the keys to the lockers of where they hold the binoculars for the lookouts. Now, that second officer that went home, took the keys with him. He didn't he didn't mean to, he didn't realize that he did. But he took the keys to where the lockers where all the binoculars were, took those keys home with him. So when Titanic's out sailing around, they've got these watches, you remember these watches in the uh in the movie, right? What don't you see these guys using?
[00:34:12] Binoculars. They they didn't have their binoculars because they couldn't get to them. I mean, I suppose if they really wanted to, they probably could have busted open the lockers, but they didn't. Um, these guys didn't have didn't have those. So, um, that delay, that one delay, and I know this is kind of an extreme case, but that one delay from March until April, increased their chances significantly of hitting an iceberg. And it got worse because that crew shakeup, um, that delay caused that crew shakeup causing some of these other things to happen. So, to sum up, uh, most of the unpredictability in your process, whether you know this or not, most of the unpredictability in your process is down to poor pull policies. I know when you're in your in your world, you've probably been told a lot that, hey, we should implement something like an expedite process because an expedite process will make our our process more predictable. And the truth is that if anything, it makes it much, much, much less predictable. For all kinds of reasons even that we haven't even talked about here. So don't believe the hype around expedites, make your process more predictable, generally speaking, they make them less predictable unless certain things are done. Uh and in general, in general, it's better to start work that's already, um, it's better to finish work that's already been started before starting any new work. Um, and so to ask yourself, does the work that you want to start, uh, is it, does it really need to be expedited? Right? That's probably the bigger question. Generally speaking, if you have this item that comes in that that somebody thinks needs to be expedited, the best strategy is to finish something that's already in progress so that we're free up capacity to start this new thing. In general, right?
[00:35:49] Uh, we we saw that. Okay, I want to be clear, I want to be clear. Here's not what here's what I'm not saying, right? Because we live in the real world, I don't want to I'm not here saying that expedites don't exist, right? Of course they exist. Right? If your if your company is based on taking credit card payments off of a website, and that payment processing thing goes down, that's probably a pretty good definition of an expedite, right? But the fact that maybe your CEO is golfing buddies with this other guy and had this great idea about this new feature we should implement because they're out on the golf course talking about it. That's not an expedite. Right? So, um, chances are that um all the things that you think are expedites really aren't, and you'd be much better served by not expediting. So one last thing, one last thing about about expedites and this whole notion of value, uh, because I find this I I find this fascinating. I don't know if anybody remembers these guys from the movie. These are the radio operators. Right? Titanic was one of the very first ships that was um fitted with ship to shore radio. Right? Now, what's interesting about these these radio operators right here is strictly speaking, they were not white star employees. Right? White star did not pay them. Can anybody guess how these guys got paid? If they're not paid by White Star.
[00:37:15] Per per message, yeah. So their job was, um, so this is kind of think of this as like the early days of texting, right? Wealthy passengers would come to them and say, hey, can you send this message for me? Can you send this message for me? Can you send this message for me? And they got paid per message. Right? That's how they got paid. Now, unfortunately, one of their other jobs is to relay iceberg warnings from other ships to the captain. Now, think about this from a value perspective. You're not a White Star employee, you're not paid for relaying iceberg warnings to the captain. You're paid by sending messages from the wealthy passengers. Oh, I've got pictures of all this. You're paid, um, to send, well, um, to send messages for the wealthy passengers. Which work are you prioritizing right now from a value perspective? Which work are you prioritizing?
[00:38:07] The messages, right?
[00:38:10] The last iceberg warning that was placed in the, um, in the captain's hands on the day that Titanic sunk, was I think at 5 o'clock at night. Titanic sunk at 11, I want to say 11:30, something like, you know, several, several hours later. There were hours and hours and hours, multiple, multiple iceberg warnings that never got put in the hands of the of the captain because our our radio operators were were too focused on on value. Right? So,
[00:38:41] um, a couple of uh, couple of books written by the world's best author on these topics. Um, actually Magical Metrics for Predictability. Uh, this is my contact information. Feel free to um, feel free to shout at me on Twitter. Most everybody does. I'm usually just complaining about tennis or something like that. So, um, questions.
[00:39:06] Questions about anything? I don't even know, how are we doing on time?
[00:39:09] Good time.
[00:39:10] Do I have time?
[00:39:10] More than 10 minutes.
[00:39:11] Yes. See, I promised I'd get you guys, we started late and I promised I'd get you guys out early. Any any questions about any of this? I know I kind of went over over it very, very, very quickly, but uh. What do you think? You surprised, you're not surprised? How many how many people have like an expedite process built into how they work right now?
[00:39:30] About half, looks like about maybe a quarter to to half the the audience. Do you think it's working for you?
[00:39:36] Of course.
[00:39:39] You know, um, another thing that's interesting about our fantasy our fantasy land situation is, remember we had those expedites? What do you think the the cycle time of those expedites was? So remember, they're skipping queuing columns, um, and they're they're always violating work in progress limits, and they still take 10 days. What do you think the expedite of those ones were?
[00:40:02] If you had to think.
[00:40:04] Yeah, somewhere somewhere closer to 30 or 40, right? All the expedites were probably getting done in 30, 40, 50 days. But remember in that last example, everything was taking 100 days? Your expedite takes 30, 40, 50 days, everything else takes closer to 100 days. Now your product owner, I know if I hand you an expedite, you get it done in about 30 days. If I hand you a regular item, it could be as much as 100 days. What am I going to do now?
[00:40:29] Everything becomes an expedite. Right? What else? Questions?
[00:40:36] I've stunned you into silence, you're like, oh God, please nobody ask a question because my train is leaving and if there is a train leaving. Is the no, is the number four running? Are we are we going to have an opportunity to get to the number four? Or is that? Oh, question, please, yes. We we could talk about that.
[00:40:52] you're telling me that in your simulation and the right item then stay the same.
[00:40:58] Yes.
[00:40:59] So, how does it behave in the real world where you can have something that takes five days? 10 days.
[00:41:07] Yeah. Great question, so the so I'm I'm glad you brought this up because I meant to talk about this. In this simulation, the question was in this simulation, everything took exactly 10 days of active time. Everything, but of course, we know that that's not the real world at all. So what happens when we have variability of the items themselves, right? Some some are taking five days, some take 20 days, some take whatever. How do you think that would play into this? Do you think that would make things better or worse?
[00:41:33] Worse.
[00:41:33] Yeah, it makes it makes it much worse. This is the absolute best case scenario. So if we add in the variability of our items as well, it's going to make things even worse. This is why you have to be even more diligent in terms of making sure you understand those those pull policies because not only do we have to deal with the variability of the items themselves, but now we've got the added variability of we're making poor decisions. So, thanks for bringing that up, that that that's a great question.
[00:42:02] Anything else?
[00:42:05] All right. I hope everybody gets home safely. Um, I'm not I haven't decided if I'm going to walk yet. I don't know, it's about an hour and a half walk, but I'm I'm thinking I might. Is it raining? Does anybody know? Is it raining? Yeah, we'll see. Um, there there's that commute time. Thank you very much to all of you. and good evening Thank you.

Transcript