Second-order thinking in case interviews

What the feedback actually means

Candidates are given this advice constantly, and it is close to indecipherable until you push back and ask for a solid example of what the interviewer meant. Without that example, "go deeper" can become a catchall for a case that simply went badly. The first move in any debrief is to ask exactly that: point at the moment in the case where I stopped too early. Interviewers can nearly always name it, and once they do, the note collapses into one of a few specific gaps.

• Your framework was shallow. It was not exhaustive, or it was the obvious textbook version, or it was not tailored to this client and this prompt.
• You didn't consider the effects of your proposed plan. The recommendation was treated as the finish line rather than the middle of the analysis.
• You didn't consider remedial plans to countermand those effects. You spotted a risk, said it out loud, and left it hanging instead of designing around it.

To work out which one hit you, replay the case and find the part of the framework you went to, then work back from there. The missing depth is almost always the layer directly beneath the branch you were standing on when the case ended.

The map: first, second and third order

Strip away the jargon and the orders are just a chain of questions. First-order thinking answers the question in front of you: the move and its direct arithmetic. Second-order thinking asks what happens next if the client actually does it, to customers, competitors, your own people, and to the headline number itself. Third-order thinking designs the response to those effects: the counter-plan, the sequencing, and the rival's counter to your counter.

The map below shows where those layers sit in the flow of a case. The top row is the case every candidate runs. The depth lives underneath the first-order answer, and the dashed line on the right is the first-order candidate's shortcut: from answer straight to recommendation. Notice where the solid path ends up: the deeper layers don't replace the answer, they change it.

Read the map left to right, then downwards. The top row earns you the interview; the two bands underneath are what the feedback is asking for; and the black box at the bottom is what a synthesis actually sounds like when the layers are included. The dashed line is the path most candidates take without noticing.

The worked example: call centres to AI

A very crude example to illustrate the point. The interviewer hands you a cost case that begins like this:

"One of our client's major cost categories is its call centres, which are domestically domiciled today. Competitors have either outsourced this completely or moved to AI solutions for first-line customer support. That puts us at a cost disadvantage. Moving toward an AI first line would save the company $60 million per year."

The first-order answer is the one most candidates give: take the saving, size it confidently, recommend the move. It is not wrong, and the arithmetic may be flawless. It is simply not the end of the case.

The second-order step prices what the move does to the business around it:

"However, the effects on customer retention eat into that saving. Satisfaction drops through the sales and support process, the customers who churn are disproportionately the high-value ones who wanted a human, and the up-selling that agents used to close quietly disappears with them. Net, maybe a third of the $60 million never arrives."

There are more ripples available at this layer if the case goes there: the remaining domestic centres take a morale hit, and a regulator may take an interest in how a bot handles vulnerable complainants. You do not need all of them; you need the one or two that move the number.

The third-order step designs the remedial plan that countermands those effects:

"To countermand this, we propose a two-tier customer support model: a human-handled tier for Tier A customers, and a digitised self-serve tier for free and Tier B customers, piloted in one region first, with satisfaction thresholds that gate the national rollout."

Those latter two moves are the difference between being read as a first-order thinker and a higher-order one. The first candidate found the saving. The second priced the saving's side effects. The third redesigned the solution so the saving survives contact with reality, and turned a headcount cut into a service redesign in the process.

Playbook: profitability cases

Profitability is the most common case type and the most common place to be told you were shallow, because the first-order answer is genuinely good: segment the P&L, find the driver, fix the driver. The depth is in what the fix does next.

Playbook: cost-cutting cases

The call-centre example above is a cost case; here is the general shape of the layers. Cost cases punish candidates who treat the benchmark as the answer, because every cost line is attached to something the business actually does.

Playbook: M&A cases

In an M&A case the first-order answer is the synergy math, and it is usually the part every prepared candidate recites well. The differentiation is entirely in what happens to those synergies when a hundred integration decisions start landing.

Playbook: market entry cases

Market entry cases hand you a market size and an entry mode and invite you to stop. The layer most candidates miss is that the market has incumbents, and incumbents are not spectators.

Building the habit in the room

The layers are generative, not mystical. At every recommendation, run four questions: what breaks? what improves? who retaliates, and how fast? what does this cost us in flexibility later? The angry parties are usually a short list: customers, competitors, regulators, employees, suppliers. Run the list, keep the one or two answers that move the number, and design around them.

Budget the depth. Third-order thinking on every branch is boiling the ocean in reverse, and interviewers penalise depth without judgement just as they penalise shallowness. Go deep on the branch that carries the decision, say out loud that you are doing so, and leave the other branches deliberately shallow. One deep branch beats three shallow ones, and choosing which branch to deepen is itself the judgement being tested.

Finally, signpost the layers as you speak, because interviewers grade what they hear. A synthesis that carries all three sounds like this:

"Headline: the move to an AI first line saves $60 million a year. Before I recommend it, two effects could erode it, churn among high-value customers and the up-sell the agents were closing, worth roughly a third of the saving. With the two-tier service model gating the rollout, I would underwrite about $40 million net and phase the switch by region."

And when the debrief hands you the feedback, push back and ask for the solid example. That conversation is where "you didn't go deep enough" turns from an insult into a to-do list, and candidates who ask it stop receiving the note.

Frequently asked questions

What does 'you didn't go deep enough' mean in a case interview?

It is shorthand for one of a few specific gaps: a framework that was generic rather than tailored to the prompt, a recommendation made without pricing its knock-on effects, or risks identified with no counter-plan attached. Ask the interviewer for the exact moment you stopped early; the missing layer is usually directly under the branch of the framework you were standing on.

What is second-order thinking in a case interview?

Asking what happens next if the client actually adopts your recommendation: how customers react, how competitors respond, what it does to your own people, and how the side effects change the headline number. The move itself is first order; its consequences are second order.

What is third-order thinking in a case interview?

Designing the response to your recommendation's side effects: the counter-plan (such as a two-tier service model), the sequencing and quality gates, and the anticipation of the competitor's counter-move. It is the layer that turns a finding into a recommendation that survives contact with reality.

How do I show deeper thinking without running out of time?

Go deep only on the branch that carries the decision, and signpost the layers out loud: headline first, then the effects that could erode it, then the counter-plan and the net number. One deep branch with a clear structure beats three shallow ones, and interviewers grade what they hear.