Using Reference Classes to Keep Projects on Track

I recently had an eye opening experience during a planning bootcamp we are running in Guadalajara with our team of 35. We had all read the book How Big Things Get Done as pre-work, and it introduced us to the concept of reference classes. This idea has been bouncing around in my mind ever since, and I realized we are not doing nearly enough to leverage it in construction planning.

During the bootcamp, Kate challenged me on this topic, and I am grateful she did. She asked if we had really done enough research for the reference class of our upcoming 58 million dollar project with Lean Build, our first major project under this approach. My instinctive answer was yes, we had looked at similar projects, gathered experience, and thought through a macro plan, a norm, and a backup. But Kate pressed further. Were these projects recent enough? Did they involve Elevate’s influence or were they before our team had refined our systems? Were we accounting for the fact that our trade partners in Phoenix were new to working with us?

Her questions revealed something uncomfortable but true. I had not dug deeply enough into reference classes. I had taken the surface level comfort of having some data but had not critically analyzed whether it was robust enough to forecast our outcomes with confidence.

So what exactly is a reference class? In simple terms, it is a collection of similar past projects used to understand the likely outcomes for the project you are about to start. By studying how much those past projects overran on budget or schedule, you create a realistic dataset for forecasting. Instead of relying on optimism or assuming our project is unique, we ground our promises in evidence.

This matters because as much as we would like to think otherwise, our projects are not special. Nothing in construction is truly unique. Nearly everything we build has been done before, somewhere, in some form. And we almost always have data if we are willing to look for it. The real danger comes when we fall into the trap of wishful thinking, believing our project will magically outperform the average without a proven reason.

Ideally, a reference class includes dozens of similar projects. Fifty is a dream number, but even having 25 projects from your local region provides powerful forecasting insight. Imagine the advantage of walking into a hospital building knowing that historically these projects overrun by 12 percent on schedule and 8 percent on cost. You would then build those expectations into your promise and use your planning tools to actively reduce those risks.

At Elevate, we have already built our systems to accommodate this mindset. Our macro level tact plan serves as the promise, the slowest reasonable speed aligned with contractual milestones. This should be tied directly to the reference class so the owner gets a promise rooted in reality. Then we have a norm level plan, our target, which aims higher by removing risks, increasing flow, and applying lean techniques. Same plan, different speeds. One is the conservative promise. The other is the stretch target that reflects what is possible when the team executes well.

The key is discipline. When building the macro plan, we must resist the temptation to think our project is different. We must not shortcut reality. Reference class data gives us the average, the normal distribution, the baseline. That is what we promise. Then, with lean tools like Last Planner, tact planning, and production control, we can improve upon the baseline and aim for the target.

Of course, this only works if we are actively collecting and storing reference class data. This is where the industry is sorely lacking. Too often, companies say they want to track production rates, but they make it overly complex and try to measure activity by activity. In my experience, that has never worked. What would work is a simple database. For each program type such as hospital, lab, multifamily, or power plant, we record the baseline budget, the baseline schedule, and the actual outcomes. We calculate the variance. Over time, this database becomes our most valuable tool for forecasting.

If we know, for instance, that a certain program type consistently experiences delays in switchgear procurement, we do not just shrug and hope our project will be different. We plan ahead, confront the risk, and take action to prevent it. This is how we respect the data while still striving for better results.

So here is my challenge to you. Start by identifying your own reference classes. What types of projects do you typically build? Gather as much data as possible about past performance. Then, when you plan, separate your promise and your target. Use the reference class to ground your promise, and then use lean techniques to aim for a better target. And above all, do not assume you are unique or exempt. The data tells the story. Our job is to listen and act.

This is the future of reliable planning. If we start collecting and using reference class data, we will stop overpromising, stop underdelivering, and finally build a culture of trust and predictability in construction.

Key Takeaway
Reference classes stop us from making wishful promises. By grounding project plans in real data and then targeting better outcomes with lean methods, we can protect our commitments and still strive for improvement.