The Pull Plan & The Reference Class

The Pull Plan and the Reference Class: Why Data Has to Come Before Optimism

There is a moment in every pull planning session that feels like a breakthrough. The trades are in the room. The sticky notes are on the wall. The sequence is taking shape. And then somebody stands back and looks at the overall duration the collaborative process just produced and it is dramatically shorter than what any previous experience with this type of work would suggest is realistic. Everybody in the room is excited. The energy is good. The plan feels achievable because the people who built it believe in it.

That moment is where projects go wrong. Not because pull planning is wrong, pull planning, done correctly, is one of the most powerful production tools in construction. But because collaborative optimism, when it is allowed to override historical data, produces timelines that cannot be executed and milestones that will be missed. The pull plan is a sequence. The reference class is the reality check. You need both, and you need them in the right order.

The Pendulum: CPM on One End, Rigid Single-Train Takt on the Other

Before getting to the reference class, there is a related pendulum that needs to be named because it shapes how pull plans get built. On one end of the pendulum is CPM, trade stacking, trade burdening, no flow, everything piled into zones without regard for the sequence dependencies that allow trades to move cleanly. CPM produces overloaded zones, constant interruption, and schedules that look full of activity while the project falls behind.

On the other end of the pendulum is something equally problematic: single-train Takt planning applied rigidly, where one universal Takt time is forced on every trade regardless of their natural crew compositions, natural cycle times, and natural production rhythms. When trades are forced into a single Takt time that does not fit their scope, the crew compositions get distorted, the work gets done in ways that do not match how those trades actually perform, and the production system looks theoretically clean while producing real-world dysfunction. The United States has gotten CPM wrong. Some European practitioners have gotten rigid single-train Takt wrong. Both extremes fail the trades and the project.

The right answer is multi-train Takt planning where trades flow at the rhythms that are natural for their scope, trains that share phases move at compatible speeds and proper spacing, and multiple trains intersect correctly rather than being forced into an artificial uniformity. The pull plan is the tool through which that multi-train logic gets built collaboratively with the people who will actually execute it. But the pull plan only works if the duration it produces is grounded in historical reality.

What a Reference Class Actually Is

A reference class is a historical dataset that tracks the schedule and budget outcomes for similar projects of a certain type, a certain program category, a certain size and complexity range, so that planners have a data-driven baseline to evaluate their plans against. Not one person’s memory of how long something took. Not a gut feeling from the most experienced person in the room. A systematic record of how this category of work has actually performed across multiple real instances.

The Macro Takt plan establishes the high-level schedule that the owner and the project will be held to. The reference class is one of the most important inputs to that plan, because it tells the team what this type of work historically requires before anyone has the chance to talk themselves into something more optimistic than the data supports.

The principle behind reference class forecasting developed in planning research and applied in major infrastructure and construction programs worldwide is straightforward: the single best predictor of how long a future project will take is how long similar past projects actually took, not how long the people planning the current project believe it will take. Individual project teams are almost always optimistic. The data is almost always more accurate.

A Story That Makes This Unforgettable

Here is a real example of what happens when the reference class gets ignored. A project team was planning a phase. The historical data for phases like this, same project type, same program, same complexity showed consistently that the phase took somewhere between twelve and a half and thirteen months. That was the reference class. That was what the data said.

The team did a pull plan. The collaborative session was energetic. The trades came together, worked through the sequence, committed to the zones, and produced an eight-month duration. Eight months. The room was excited. The plan felt like a breakthrough, a genuinely improved outcome produced by better coordination, better sequencing, and stronger trade commitment than previous projects had managed.

The team talked themselves into it. The rose-colored glasses went on. The reference class got set aside because the pull plan felt so convincing, so collaboratively built, so committed to by the people who would actually do the work. The milestone was set at eight months.

It took thirteen months. Exactly what the reference class said it would. The enthusiasm in the planning session did not change the underlying production reality. It just delayed the point at which that reality became visible.

Why This Happens: The Psychology of Pull Planning

The reason this failure pattern is so consistent is not that pull planning is wrong. It is that pull planning, when done well, creates genuine conviction. The trades who built the plan feel ownership of it. The sequencing feels tight and logical. The collaborative energy produces a sense that this time will be different, that the coordination, the buy-in, and the shared commitment will overcome whatever caused previous phases to take longer.

Sometimes that conviction is justified. Genuinely improved preconstruction, better trade selection, better logistics, and a more disciplined production system can compress a phase beyond what historical averages show. But the improvement has to be real and accountable, specific changes to specific constraints that explain specifically why this phase will perform differently than the reference class. “We’re more committed this time” is not a sufficient reason to override thirteen months of historical data with an eight-month plan. Commitment is an input to the plan. It is not a reason to ignore the data.

The Right Relationship Between Pull Plan and Reference Class

Here is the discipline that protects the project without dismissing the value of the pull plan. The pull plan and the reference class are not competing tools. They are complementary, and the reference class comes first.

Before the pull planning session, the team reviews the reference class for this type of work. They know going in what the historical performance data says. They know the range of outcomes for phases like this one, what the fast end looks like, what the slow end looks like, and what the typical case looks like. That knowledge shapes the conditions of satisfaction they bring into the session and the validation they apply to the duration it produces.

After the pull planning session produces a duration, the team asks one question before accepting it: does this match the reference class? If the pull plan says eight months and the reference class says thirteen, the team does not accept eight months as the milestone. They may target eight months as an aspirational performance goal, something the team works toward without promising to the owner or locking into a contract. But the milestone, the external commitment, the number the owner is counting on that number reflects the reference class until the pull plan can demonstrate specifically how the constraints that caused previous phases to take thirteen months have been eliminated.

The pull plan is for sequence, trade buy-in, and constraint identification. The reference class is for schedule reality. Use both. Weight the reference class when they disagree.

Warning Signs That the Reference Class Is Being Ignored

Before the milestone gets locked to an optimistic pull plan duration, watch for these signals that the data conversation is being skipped:

• The team references how committed the trades are but cannot identify specific constraints from previous phases that have been structurally eliminated in this one.
• The pull plan duration is dramatically shorter than anything the reference class for this project type has ever achieved, without a specific explanation for why this phase is fundamentally different.
• The energy in the room has shifted from “does the data support this?” to “everyone believes in this plan” with conviction replacing analysis.
• The milestone is being set to the pull plan output before anyone has confirmed that the output is consistent with historical performance data.
• The project team is using the phrase “this time will be different” without being able to say exactly which specific conditions have changed.

Every one of those signals is the pull plan running ahead of the reference class. The correction is straightforward: go back to the data. Compare. If the data supports the plan, move forward with confidence. If it does not, adjust the milestone to reflect what the data actually says.

Data Is the Foundation of Every Good Plan

The pull plan is a remarkable tool. Done correctly with real trade input, genuine buy-in, a backwards pass that honors predecessor needs, and a diagonal flow check that confirms the sequence holds, it is the most accurate construction schedule that exists, because it was built by the people who will actually do the work. But it is still built by people who want to succeed and who naturally believe in the plan they just made. The reference class is the check on that natural optimism. It is the historical reality that holds the plan accountable to what this type of work actually requires. Together, they produce a plan that is both collaboratively owned and data-grounded, the only combination that consistently produces accurate milestones.

We are building people who build things. That includes building the discipline to check our optimism against the data and to make commitments we can actually keep, to owners, to trade partners, and to the families whose plans depend on the projects we deliver on time. If your project needs superintendent coaching, project support, or leadership development, Elevate Construction can help your field teams stabilize, schedule, and flow including the preconstruction discipline that uses both pull plans and reference class data to produce milestones grounded in reality.

A Challenge for Builders

On your current project’s next phase pull plan, build in one step: before accepting the duration the session produces, pull the reference class data for this project type and compare. If there is a gap, name it explicitly and explain what specific structural changes justify the deviation from historical performance. If you cannot name those changes specifically, revise the milestone to reflect what the data says. The commitment you protect by doing this is worth far more than the optimism you preserve by skipping it.

As W. Edwards Deming said, “Without data, you’re just another person with an opinion.”

On we go.

Frequently Asked Questions

What is a reference class and why does it matter for pull planning?

A reference class is a historical dataset of actual schedule and budget outcomes for similar projects, same type, program, and size that gives planners a data-grounded baseline to validate their pull plan against. It matters because individual project teams are almost always optimistic, and the historical record is almost always more accurate than the plan produced by a collaborative session.

What should happen when a pull plan produces a duration that disagrees with the reference class?

The reference class should be given precedence for the external milestone commitment. The pull plan duration can become an internal performance target, something the team works toward but the owner-facing milestone should reflect what the historical data says until the team can identify specific, structural reasons why this phase will perform differently than previous comparable phases.

Why is multi-train Takt planning the right answer between CPM and rigid single-train Takt?

Because CPM stacks trades without regard for flow, and rigid single-train Takt forces all trades to one Takt time that distorts their natural crew compositions. Multi-train Takt allows trades to run at the rhythms natural for their scope, with trains moving at compatible speeds and proper spacing, and multiple trains intersecting correctly. It respects how each trade actually works while still protecting Trade Flow and buffers across the whole phase.