Your Forecast Is a Negotiation, Not an Estimate

Ask an analyst with no stake in the outcome what next year's occupancy will be, and you get an estimate. Ask a regional manager the same question when their bonus depends on beating the answer, and you get a position.

Both numbers arrive in the same template, under the same column heading, and get consolidated into the same plan. Only one of them is trying to be right. And most organizations spend considerable effort trying to improve the accuracy of a number that was never attempting accuracy in the first place, which is why the effort so rarely works.

The system is paying for this

This is not a claim about anyone's character, and it is worth grounding in the research rather than presenting as cynicism. Michael Jensen of Harvard Business School made the definitive argument in a paper with a title that leaves little ambiguity: Paying People to Lie: The Truth About the Budgeting Process, published in European Financial Management in 2003, with an executive summary that ran in Harvard Business Review.

His finding is that paying people according to how their performance relates to a budget or target causes them to game the system, and destroys value in two distinct ways. First, both superiors and subordinates distort the numbers during the formulation of budgets, which strips the process of the unbiased information the organization needs to coordinate activity across its parts. Second, they game the realization of those budgets once set, manipulating actual results to land where the incentive points.

The sentence in his paper that ought to stop a CFO is the explanation of why this is universal rather than exceptional. People are taught to lie in these systems, he argues, because telling the truth gets them punished and distorting the number gets them rewarded. Nobody has to be dishonest by disposition. The system selects for the behavior, teaches it, and pays for it, and then everyone expresses surprise that the forecasts are unreliable.

What it looks like in a property portfolio

The general mechanism takes a specific and recognizable form in real estate, on both sides of Jensen's split.

During formulation, the distortion is conservative in a particular direction. Occupancy assumptions land a little below what the market would support. Rent growth is modeled cautiously. Expense assumptions carry comfortable headroom. Each individual choice is defensible, which is precisely what makes the pattern hard to challenge, and the aggregate result is a target that can be beaten by a competent manager having an ordinary year.

During realization, the gaming is more costly, because in property it frequently involves borrowing from the asset. Maintenance gets deferred to protect a net operating income figure, which improves the period number by degrading the building. Capital projects slip a quarter. Concessions are timed to fall on the favorable side of a boundary, and renewals are pushed or pulled across period ends. In the final quarter, remaining budget gets spent to protect next year's allocation, which is the same distortion running in reverse.

The maintenance one deserves particular attention, because unlike most budget gaming it has a physical consequence. The number improves, the asset gets worse, and the deficit compounds until it surfaces as a capital event that someone else inherits.

The problem is that one number is doing two jobs

Underneath all of this is a structural conflict that no amount of process improvement resolves.

A planning number needs to be the best available estimate of what will actually happen, because treasury plans cash against it, procurement plans volume against it, and the organization coordinates itself using it. A target needs to be a commitment that motivates and against which performance is judged and paid.

These are incompatible requirements for a single figure. A number optimized to be a beatable commitment cannot simultaneously be an unbiased estimate, and asking someone to produce both in one cell asks them to work against their own compensation. This is not a failure of integrity, it is arithmetic.

Jensen's emphasis on the coordination cost is the part most organizations underweight. The obvious damage of sandbagging is that targets are too easy. The larger damage is that the entire organization is now planning on numbers it knows are wrong in an unknown direction, which means every downstream decision, staffing, cash, procurement, capital allocation, is being made on distorted information. The distortion does not stay in the compensation system. It propagates into every plan built on top of it.

Why AI forecasting will not fix this

This matters now because a great deal of money is being directed at forecasting accuracy, and much of it is being spent on the wrong problem.

The pitch is straightforward: better models, more data, higher accuracy. But if the number is a negotiated commitment rather than an honest prediction, the error was never in the mathematics. Applying a better model to a gaming problem produces one of two outcomes, and both are bad.

In the first, the model produces a genuinely unbiased forecast, which is unwelcome. It removes the manager's ability to set a beatable target, so it gets overridden, adjusted for local knowledge, or quietly ignored in favor of the submitted number. The organization has purchased an accurate forecast that nobody uses.

In the second, and more common, the model gets tuned until its outputs are acceptable. Assumptions are adjusted, inputs are curated, the scenario is selected. This is the same gaming as before with additional steps, and it is worse than the manual version, because the resulting number now carries the authority of having come from a model. A sandbagged forecast presented as an algorithmic output is harder to challenge than a sandbagged forecast presented as a manager's judgment. This is the pattern of a number that looks rigorous while misleading, examined more generally in the metrics that lie about AI success.

No forecasting technology fixes an incentive problem, because the technology addresses the estimate and the problem is in what the number is for.

The honest part

Several qualifications matter here, because the argument can be pushed further than the evidence supports.

Targets are genuinely valuable. Commitment focuses effort, accountability improves performance, and an organization where nobody is answerable for a number is not obviously better than one where the numbers are somewhat gamed. Jensen's own prescription, making compensation a linear function of actual performance and independent of any budget or target, is intellectually clean and represents a significant compensation redesign that many organizations cannot or will not undertake. The Beyond Budgeting movement offers a more thorough alternative, and it works for some firms, but recommending wholesale abandonment of budgeting to a mid-size property company is not practical advice.

There is also a genuine difficulty distinguishing prudent conservatism from strategic sandbagging. A manager who builds buffer against real uncertainty is exercising judgment, not gaming, and the two look identical from above. Anyone claiming they can reliably separate them is overstating.

And forecasts are simply hard. Markets move, tenants behave unexpectedly, and a miss is not evidence of manipulation. The argument here is not that every variance is gaming. It is that the incentive structure introduces a systematic bias in a known direction, on top of the ordinary difficulty of prediction, and organizations rarely account for it.

Separate the two numbers

The practical move does not require redesigning compensation, which is why it is worth trying first. It requires producing two numbers explicitly rather than one number implicitly serving both purposes.

The forecast is the best available estimate of what will happen. No compensation attaches to it. It can be revised at any time, in either direction, and the person producing it is evaluated on its accuracy rather than on whether reality exceeded it.

The target is the commitment. Compensation attaches to it, it is set deliberately at the start of the period, and it may reasonably differ from the forecast, because a target can legitimately be a stretch.

Once the two are separated, several things become possible that were not before. Forecast accuracy becomes something you can actually reward, including upward revisions, which under the combined system are punished. Planning improves, because treasury and procurement work from a number that nobody had reason to distort. And the negotiation about the target becomes explicit and honest, conducted openly as a negotiation, rather than disguised as a technical exercise in prediction.

There is a single question that reveals which system you currently operate. If a regional manager revised their forecast upward in the second quarter, would anything bad happen to them? If the honest answer is yes, that they would be raising their own bar, inviting scrutiny, or setting an expectation they now have to meet, then you do not have forecasts. You have targets wearing a forecast's name, and every plan you build on them inherits a bias you have not measured.

FAQs

Q1. Isn't this just accusing managers of dishonesty?
No, and Jensen's argument is explicitly the opposite. The behavior is a rational response to how the system pays. When telling the truth about an achievable number reduces your compensation and shading it improves your compensation, distortion is the predicted outcome for reasonable people. The design produces the behavior, which is why exhorting people to be more accurate has no effect.

Q2. What are the two ways budget gaming destroys value?
Jensen identifies distortion during formulation, where both superiors and subordinates shade the numbers, which deprives the organization of the unbiased information it needs to coordinate across its parts. And gaming during realization, where actual results are manipulated to land favorably. The second is more visible, while the first is generally more costly because it corrupts every plan built on the numbers.

Q3. What does this look like specifically in property?
In formulation, cautious occupancy and rent growth assumptions with comfortable expense headroom, each defensible individually. In realization, deferred maintenance and capital projects to protect a period figure, timing of concessions and renewals around period boundaries, and end-of-year spending to protect next year's allocation. Deferred maintenance is the most damaging, because it improves the number by degrading the asset.

Q4. Why can't one number serve as both forecast and target?
Because they have incompatible requirements. A planning number must be an unbiased estimate so the organization can coordinate against it. A target must be a commitment that motivates and determines pay. A number optimized to be a beatable commitment cannot also be an honest estimate, and asking one person to produce both in a single figure asks them to work against their own compensation.

Q5. Will AI forecasting tools improve our accuracy?
Not if the underlying number is a negotiated commitment, because the error is not in the mathematics. Either the model produces an unbiased forecast that gets overridden because it is unwelcome, or it gets tuned until its outputs are acceptable, which is the same gaming with a veneer of objectivity. The second is worse, since a sandbagged number carrying algorithmic authority is harder to challenge.

Q6. What is the practical fix if we cannot redesign compensation?
Produce two explicit numbers. A forecast that carries no compensation, can be revised freely in either direction, and is evaluated on accuracy. And a target that carries compensation and is set deliberately at the start of the period. This separates the estimate from the commitment without requiring the full linear pay-for-performance redesign Jensen recommends.

Q7. How do we tell prudent conservatism from sandbagging?
Often you cannot, at least not case by case, and anyone claiming otherwise is overstating. A manager building buffer against genuine uncertainty is exercising sound judgment, and it looks identical from above to strategic shading. What you can observe is the pattern: systematic bias in one direction across many periods and many managers indicates structure rather than individual caution.

Q8. What single question exposes the problem?
Ask whether anything bad would happen to a manager who revised their forecast upward mid-year. If revising upward raises their own bar, invites scrutiny, or creates an expectation they must now meet, then the number is functioning as a target rather than a forecast, and every plan built on it carries an unmeasured bias.