An owner who watched the estimate for their forty-two-million-dollar K-12 build move from a conceptual number to a schematic number to a design-development number to a final GMP wants one thing before they sign: an explanation of why it moved, stage by stage, that makes the movement make sense. Did the GMP land where the progression predicted, or did it jump in a way that signals something went wrong? The variance reconciliation memo answers that question, tracing the estimate through its stages against the named delta thresholds and explaining each movement, and it is what converts a series of changing numbers into a story of controlled convergence the owner can trust. This lesson shows you how to use the AI-assisted estimate sandbox to run the stages and produce the reconciliation memo, while keeping the estimating judgment about what each variance means firmly human.

The Question the Reconciliation Answers

An estimate is not a single event; it is a sequence, refined at each design milestone as the design develops and the unknowns resolve, so the same project has a conceptual estimate, a schematic estimate, a design-development estimate, and a final GMP, and the numbers differ because the basis differed. The owner, watching those numbers change, needs to understand the changes, because an estimate that moved a lot between stages either reflects normal convergence as the design firmed up or signals a problem, scope that was missed early, a market shift, an error, and the owner cannot tell which from the numbers alone. The variance reconciliation memo is what tells them which, by explaining each stage-to-stage movement against the expected progression.

This connects directly to the named delta thresholds from the GMP narrative lesson: the thresholds set the expected range of movement between stages, and the reconciliation memo compares the actual stage-to-stage deltas against those thresholds, showing where the estimate converged as expected and flagging where it moved more than the threshold, with an explanation of why. A movement within the threshold is normal convergence and reassures the owner; a movement beyond the threshold is a signal that needs explaining, and the explanation is either a legitimate reason, a major scope addition the owner requested, a documented market change, or a problem the contractor needs to own. The reconciliation memo is the document that demonstrates the estimate matured in a controlled, explicable way, which is what earns the owner's confidence in the final GMP, and it is the owner-facing companion to the internal estimate progression.

The AI-Assisted Estimate Sandbox

The AI-assisted estimate sandbox is the environment where you run the estimate at each stage, exploring the cost implications of the design as it develops, and AI accelerates this in the ways the estimating chapter established: faster takeoff from the evolving drawings, faster comparison and rollup, faster what-if exploration of cost decisions. The sandbox lets the precon team see the cost picture at conceptual, schematic, and DD without each stage being a multi-day manual estimate, which means the team can keep the estimate current with the design and explore the cost trade-offs that target-value design depends on.

The discipline in the sandbox is the same dollars-gate discipline from the takeoff lesson, applied across the stages: the AI accelerates the estimating, but the quantities and the costs at each stage are verified to the rigor the stage demands, lighter at conceptual where the estimate is intentionally rough, tighter at DD where it is converging toward a commitment. The sandbox's speed is what makes running multiple stages and exploring trade-offs affordable, but each stage estimate is still the estimator's reconciled number, not the AI's raw output, because each stage estimate feeds the reconciliation and ultimately the GMP. The sandbox is fast estimating across stages; the estimator's verification at each stage is what makes the stage estimates real, and the reconciliation memo then traces the real stage estimates against the thresholds. The speed enables the staged estimating; the verification makes the stages trustworthy enough to reconcile.

The sandbox's speed makes running conceptual, schematic, and DD estimates and exploring trade-offs affordable, but each stage estimate is the estimator's reconciled number, not the AI's raw output, because each stage feeds the reconciliation and the GMP.

The Judgment in Explaining Each Variance

The heart of the reconciliation memo is not the numbers, which the sandbox produces, but the explanation of each stage-to-stage variance, which is estimating judgment the AI cannot supply. When the estimate moved from schematic to DD by some amount, the memo must explain why, and the why is a judgment about what drove the movement: design development that added detail and cost, a scope clarification, a market change in material or labor pricing, a correction of an earlier assumption, or a problem. Identifying the real driver of each variance requires understanding the estimate, the design changes, and the market, which is the estimator's knowledge, not the model's.

So the AI can draft the memo's structure and prose, presenting the stage estimates, the deltas, and the comparison against the thresholds, but the explanation of why each variance occurred must come from the estimator, who attributes each movement to its real cause. The danger of letting the AI explain the variances is that it would generate plausible-sounding explanations that may not be the real drivers, a fabricated narrative of why the estimate moved that does not match what actually happened, which is worse than no explanation because it misleads the owner about the project's cost dynamics. So the estimator identifies the real driver of each variance and the AI renders that attribution clearly, and the verification confirms each explanation reflects what actually drove the movement, not a plausible story. The memo's credibility rests on the variances being explained truthfully, which is the estimator's judgment, and the AI's role is to present that judgment in the clear, owner-readable form the reconciliation requires.

The AI ROI Memo: Justifying the Tool to the Owner

There is a second document that often rides along with the precon reconciliation on a CM-at-risk project, and it is a distinctive one: the AI ROI memo, justifying to the owner the investment in the AI tools used in precon, because on a CM-at-risk project the owner often pays for the preconstruction services and may question the cost of the AI tooling. When the GC has used Togal.AI for takeoff or other AI tools in the precon phase, and the owner is paying for precon inside the GMP, the GC may need to justify that tool spend as delivering value, which is what the AI ROI memo does.

The ROI memo makes the case that the AI investment delivered value: faster estimates that kept pace with the design, more thorough exploration of cost trade-offs in target-value design, more reconciled and reliable numbers, and the time savings that let the precon team focus on the judgment that improved the estimate. The honest framing, from the strengths and tier lessons, is to present the value as real and specific without overclaiming, the actual time saved and the actual capability gained, not a vague productivity percentage, with the named payback if you can substantiate it, typically the precon labor cost the tools saved against the tool cost. The ROI memo is itself a document AI can help draft, presenting the value case clearly, with the estimator owning the actual numbers behind the case, because an ROI memo that overclaims is as damaging to owner trust as a variance memo that misexplains. The reconciliation memo justifies the GMP and the ROI memo justifies the tools that produced it, and both rest on honest, estimator-owned numbers presented in AI-drafted clarity.

The Reconciliation and Target-Value Design

The variance reconciliation is especially valuable in a target-value-design process, where the team designs to a cost target rather than estimating whatever gets designed, because TVD makes the relationship between design decisions and cost the central management question, and the reconciliation is how that relationship stays visible. In TVD, the team sets a cost target early and makes design decisions to hit it, so each stage estimate is a checkpoint on whether the design is tracking the target, and the variances between stages are the record of the decisions that kept it on target or pushed it off. The AI sandbox is what makes TVD practical, because designing to a target requires knowing the cost implications of decisions quickly enough to steer, which the fast staged estimating provides.

The reconciliation memo in a TVD project tells the owner not just that the estimate converged but that it converged because the team actively managed the design to the target, which is a stronger story than passive convergence: it shows the contractor steering the cost rather than just reporting it. The variances become evidence of management, this stage moved down because we value-engineered the structure, this stage held because the design decisions were cost-neutral, and the estimator's explanation of each variance is the account of the cost management that TVD is about. So the reconciliation is not just a backward-looking explanation of why the number changed; in a TVD context it is the documentation of an active cost-management process, and the estimator's variance explanations are where that management is made visible to the owner, which is exactly the value-add an owner paying for sophisticated precon services is looking for. The AI makes the fast staged estimating that TVD needs affordable; the estimator's management of the design to the target and explanation of the variances is the service the owner is buying.

Honesty on the Variance That Went the Wrong Way

The hardest part of the reconciliation, and the part that most determines whether the owner trusts you, is how you handle the variance that went the wrong way, the stage where the estimate jumped beyond the threshold for a reason that reflects a miss rather than a request. The temptation is to bury it, to attribute it vaguely or fold it into other movements so it does not stand out, and AI's fluency could help you do exactly that, drafting a smooth narrative that obscures the bad variance among the good ones. That is precisely the wrong use, because the owner who later understands that a variance was misattributed loses trust in the entire reconciliation and the GMP it supports.

The honest and ultimately stronger approach is to flag the bad variance clearly, explain its real cause, and address what it means, because an owner trusts a contractor who is candid about a miss far more than one who is later caught obscuring it, and the reconciliation's whole purpose is to earn trust through transparency. If a scope was missed at schematic and caught at DD, causing a jump beyond the threshold, the memo says so, explains it, and addresses how it affects the GMP, rather than smoothing it away. This is the same principle as the GMP narrative's transparency on an over-threshold stage, applied to the moment it is hardest to honor, and it is where the estimator's ownership of the variance explanations matters most, because the AI would happily draft the smooth obscuring version if asked, and the estimator's judgment to be honest about the bad variance is what keeps the reconciliation trustworthy. The reconciliation that candidly explains its worst variance is more credible than the one that hides it, and that candor is an estimator's choice the AI cannot make for you.

The Applied Problem: Reconcile Three Stages to the GMP With an ROI Memo

Here is the exercise. Take a real or representative forty-two-million-dollar K-12 build and produce the three-stage estimate-to-GMP reconciliation, conceptual to schematic to DD to final GMP, with the owner-facing variance memo and the AI ROI memo attached. Run the workflow: use the AI sandbox to run the estimate at each stage with the verification each stage demands; compute the stage-to-stage deltas and compare them against the named thresholds; identify the real driver of each variance as the estimator; have AI draft the variance memo presenting the progression, the deltas against the thresholds, and your variance explanations; and draft the AI ROI memo with the substantiated value case.

Produce three things. First, the three-stage reconciliation, the estimates at conceptual, schematic, DD, and final GMP, with the deltas and the comparison against the named thresholds. Second, the owner-facing variance memo, explaining each stage-to-stage movement with its real driver, flagging any movement beyond threshold with its explanation, in the clear form that earns owner confidence in the GMP. Third, the AI ROI memo, with the honest, substantiated value case for the AI tools used in precon and the named payback if you can support it. Verify throughout: each stage estimate to its rigor, each variance explanation against what actually drove it, and the ROI numbers against the actual savings.

The deliverable is the three-stage reconciliation, the variance memo, and the ROI memo, and the lasting product is a precon close-out workflow that lets you present the owner a controlled, explained estimate progression and a justified tool investment, fast, so the precon process is both faster and more transparent. This completes the estimating chapter, and it ties its threads together: the takeoff feeds the stage estimates, the GMP narrative explains the final number, and the reconciliation memo explains the progression to it, all with AI accelerating the estimating and the drafting while the estimator owns the numbers and the explanations. The precon professional who masters this delivers owners a transparent, credible estimate story and a defensible tool-investment case, which is what builds the owner trust that wins the next CM-at-risk pursuit, achieved because the estimating was fast and the explanations stayed honest and human.

Key Takeaways

• An estimate is a sequence (conceptual, schematic, DD, GMP), and the owner needs to understand why the numbers changed. The variance reconciliation memo answers that by tracing each stage-to-stage movement against the named delta thresholds and explaining it.
• A movement within the threshold is normal convergence that reassures the owner; a movement beyond it is a signal that needs explaining, either a legitimate reason or a problem the contractor must own. The memo demonstrates the estimate matured in a controlled, explicable way.
• The AI-assisted estimate sandbox accelerates running the estimate at each stage, but each stage estimate is the estimator's reconciled number to the rigor the stage demands (lighter at conceptual, tighter at DD), not the AI's raw output, because each feeds the reconciliation and the GMP.
• The heart of the memo is explaining each variance, which is estimating judgment the AI cannot supply: the real driver (design development, scope clarification, market change, corrected assumption, or problem). Letting AI invent explanations produces a plausible narrative that misleads the owner.
• The estimator identifies the real driver of each variance and the AI renders it clearly; verification confirms each explanation reflects what actually drove the movement, not a plausible story, because the memo's credibility rests on truthful variance explanation.
• The AI ROI memo justifies the tool investment to the owner who pays for precon on a CM-at-risk project: real, specific value (time saved, capability gained) with a substantiated payback, not an overclaimed productivity percentage, because an overclaiming ROI memo damages trust as much as a misexplained variance.
• The artifact: a three-stage estimate-to-GMP reconciliation for a $42M K-12 build with the owner-facing variance memo (each movement explained with its real driver) and the AI ROI memo (honest, substantiated value case), all verified, completing the estimating chapter by tying takeoff, GMP narrative, and progression together.