This interview was provided courtesy of the Computer History Museum.
The SEI 5-Level Maturity Model
Humphrey: So now, back at my
SEI job, talking with the guys at MITRE, we now had a
list of 100 things that people ought to do if they were doing good work. I was
sitting in the airport in
Booch: A revelation.
Humphrey: So, a five-level model and the whole purpose of it was really to try to judge how are people doing. Are they really doing the right kind of stuff? We wanted to keep it kind of general. We didn’t want to get into how they do any of this stuff. We didn’t want the acquisition people telling people how to plan and how to do anything else like that. So, we wanted it to be fairly abstract. It wasn’t what I would call a measurement of the organization. It was an indicator of how an organization is doing and whether they were following the right kind of practice. It was basically a motivational framework. By late in 1986 we had the first pass at this thing. About that time, Larry Druffel got a call. Larry at this point had become the director of the SEI. They finished the search. They really didn’t like anybody who had applied and the search committee concluded Larry was probably the best choice. So, they picked him and he accepted.s
Booch: And, he came from Rational. He
had been working with us there for a while. The other small world story is
Larry was actually one of my teachers at the
Humphrey: Oh, is that right?
Booch: Yes, small world.
Humphrey: He was also at DARPA for a while I believe.
Booch: Yes, that’s correct, and I got
involved with him because of the Ada Program Office. So,
he went from the
Humphrey: Yes. Well, he had
been in a lot of good places and did exactly the right kind of work. In any
event, so we put this together, but Larry got a call from some general in
Washington -- in the Pentagon -- who wanted the SEI to go evaluate what was
going on at the
I strongly urged that the general not beat them up but use the results of the assessment as a guide for improving. And so he went back and talked to the general and the general bought it. So, I went to down to Gunter, talked to the colonel, told him what was up, and he told me; he said, “Look, we have all these studies.” He said, “My closet is full of them. The last study we had made something like 50-60 recommendations. To put them in place would have cost $13 million. We didn’t have that kind of money. There’s no way in the world we could do it.” He said, “How is this going to do any better?” I said, “I don’t know, but we’re glad to go through it and you can take a look at it. My guess is you’ll come up with something where you can change what you’re doing without adding a whole lot of money, but I don’t know.”
So, he bought it and we went down and did an assessment and it was great. We identified a lot of things that ought to get done. They put in place some action plans. It turned out that they put this in place, and he actually went back and reported to the general what they were doing and what their action plans were and all of that sort of thing. There hadn’t been a commander out of that lab that had been promoted in history. They’ve never been promoted. They’d always failed. This commander was a colonel and he was made a one star [General] and moved on. Unfortunately, that was a disaster because when he was gone, all of a sudden the new guy wasn’t interested. He had no skin in it.
And so they went through like six or seven commanders in the next five or six years and they never did anything on the action plan. They finally got a lady in who took a look at all this stuff and finally said, “Okay. Quit screwing around -- do it.” So, they actually did it later, but it took them a long time. But in any event, the structure we had worked. What we put together at the SEI actually did work because now we could connect actions to our evaluations. So, when people got better or worse, we could tell them why. All of a sudden, we had something that allowed and motivated improvement and the CMM [Capability Maturity Model] took off from that. That’s where it went.
Booch: Wow. It’s interesting to hear the beginnings. I had never heard the beginnings.
Humphrey: Yes. It was very effective. It was quite amazing. Now again, I really want to emphasize (and people lose this; they just don’t get it) that the CMM is not for the purpose of measuring an organization. It’s not trying to characterize it in any detail in terms of a model of it. Refining it to get more and more precise is a mistake. You’re not actually doing that at all. The whole idea was to motivate people to think about how they’re working and how to improve it, and you want to keep it simple. It has gotten way away from that. I’ll come back to that later.
So, that’s what we did with the CMM. During some of the early CMM development, we put that together. I wrote my book Managing the Software Process. It sort of describes what we were doing. It was sort of the bible for it, at least for a while. I remember going out and giving a talk. Remember we put out this little technical report I mentioned? An SEI technical report. This was like in 1987, I think, or early 1988. It sort of described the method and the levels and all that sort of thing. It was before we actually had published CMM. I remember going out and giving a talk on it at AIAA -- is that one of the associations?
Booch: The American Institute for Advancement of - what do they stand for? I’ll have to look those up.
Humphrey: Aeronautics and Astronautics -- something like that. But I talked to a number of these conferences. I have it - Aeronautics and Astronautics.
Booch: Ah, that’s the one, yes.
Humphrey: I think I talked to one of their meetings, and I took out a couple of dozen copies of the report, and I started my talk and I said, “Oh, by the way. I have some copies of the report here if anybody wants it.” It was a stampede. All of a sudden, people rushed up to get the report and they were all gone in about three minutes because everybody wanted it. Of course, there were about 200 people in the meeting. So, it didn’t work. But in any event, it was extremely popular. People were very interested in it and it took off. As you know, that’s sort of what happened with the CMM and then the CMMI. So, that’s what we did there. As we got going with that, we had to put together a group to structure it, to put it together in a much more formal kind of a framework, which they did with the CMM.
CMM Steering Committee
One of the problems I had was as we were going through it, the guys were putting together a model for what was the CMM. They were making it much more formal than the sort of maturity model framework that I had put together myself. That was fine. I wanted them to do that. And so, we called in a whole lot of people from the Defense Industry who were very interested in this. A lot of people were interested. So, they all came in to a big meeting we had at the SEI, and our guys went through a presentation about what we were doing with the initial version of the CMM. We damn near had a riot. There were a whole bunch of people who just objected violently to what we were doing. “You didn’t talk to us about it. How come, etc., etc.” So, we decided we would co-opt the people that were most vociferous and we did. We started inviting them. We put together a steering committee. We got the people who were really the most outspoken and most concerned and put them on the steering committee.
So, we brought them
in to work with us on making sure it did really meet their needs. There were a
whole bunch of meetings for that. I had participated in some of the initial
meetings and I realized that I couldn’t sit in on the meetings. It wasn’t that
I wasn’t interested. I was dying to sit in on the meetings, but the problem was
I tend to think out loud, like a lot of us do and said, “Well, how about this”
to bounce it off people. Well, the minute I would say, “How about this,”
everybody would say, “Okay.” It was sort of I had spoken: It was engraved in
stone. That was it.
So, I was no longer involved in designing the thing. I would review it later. I had lots of questions and stuff like that, and I’d poke at it. My own guys, I could work with them, but not with the whole community. As we went through all of this stuff, I began to realize fairly early on as we started doing the CMM (the CMM in particular -- it wasn’t the PSP yet), I was concerned because this was really kind of abstract at a fairly high level. People were talking about how we could use this in a very small organization. No one knew how to answer that. We didn’t know. That particular question hit me. And so, I started wondering about that. How would you really apply the CMM to a very small organization?