The model must prove harm for the class with common evidence, measure only the harm alleged, and survive the members who were never injured at all.
Start a conversation with Cournot, the Institute’s damages concierge, already scoped to class-wide models. Pick a starting point, or describe the dispute directly.
A class-wide damages model carries three burdens at once. It must be common: capable of establishing impact and measuring damages for the class as a whole, from evidence and methods that apply to everyone, because a model that requires claimant-by-claimant inquiry concedes the predominance argument. It must fit: measure the damages flowing from the specific theory of liability being certified, and nothing else, because a model that bundles harm from conduct the court did not certify, or measures a loss the theory does not assert, is a recognized and case-ending defect. And it must be honest about variation: real classes contain members who paid different prices, bought different quantities, valued the product differently, and some who suffered no harm at all, and the model has to absorb that heterogeneity rather than bury it inside an average. The working forms are familiar, regression models of price or pay structure, formulaic damages built on common records, survey-based valuation of the misrepresented attribute, but the test they face is distinctive: an opposing expert whose entire assignment is to show that the method conceals individual questions. Building for that examination from the start, rather than retrofitting after the rebuttal lands, is the difference between a model that gates the case open and one that gates it shut.
Commonality, fit, and heterogeneity. Every certification fight visits all three.
Evidence that the conduct affected the class generally: pricing structure, wage policy, uniform representations. The predicate for everything.
The model measures the certified theory’s harm, only that harm, and can separate it from harm the theory does not assert.
Different prices, volumes, preferences and circumstances inside the class, absorbed by the model rather than averaged away.
The share of the class the conduct never touched, and whether the model can identify or bound it.
Transaction, payroll or pricing records covering the class. The model’s claim to commonality is only as good as its data’s coverage.
How an aggregate figure would reach individual members. Planned with the model, not after it.
How the Institute approaches a class-wide model.
In aggregate litigation, often whether there is a case at all.
A model that measures damages from four theories of harm when only one was certified does not merely overstate the number. It fails to measure the certified harm at all, and the case law treats that as a reason to deny certification. Trace every dollar the model measures back to the theory it flows from.
A mechanism by which the conduct plausibly affected class members generally, supported by evidence about how the relevant market or workplace actually operates. In a price case, that is usually evidence of pricing structure: that prices moved together across customers and products, or were set by common list prices, formulas or negotiations anchored to them, so that an overcharge at the top propagated broadly. In a wage case it is a uniform policy applied through common payroll systems. The structural evidence matters because it converts an assertion, everyone was harmed, into an economic account of why harm would be general, and it is tested against the defendant's counter-evidence of individualized negotiation, discounting or exemptions that would break the propagation.
They can summarize; they cannot substitute for a method that engages with variation. An average overcharge or an average unpaid hour is a legitimate output of a well-specified model, and aggregate damages are routinely expressed that way. The failure mode is using the average to manufacture commonality: applying a uniform figure to members whose actual experience varied so much that many fall at or below zero, so the average is doing the work of hiding individual questions rather than answering common ones. The line between the two is whether the underlying model explains the variation, with member-level or segment-level analysis showing harm across the distribution, or suppresses it. Rebuttal experts head straight for the distribution, and a model that has already mapped it holds up.
Identified, bounded or carved out, and above all engaged with rather than denied. The strong approaches are structural: define the class or the model so the uninjured are excluded from the start, purchasers of a product line the overcharge never touched, employees below the relevant threshold; or build the model at a granularity that identifies zero-harm members so they can be handled at allocation; or, where identification is impossible, quantify the plausible share honestly and let counsel argue its legal significance, which varies by forum. What fails is the reflexive claim that literally everyone was harmed, in the face of obvious counterexamples, because it stakes the model's credibility on its weakest assertion and hands the defense a demonstrative exhibit.
It grows up, and consistency between its two lives matters more than most teams expect. At certification the model is a methodology: a demonstration that class-wide proof is possible, often run on preliminary data. At the merits it becomes the actual damages calculation, on full discovery. The defense will compare the two, and material shifts, a different specification, a conveniently changed benchmark, results that moved sharply once the method was applied for real, become impeachment: either the certification showing was inflated or the merits model is. The protective discipline is to build the certification model as close to trial-grade as the data then allows, and to document the reasons for every later refinement, so evolution reads as diligence rather than as retreat.
Describe the class and the theory. The Institute will help you see whether the model can carry both burdens.