Cornerstone Research Case Interview: Two Rounds, Two Cases

The two Cornerstone Research case interview rounds it prints

The analyst interview page states the shape in one sentence: "The interview process for the Analyst and Summer Analyst roles is identical. You will participate in two rounds of interviews: two virtual 30-minute first-round interviews, and a multi-hour final round in one of our offices."

The final round row leaves the office unnamed, and the firm names them elsewhere. Its own offices page lists nine: Boston, Brussels, Chicago, London, Los Angeles, New York, San Francisco, Silicon Valley and Washington, checked 8 October 2026 at cornerstone.com. That matches the over 1,000 professionals across nine offices its London posting states, so a Cornerstone Research interview that reaches the final round lands in one of those nine, and Chicago is on the list.

Two of those rows repay a second look.

The first is the 30-minute figure. Two 30-minute conversations that each carry a case leave very little room for warm-up. The firm's stated purpose for the round covers a lot of ground in that time: "Our interview process is designed to provide a glimpse into the nature of our work while allowing us to assess your thought process, comprehension, communication skills, and ability to receive and apply real-time feedback." The last item on that list is the one candidates underprepare. Receiving a correction mid-case and using it is part of what gets measured.

The second is the associate row. The URL that would hold an associate interview page returns a 404, tested on 29 September 2026. Doctoral and M.B.A. candidates get a published materials list and no published format. Analysis Group publishes the comparable shape for an analyst, two rounds with the second held at the office you applied to in front of roughly three to five consultants.

Two practice cases Cornerstone Research gives away

The firm hosts a page of analyst case examples with two full cases. Each carries four questions, a hint for each question, a worked answer for each, and a closing summary of what the real case team did. It is a free, official, firm-authored practice set, and it is the single most useful thing in this dossier.

Case one is a securities matter. A high-tech company, Zilo, goes public at $20 a share, trades above $60, then falls to $16 after losing two customers who together made up 40 percent of sales. Shareholders sue. Plaintiffs claimed $40 a share, the entire drop across the class period, and the case asks you to say why that number cannot be right.

The first question is the whole trade in one line: "What is the problem with Plaintiffs' approach to estimating damages?" The published answer says plaintiffs "effectively attributed all stock price movements during the class period to the alleged misconduct." The remaining questions walk you to a regression on the pre-period, a market index, an industry index, and then an event study. The closing summary gives the punchline: "the team concluded that only $3 of the price drop might relate to the allegations."

Case two is intellectual property. A manufacturer claims lost profits from a rival's use of its patented techniques. The questions test whether the two firms even sold to the same customers, then what else could explain the decline, then the step most candidates skip. The hint for that one is a single line reminding you that profit equals revenue minus costs.

Work both of them out loud before your first round. They are the firm's own answer to what it wants to hear.

A securities damages walk, and the factor of 24

Cornerstone Research publishes its cases without arithmetic. The numbers below are ours, built in the shape its securities case describes, so you can feel where the money actually goes.

The setup. A company has 80 million shares outstanding. Across a 30-day class period the share price falls from $52.00 to $38.00. Plaintiffs multiply the $14.00 drop by all 80 million shares and claim $1.12 billion.

Move one: take out the market and the sector. A regression on the year before the class period gives the stock a 0.8 sensitivity to the broad index and a 1.2 sensitivity to its industry index. Over the class period the broad index rose 1.0 percent and the industry index fell 9.0 percent. Predicted return is 0.8 times 1.0 percent plus 1.2 times minus 9.0 percent, so 0.8 percent minus 10.8 percent, which is minus 10.0 percent. The but-for price is $52.00 times 0.90, so $46.80. Market and sector explain $5.20 of the $14.00.

Move two: isolate the disclosure. The company-specific residual is $46.80 minus $38.00, so $8.80. An event study matches each significant daily move to the news that day. A guidance cut and a CFO departure account for most of it. Only $3.20 a share lands on the disclosure at issue.

Move three: count the shares that can claim. Only shares bought during the class period and still held at the disclosure are eligible. A trading model puts that at 18 percent of shares outstanding, so 80 million times 0.18, which is 14.4 million shares.

Move four: land it. Damages are 14.4 million times $3.20, which is $46.08 million.

Say what that shows, precisely. The claim fell from $1.12 billion to about $46 million, a factor of roughly 24, and no single adjustment did it. Three multiplied: $14.00 to $8.80 on market and sector, $8.80 to $3.20 on unrelated news, and 80 million shares to 14.4 million on eligibility. Sanity check before you speak: $46 million against a market value of 80 million times $52.00, which is $4.16 billion, is about 1.1 percent. That is a plausible size for a securities damages figure. Ten times that would not have been.

How we score a quantitative answer

Cornerstone Research publishes no marking scheme of any kind. The hundred points below are ours, written by us and read from the code that runs our grader. The firm has never seen them and endorses nothing here.

What the application asks for, and what kills it

The how to apply page is short and unusually strict. It opens with a warning most careers pages soften: "Applications with missing materials will not be considered."

Four consequences follow. The office preference belongs in the cover letter, so it needs a sentence rather than a checkbox. The transcript requirement makes this a slower application to assemble than a one-click apply. The portal instruction means a school system submission does not count on its own. And the deadline belongs to the posting rather than to the firm: the London analyst role printed 15 November 2026, and the US portals carried different dates or no open roles at all when we read them.

One more rule sits on the interview page and on the associate page alike: "During interviews and assessments, we do not permit the use of AI. We expect candidates to rely on their own thoughts, analysis, and logic."

A preparation order from what the firm already gives you

The order below is ours, and every step uses material the firm publishes rather than material you have to go and find. It is written for the two 30-minute first-round slots, because that is the only round with a published length.

1. Work the securities case out loud, hints covered. Four questions, four hints, four worked answers. Cover the hints, answer each question aloud, then read the firm's version. The gap between your wording and theirs is the whole diagnostic, and it costs one sitting.
2. Work the intellectual property case the same way, two days later. The gap matters more than the case. A second attempt on the day you learned the first teaches you the answers rather than the method.
3. Rehearse being corrected. The firm names the ability to receive and apply real-time feedback in its stated purpose for the round. So have somebody interrupt your second pass with a wrong objection, and practice taking the correction and continuing. Defending a first answer is the behavior this round is built to detect.
4. Fit the whole thing inside thirty minutes, once. Prompt, structure, calculation, close. A candidate who has never run the sequence against a timer discovers in the room that the close has no minutes left in it.
5. Prepare the behavioral half as heavily as the case. The firm's own manager says to consider the behavioral components as much or more than the case element. Two 30-minute slots with a case in each leave perhaps ten minutes apiece for everything else, so two or three episodes rehearsed to ninety seconds is the right amount of material.

Assemble the application before any of it. Missing materials are refused outright, and a transcript takes longer to obtain than a cover letter takes to write.

Six misreadings of a Cornerstone Research case interview

1. Rehearsing a profitability tree. Both published cases ask what caused a specific loss and how much of it. Neither asks for a recommendation. A framework with no counterfactual branch answers a question the firm did not pose.
2. Skipping the two published cases. They carry the firm's own hints and its own worked answers. Preparing from anything else when these exist is a choice worth reconsidering.
3. Treating the case as the whole interview. Spencer Pigg, a manager in Washington quoted on the firm's page, says candidates "should make sure to consider the behavioral components as much or even more than the case element." He also says the cases are built to suit varied academic backgrounds.
4. Ignoring the feedback loop. The firm names "ability to receive and apply real-time feedback" in its stated purpose for the round. Defending a first answer against a hint spends that point rather than earning it.
5. Preparing econometrics you cannot defend. The published cases describe regressions and event studies at the level of what they do. Claiming a method you cannot explain plainly is worse than naming the one you know.
6. Assuming the associate day matches the analyst day. The firm publishes an interview page for analysts and summer analysts only. Doctoral candidates have a materials list and nothing else.

Our practice hub shows what else sits here for a specific firm, and the behavioral practice covers the probe that breaks a thin story.

What Cornerstone Research does not publish

How many people you meet in the final round. The page says multi-hour and stops. No headcount, no schedule.

What the associate interview contains. The analyst URL pattern has an interview page. The associate equivalent returns a 404.

Any pay figure. Neither the careers pages nor the UK analyst posting carried one when we read them on 29 September 2026.

An academic threshold. The London posting says "first-class honours is common among successful applicants," which describes who gets in rather than who may apply. Treating it as a cutoff reads more into the sentence than it holds.

Whether a written test exists. The AI guidance mentions "assessments" in the plural and no page describes one. That is not proof none exists. It means nobody should tell you what to prepare for.

A passing bar, an acceptance rate, or a number of seats. None of these appears anywhere on the site or the portals.

If your invitation email contradicts anything above, the invitation is current and this page is not. Send us what you saw with the date, and we will read it against the firm's pages before changing a word.

What changed on this page

We correct in public and we date the correction. The body above carries what the firm publishes today. This log carries what this page used to carry.

29 September 2026. The previous version, 277 words long, said that Cornerstone Research's screening steps could not be described because we had no way to verify them. That was the page's central claim and it was false. The firm publishes the number of rounds, the length of the first round, the presence of a case in each round, five preparation tips and two complete practice cases with worked answers. All of those documents now sit in the source register at the head of this page with the date each was read.

29 September 2026. The earlier version called the math drill the highest-yield preparation for this role, phrased as a property of the job. It is our recommendation. The four math weights are now named as ours at the point where the chart appears, and where we point at the firm's published case as support for one of them, the page says the connection is our inference.

29 September 2026. The worked damages example, the application table, the AI rule and the UK deadline are new. None of them appeared in the previous version.