This is a transcript of episode 317 of the Troubleshooting Agile podcast with Jeffrey Fredrick and Douglas Squirrel.
How can your organisation be simultaneously fast and collaborative? Foster willingness to fail successfully and collaboratively, while applying strong leadership and accountability.
Show links:
Listen to the episode on SoundCloud or Apple Podcasts.
Teal Organizations, from Reinventing Organizations
Listen to this section at 00:14
Squirrel: Welcome back to Troubleshooting Agile. Hi there, Jeffrey.
Jeffrey: Hi, Squirrel.
Squirrel: So we had a wonderful question, not from a listener this time. I mean, he may listen actually, but he’s one of my clients. And we were discussing a approach that he sees in the data science team at the place he works at, where I’m coaching him. And he said, ‘this approach can never work! These folks are just not ever going to get anywhere. This is not how good machine learning teams operate.’ And I tried to tell them about some examples I knew of teams who operated in this this way, the sort of teal organization approach that were successful.
Squirrel: And he said, ‘well, those aren’t those aren’t data science ones, Squirrel. I don’t really count those because it may work in less complex environments. But here where we are, we just can’t take that kind of academic approach. We can’t take the communitarian, ask everyone their opinion, stop for lots of checks and and making sure people are brought along on the journey. We need to make some tough decisions, get things rolled out. Because look at how fast the environment, the whole industry is moving. You know, it seems like a new AI breakthrough every couple of days. And Squirrel, we got to move faster here.’
Squirrel: And I said, I think I know somebody who has led data science teams in this way. And and it happens that I’m going to be recording a podcast with him tomorrow. So here we are. And I thought that we might just talk more- we talk a lot about this approach to building software, but I thought we might kind of focus in on data science because you, Jeffrey, have led I know two data science teams that have this more inclusive approach that are less top down and more bottom up. So I thought maybe I could ask you some questions about that, and that might help out my client and lots of our listeners, too. What do you think?
Jeffrey: That sounds great. I’d love to share my experience here, although I just will say I was leading the the data science in the sense of it was part of my remit as a CTO or VP of Engineering to include data science as well as engineering. So it was a bit broader, but definitely covers the whole range of product development, including the R&D, machine learning, data science part, as well as the then the the transmission from that into delivery, which I think is what your client is very concerned about.
Squirrel: Like almost every organization today, there’s a data science component, and then there’s an engineering, traditional engineering, non-data science component. So he’s interested in both. So I think this will meet what he’s looking for. So Jeffrey, can you just tell us a little bit about what one or both of these organizations and maybe also then tell us what was different about them? What’s this teal organization idea? How does it work? What do you what do you see when you walk in that’s different from other engineering or data science organizations?
Jeffrey: Sure. Yeah, I’ll do that. So I’ll also give a bit of context. So the kind of time span we’re covering here covers about 12 years and two very different teams and different domains. We’ll start a bit with teal. And, uh, one thing I’ll say is that what we’re going to talk about here, I think, is a lot about having high performing teams. And it’s important to do to ground it in that, like the goal of teal organizations is actually about performance and the claim and this comes from a book, Reinventing Organizations by Patrick Leroux. I think I have that name right- No, sorry, Frederic Laloux. I had it wrong, but I had the important part. His surname. Correct. So, uh, Reinventing Organizations.
Squirrel: It’ll be in the show notes for listeners who want to check it out. Go ahead.
Jeffrey: He put out this framework about the evolution of organizational culture, and he basically said that there’s been a number of breakthroughs that allow organizations to be much more effective than what came before. And he posited this evolution from red to amber to green to finally teal.
Jeffrey: There’s a level in here I’m missing. [Transcription Note: the other level is orange, between amber and green.] But the idea is that each of these has some sort of breakthrough and it goes from the original part is like, the organization as a wolf pack, and it’s all about the strongest person is the leader. And then you get to something that’s more about being more hierarchical and organizing, and that the now the analogy is something like the military, where there’s a chain of command.
Jeffrey: Then you get to bureaucracy, you have something more like the Catholic church. Then you get analogy of organization like a machine, and this would be the big industrial companies in the 20th century, where everyone has a place and they’re supposed to be a part in a machine. And then you get to green, which is the analogy of like, we’re one big family and kind of a paternalistic sort of view.
Jeffrey: And then teal is about what he sees as the next step, where everyone is showing up as an individual. So the difference here is it’s less of the paternalism. You’re definitely not a machine. You’re not interchangeable parts. Everyone shows up, and it’s really the idea is that you can get the most out of people individually and therefore the most collectively when you have certain attributes that allow people greater freedom of action to take more responsibility while also being accountable. And they can do this, uh, and they don’t- What really is big in teal is that people understand the mission and what you’re trying to accomplish. And then they can organize themselves to get that done.
Jeffrey: They don’t need as much management top down because they’re intelligent, engaged problem solvers. And if you ever hear about, you know, read Harvard Business Review, you’ll see the word engagement all the time. ‘Our people engage-‘ ‘how to increase engagement-‘ ‘how to know when people aren’t engaged-‘ And teal really optimizes engagement and therefore optimizes the contribution of everyone. So that’s kind of what it is.
What’s Different
Listen to this section at 06:20
Jeffrey: Now you asked, actually, what it what does it look like? And maybe the main thing that’ll will look different from what our listeners are used to is it looks like a high autonomy environment where a lot more people are bringing ideas forward. There’s a lot more questioning of, like, ‘do we understand why we’re trying to do this?’ Really understanding the story and the direction.
Jeffrey: So on an engineering side, something that someone reminded me of yesterday, there’s a certain engineer, who’s really well known for when the product people come with a design, we’ll say, like, ‘well, how do you know this is the right thing? What have you done to understand the client?’ In other words, help me understand the whole story of how you got here. Don’t just give me the the job to do, but help me understand the story, so that I can then go into a better job applying my own judgment so that and dialog-
Squirrel: And possibly, I can contribute to your design and tell you that part of it isn’t feasible or part of it, we tried last week and it doesn’t work.
Jeffrey: Yeah, or ‘here’s something even better, given what you’ve told me. Have you thought about this? You might not have realized this was possible! Let me take a day to give you a prototype?’ And maybe that changes the conversation, and it often does. And so that kind of innovation can come from anywhere, kind of thing, I think is one of the things I’d say is a real characteristic that you would notice as different if you just got dropped into an organization. Or have you ever heard of the miracle question, if the miracle happened last night while you were sleeping and then you woke up the next day, but you didn’t know the miracle had happened because you were asleep. What’s the thing you would notice? So if you were to think of your company, if you suddenly were teal, what would you notice the next day? You suddenly notice a lot more engagement, a lot more ideas, people with questions, a lot more offering of ideas and solutions. That’s that’s what you would see.
Squirrel: I got it. Okay. So now one of the specific concerns that my client has is about a phenomenon that I have seen over and over again, and I think he’s conflating this with the the teal organization model. But I want to check with you whether this is something that you saw, whether it was something that you saw as a problem, or maybe you saw it as a positive, I’m not sure. But a tendency among heavily research oriented organizations, people who are building really new stuff. I’ve not been to SpaceX, but I kind of feel like those folks, must be operating this way.
Squirrel: There’s an awful lot of paper writing, of polishing of things that you’re working on, making sure that you’ve covered off all the angles, that you’ve shaved off all the possible ways that you could be wrong, and a tendency not to experiment. And SpaceX clearly doesn’t do that because they blew up a spaceship recently. That was that was an example of trying something that definitely had not had all the bugs worked out of it, but I suspect they’re fighting this because they have literally rocket scientists there who come from an academic environment.
Squirrel: And that is one of the characteristics that I see often in data science teams. This sort of ‘we have all the time in the world, we need to make sure that we cover all the bases so that our paper isn’t rejected.’ And you have somebody like my client who says, ‘My God! Could I just get this in the market because somebody is about to release their version tomorrow? Could I get this out?’ But I don’t think that’s what you’re describing as the characteristic of a teal organization. When you said it was outcome focused, I think that was really important to underline. So what’s the difference there? And have you seen this in the data science organizations you’ve led?
Jeffrey: Yeah, I think it’s a really good point. And I’ll also say, you know, the Head of Machine Learning who reports to me is a podcast listener. So he’ll be double checking what I say. I think there’s an element of truth in what you’re describing, but it’s not in a sense congenital. It’s sort of environmental and can and can be offset.
Jeffrey: So here’s what happens, in my experience. A lot of times people in the data science background have come from an academic background and sort of they and they’re used to the kind of things that you’re describing, where they’re producing papers and you want to get all your ducks in a row. And you said, use the phrase, ‘well, they don’t want to experiment,’ but then you use SpaceX. And the difference here is experiment in production versus, you know, experimenting in the lab. And I think they’re quite happy to run experiments, kind of as they are working through their own models. But and they’re more wary about bringing something of like, ‘hey, let’s try this in production.’ Um, and I-
Squirrel: Kind of like how SpaceX would try something in a wind tunnel rather than actually launching a rocket. So there must be some pressure there, I’m sure Elon Musk is running around beating people over the head, but, what do you do in a teal organization if you have this tendency, this academic caution?
No Silos and Aligned on Outcome
Listen to this section at 11:21
Jeffrey: Yeah, well, the main thing you do is, and this is a very relevant to the teal part, which is often there’s not strong silos between organizations. The different disciplines have their own expertise, which they’re assumed to be experts in. And you ensure they’re experts in. But there’s no desire actually for people to live within their silo because they’re aligned to the actual outcome. The reason I mentioned the Head of Machine Learning is, he and I were discussing yesterday the question of elephant carpaccio, effectively. Do we have the right thin slices to move forward? That we don’t need to have all of the questions answered in advance. We need to have enough data and take forward conclusions that are tentative, when they’re ready, without waiting to have everything be perfect. You might say it doesn’t need to be publication ready to be useful. And I think that comes down to-
Squirrel: Useful and informative. The experiment might be successful with a negative result. The spaceship exploded, but we learned why and we know how to not do it again.
Jeffrey: Yes! That’s the point. And that’s the thing SpaceX is very good about this, which is the alignment. What they’ve got alignment on is that, there are some things that you can only learn from trying to launch. So they will certainly try to learn all they can prior to launch. They will run all the experiments they can, but they also know there’s a limit to what you can do before you go test in the real world. And there’s alignment in that. And part of what I would guess, and I’ll say I’ll move away from SpaceX and more to the domains I’m familiar with.
Jeffrey: There’s alignment within the data science and the engineers and product that what we’re trying to do is maximize our rate of learning and sometimes that means doing it experiments, in a data science environment in, in a different language, or different toolkit than we use in production. So maybe they’re doing something in R or Python and it’s eventually going to be translated into Java code. But at a certain stage it’s faster early on to iterate that way. And then as we learn more, we start moving to more and more production like and this goes back again to the teal part. There’s joint design and collaboration on ‘what are our options that we have’ and ‘what are the trade offs between them,’ and that you want experts involved in that discussion.
Jeffrey: So it’s not managers. It’s not me as a leader saying, ‘well, here’s the options I could think of and here’s the trade offs I see.’ I will certainly offer that because I’m an individual within the teal organization like anyone else, but I am very interested to get the input of the experts, and I stress this: ‘you are the experts; you tell me what you see as being the options, and you tell me what you see as being the trade offs.’ And because there’s alignment on the outcome and they understand the outcome is about what we actually produce and ship in the real world, then there’s alignment in the approach.
Jeffrey: As opposed to having a more siloed organization where, and this is this is I think the danger is, if you have different goals in the different groups and different measures of success, and that somehow it’s seen as successful for the data science team, like, ‘well, we created our model. There you go. We have our model. We’re done. Now it’s up to you to go out and engineer, to go make something of it.’ When they don’t see client value as the ultimate measure of success, then I think it’s problematic. And you are now starting to have that break down, that sort of unity of purpose, which I think is so important, to get the right dynamics.
Squirrel: So I imagine, I don’t know if he’ll even listen to this, but I imagine my client will be nodding along all the way to this point. And he would then say something like, ‘so that means what you need to do is break up that kind of academic, slow, super collaborative approach. Forget this teal thing. What we need is to get them aligned to delivery, and we need to get them on the treadmill here on the, on the conveyor belt, cranking out lots of new, exciting, wonderful stuff. So this sounds wonderful because what we’re going to do is cut out all of this crazy collaborative, exchange of ideas, reading papers. We’re going to stop doing all that kind of theoretical stuff, and we’re going to be very practical.’ Somehow, I have the feeling since you’re laughing, that might not be what you what you did or what you have in mind. Can you can you help us out?
Expert Judgement and Accountability
Listen to this section at 15:54
Jeffrey: No, that’s completely not the case. The point is, you’re dealing with experts here, and they have their expert judgment on what’s the fastest way to proceed. And sometimes that’s a literature review. Why would you want to go try to reinvent something that someone else has already published on? And you don’t want to just have the literature frozen at a snapshot at a time, the literature up to the point they graduated. And especially in a situation where data science is moving forward so quickly. You’d want to know what’s been published last week, and is that relevant to what we’re doing today? Maybe that changes our approach. And so we run a different approach today than we would have without that paper. So there definitely is an element here of value in keeping up with that. And there is a balance, but the question is who’s the people who can best understand that balance. And that’s going to be the experts.
Jeffrey: Now, to be clear, there is an issue here of accountability. And remember, I mean our definition of accountability, which is an obligation to render an account. People here, even if they’re experts, are not going to be perfect at this, and they need a chance to learn from their own experience. And part of that is ‘so, in this last time period, what have you done? Let’s go ahead. You told us what you intended to do. Then you did some stuff. Now let’s reflect on it. Given what you did, do you think that you made the right choices, given what you knew at the time.’
Jeffrey: And sometimes will happen is people like, ‘you know what, actually we should have done X instead.’ And X might have been ‘we should have spent more time looking at the paper, you know, at the papers.’ Or it could be, you know, ‘we should have tried to get this into engineering’s hands faster. We should have collaborated with people in engineering sooner. We should have tried this with our production data sooner.’ It can be different learning, but that learning is how people develop the intuition about what’s the right trade off. And I’ve never seen anyone who has it right at birth, it’s something that only comes with experience.
Squirrel: And that’s the puzzle for this client and many of my clients who are less technical, who are not experts on machine learning. I’ll put my hand up and say, I’m such a person as well. The difficulty is that it’s hard to tell the difference between ‘I’m polishing and I’m making sure that I cover off all the angles and that nothing is going to explode, nothing is going to break. I’m not going to give anybody the wrong answers from this AI model I’m creating,’ and ‘I’m checking their the literature to be sure that I haven’t missed a trick that would let me get to production faster.’ Those look from the outside the same. So I wonder how did you tell the difference? How did you get the accountability in place so that you could tell and stop it when someone was off on a perfection crusade and keep it going when someone was getting a lot of really great new ideas from other members of the team and the literature and so on.
Jeffrey: Well, people who know us and know our work and know our book, it shouldn’t be surprising. The answer is agile conversations. You know, it’s-
Squirrel: Whoa! Shocking! Okay, go on.
Jeffrey: I also don’t have a background in machine learning. I’m not a data scientist. I have a background in physics and nuclear chemistry. I have knowledge of statistics. I’ve worked with enough people to have picked up some things, but I’m definitely not an expert. But what I can do is have a genuine conversation where I can share my concerns and say things like, ‘hey, I’m just I’m nervous about the pace that we’re going. Do you think that there’s other ways that might be more effective? What are the trade offs? What are the options you’ve considered?’.
Jeffrey: I can I can have a dialog with this. And sometimes I can advocate for what I want, like a recent example where I’m like, ‘I think you have findings to the point where I think there’s now a value and I want to go deliver them. What I’d like you to do is create a draft presentation with your current tentative results, knowing that it’s incomplete. And then we can review that to see if there’s value.’ And so I can advocate for what I want, and then we can go ahead and and do that together. So it’s being in there. And and a lot of it is sharing the concerns-
Collaborative Discussion and Then a Clear Decision Process
Listen to this section at 20:30
Squirrel: And also possibly not together. Just I want to check Jeffrey because I know you and you, you always sound so nice on our podcast, I know that you can be tough as well. So I could imagine you reviewing that presentation and saying, we’re going live with this and the other person’s, ‘oh, wait, no, we haven’t covered this, this error condition, this edge case, this isn’t working!’ And I can imagine you then saying, ‘but we’re going live anyway.’ And I just want to check is, is that-.
Jeffrey: Yes.
Squirrel: Would that be consistent with the teal organization, and the types of things that you’ve done?
Jeffrey: Yeah, absolutely. And I think it’s because there is a question here. And I think we get into, I think an aligned problem here, which is decision making process. Right? I think a lot of people who are uncomfortable with human dynamics and interactions, they look at people talking and they’re worried it’s going to be talking forever. And there’s a difference between we’re going to have a collaborative discussion and then a clear decision process. And I think there can be cases, and there are definitely cultures where people have conversations, but they have no clear decision process. And therefore things go on forever, because they have no way to to reach closure.
Jeffrey: In our environment, we definitely have ways where we can reach closure. And ultimately there is still a hierarchy and it can still be ‘look, I appreciate your concerns, but I can make this decision that we’re going to move ahead with this.’ But that only happens after I’ve heard concerns. It’s not something that’s a priority because I know there’s a chance I could be wrong. People, these experts, know more than I do, so I’m definitely going to hear from them any concerns they have about moving ahead, but ultimately I’m not going to abdicate my role in the decision process. I don’t want them to abdicate their role either.
Jeffrey: There’s a whole chain of responsibility here that expecting people to do their part, including disagreeing. But there is that element of disagree and commit. At some point we’ve talked it, we’ve heard all sides. We’re going to make a decision and move forward. And this is a bit, remember our last week, one way and two way doors, especially in the case of a two way door, where it’s something that’s temporary and reversible and we can learn from it, we’re going to spend a lot less time, before we move ahead compared to something that we think is a one way door would be hard to come back from. And of course, we’re going to spend as much time as we can to create two way doors rather than accepting one way doors.
Squirrel: There you go. Let’s close with just one more thing that you mentioned that I want to pull out. You said there were a number of mechanisms for kind of forcing a decision, making sure the team doesn’t get stuck, doesn’t wind up in lengthy collaboration that never reaches a decision. Could you say a bit about what those are?
Jeffrey: Sure. And I think that one of the fundamental issues is about my availability. This may seem strange, but trying to make sure that I’m regularly available so that people aren’t waiting when there’s uncertainty, they know that they can get rapid feedback on choosing between alternatives. So that actually ends up being a really important point.
Jeffrey: A lot of times when there’s a disagreement among a team, it’s because they’re not sure about, given the sets of trade offs that they see, which trade off would I or would the company prefer? And I’ll say there’s often cases where I don’t actually have the answer for what the company as a whole would say, but I’m willing to say, ‘well, I’m not sure. Here’s my understanding. I believe it’s this. Let’s go ahead and move forward on that basis,’ and we might correct it later. We might end up changing direction, but helping people to get comfortable with the fact that we’re going to make an attempt using our judgment, the best we can, and also knowing that they’re empowered to do the same. So yes, you might be in a discussion. You’re uncertain as a group. Come up what you think is the best. And then if you can’t resolve it amongst yourselves and you need to escalate it, then I’m going to be available to help that move ahead quickly.
Squirrel: There we go. So something I heard strongly from you, which I think might be a good final point to make to listeners who might want to apply this is that, this sort of highly collaborative culture that you’ve created in these two examples is absolutely consonant with, and in fact dependent on, strong leadership. So we’re dependent on when Jeffrey’s around to look at things. So that that sets our cadence. Jeffrey might make a judgment about what the rest of the company wants, and we’re going to follow Jeffrey’s approach to that. We’ve anointed Jeffrey as our decision maker. We’re going to disagree and commit. It sounds like there’s there’s strong leadership that goes along with the collaborative idea soliciting approach, right?
Jeffrey: Yeah. Strong leadership to set the vision, not to try to manage the details. And that’s really important and to know and also that goes with this, the safety, if they have made their best judgment to achieve that vision that we’ve discussed, even if I disagree with it later on, their choice, they’re not going to be punished for that as long as they are using their judgment, they show evidence of learning, evidence of progress. It’s about progress, not perfection. So I think that it’s very difficult for people to take ownership and move ahead if they feel they’re going to be punished for getting the wrong answer. People have an environment where they’re encouraged to use their initiative to move ahead and make progress. And without being perfectly aligned with me on the steps, as long as they’re very clear on what the objective is.
Squirrel: There we go. Okay, so I think we’ve given our listeners plenty of food for thought. The idea of strong leadership in a teal organization may seem contradictory, but I hope that listeners find it at least provocative, if not inspiring. I’m really appreciative that Jeffrey can help us out with his expertise and knowledge. I know that he’s inspired me with a lot of these more collaborative ideas that I try to implement with my clients. So I appreciate that Jeffrey can do that for us. Come back again next week when we’ll have another edition of Troubleshooting Agile. Thanks, Jeffrey.
Jeffrey: Thanks, Squirrel.