Analyst hours are compressible. Economic judgement, testimony and surge capacity are not.
Start a conversation with Cournot, the Institute’s damages concierge, already scoped to the AI-era staffing question. Pick a starting point, or describe the dispute directly.
Something real is happening to the economics of economic consulting, and it is worth stating carefully because the exaggerated version is easy to disprove. Analysis published by the Stigler Center's ProMarket in April 2026 examined how substantially AI could compress analyst hours in the industry, with knock-on effects on cost. The major firms are not resisting the direction: Cornerstone has described deploying an agentic research platform across its case work, with the important discipline that outputs are replicated by a professional before they are used in litigation. What that industry practice tells a purchaser is precise — the labour-intensive middle of the work is compressible, and the verification of it is not. So the staffing question has changed shape. It is no longer how many analysts a matter needs, but which parts of the matter genuinely require human staffing and which are now substantially faster for everyone, including the large firms.
The top of this list is where the savings are. The bottom is where they are not.
Historically a large share of analyst hours. Substantially compressible.
Faster to write and faster to check, with review still required.
Large collections processed far faster than by hand.
Not compressible. Cornerstone replicates AI output by a professional before litigation use.
Choosing the measure, defending the counterfactual. This is the expert, and it does not automate.
Sixty requests over a weekend is a staffing problem, not a computation problem.
How the Institute assesses the staffing question.
Where the cost goes, and whether the engagement holds under pressure.
The industry standard being described publicly is replication by a professional before litigation use. A candidate who has not thought about verification is a risk regardless of how lean or how large their team is.
For some matters, closer to it than five years ago — and stating it more strongly than that is not supportable. What has genuinely compressed is the processing layer: data cleaning, coding, document review, the work that consumed junior analyst hours. Where a matter's difficulty lay in that layer, a lean configuration is now viable that would not have been. Where the difficulty is data at genuine scale, parallel workstreams, or trial responsiveness, the constraint was never computation and has not moved. The useful question is which of those a specific matter is, and it is answerable in a conversation.
Yes, publicly and substantially, which is exactly why this is not an argument against them. Cornerstone has described an agentic research platform applied across case work; the industry commentary treats adoption as general rather than exceptional. The productivity gain is therefore available on both sides of the choice, and anyone selling "lean beats institutional because AI" is not describing the market accurately. The right inference is narrower: the analytical layer is cheaper for everyone, so the case for institutional scale increasingly rests on throughput, redundancy and litigation support rather than on raw analytical horsepower.
It creates a verification obligation, and the industry practice being publicly described is a reasonable guide: replicate the output by a professional before it goes anywhere near a filing. An expert must be able to explain and defend the analysis as their own, which means understanding what was produced and having checked it — not merely having supervised a tool that produced it. This is a sensible question to put to every candidate regardless of their delivery model, and a candidate who has no answer to it is telling you something useful.
Add two questions to whatever you already ask. First, which parts of this engagement do you expect to be substantially accelerated by your tooling, and what does that do to the estimate — a candidate who cannot answer is either not using the tools or has not thought about passing on the benefit. Second, how is anything AI-assisted verified before it enters a report. Between them, those two questions surface both the cost picture and the risk posture, and the answers vary far more between candidates than rates do.
Describe the analytical load. The Institute will help you see what still needs people.