How to Practice for McKinsey Solve: Building Reasoning with AI

The old paper Problem Solving Test rewarded a narrow skill: reading dense passages and picking defensible multiple-choice answers under a clock. McKinsey Solve rewards something harder to fake — the ability to build a working model of a system, investigate messy data, and make sound decisions when the clock is against you. Most candidates who struggle with Solve are not short on raw intelligence. They practice the wrong thing, or they practice with no way to tell whether they are getting better.

You cannot rehearse the exact Solve games. McKinsey rotates them and scores your process, not just your outcome. But the reasoning underneath is completely learnable, and it is the same reasoning the live case interview tests. This guide gives you a practice loop that develops that reasoning — and shows how to run each part with AI feedback so you are training against a signal, not a feeling.

What Solve Actually Measures

Since around 2020, McKinsey replaced the PST with the digital Solve assessment, now the global default in most offices (a few may still run the older PST format). A Solve session is two mini-games, taking roughly 60 to 81 minutes total. The two you are most likely to see are:

• Ecosystem Building — you construct a food chain or reef system under resource constraints, balancing what each species needs against what the environment provides. This is constrained-optimization and systems thinking dressed up as a game.
• Red Rock Study — you investigate a dataset to answer questions, filtering, cross-referencing, and drawing conclusions from evidence. This is hypothesis-driven data analysis.

Occasionally candidates get Sea Wolf or Plant Defense instead. Scoring is algorithmic and process-based — the system tracks how you work, not only your final answers — and roughly 70% of candidates are eliminated at this stage.

Read those two core games again through a recruiter’s lens: can this person model a complex system under constraints, and can they investigate data to reach a defensible conclusion? Those are the two reasoning muscles worth building. Everything below trains them.

The Practice Loop

Random practice plateaus fast. A loop that closes the feedback gap is what moves scores. Run this cycle every session:

flowchart LR
    A[Diagnose<br/>weakness] --> B[Targeted<br/>drills]
    B --> C[Timed<br/>simulation]
    C --> D[Review<br/>with feedback]
    D --> A

Each pass around the loop should sharpen a specific weakness, not just log hours. The difference between a candidate who improves and one who stalls is almost always the review step — closing the gap between what you did and what strong reasoning would have done.

Step 1 — Diagnose Your Weakness

Before drilling anything, find out which reasoning muscle is weakest. Work one full case end to end and watch for where you break down:

• Decomposition breaks down if you jump to data or calculations before you have a structure — you analyze symptoms instead of driving toward a root cause.
• Data investigation breaks down if you read every number at the same weight, miss what an exhibit is actually telling you, or can’t form a hypothesis to test against the data.
• Mental math breaks down if you are accurate but slow, or fast but sloppy, when the clock is running.
• Decision-making breaks down if you can analyze but freeze when it is time to commit to a recommendation.

The fastest way to get this diagnosis is to have something score you. CasesCoach’s AI feedback rates every practice case on five dimensions — Structure, Analysis, Communication, Business, and Math — so your weakest dimension is not a guess. If Structure and Analysis lag while Math is fine, you have a decomposition-and-investigation problem, which is exactly what Solve’s two core games stress.

Step 2 — Targeted Drills

Once you know the weak muscle, drill it in isolation. Mixing everything together every session hides where the real gap is.

Structured decomposition. Take a business problem and force yourself to build a MECE structure before touching a single number. For a profitability question, break revenue and cost into drivers; for a market question, break demand into segments. The discipline of structuring first is the same discipline Ecosystem Building rewards — you are deciding what matters and how the pieces relate before you optimize.

Hypothesis-driven data analysis. Open an exhibit and, before analyzing it, write down what you expect it to show and why. Then test that hypothesis against the data. This is the Red Rock reasoning pattern: you are not passively reading numbers, you are interrogating them with a question in hand. Practicing this turns data from a wall of figures into evidence for or against a specific claim.

Mental math under time. Solve’s investigative games reward quick, accurate estimation. Four operations cover most of it:

• Percentages: “What is 23% of $4.2M?”
• Growth over periods: “Revenue grew 8% annually for 3 years — what’s the multiple?”
• Ratio comparison: “Is A/B greater or less than C/D?” — cross-multiply, don’t compute exact values
• Weighted averages across segments

Generate 20 random calculations a day, time yourself, and target under 15 seconds each. Our mental math guide has the shortcuts worth memorizing.

For beginners, drill with structure and hints. If you are early in preparation and your decomposition is shaky, practicing under full pressure just reinforces bad habits. CasesCoach’s Guided Mode is built for this stage — the AI interviewer offers hints and scaffolds the structure as you go, so you learn the shape of good reasoning before you are asked to produce it unassisted. Think of it as drilling with a coach standing next to you.

Step 3 — Timed Simulation

Reasoning that only works untimed does not survive a 60-to-81-minute assessment. Once a drill feels comfortable, put it under the clock.

The point of simulation is not just speed — it is holding your reasoning quality steady while the pressure is on. Most candidates lose 10 to 20 points of structure quality the moment a timer starts. You want to close that gap by rehearsing under conditions that feel real.

This is what CasesCoach’s Expert Mode is for. It simulates the pressure of a real interview — no hand-holding, a demanding interviewer, and the expectation that you drive. Use it for sprint and stress training once your fundamentals are in place. Alternating Guided Mode (to build the habit cleanly) with Expert Mode (to prove it holds under pressure) is the fastest route from shaky to sharp.

Step 4 — Review With Feedback

This is where scores actually move. After every simulated case, do not just note whether you “got it right.” Interrogate the reasoning:

What to review	Question to ask	Signal it’s your weak spot
Structure	Did I decompose before analyzing, and was it MECE?	You started calculating with no framework
Data use	Did I form a hypothesis, or just describe numbers?	You listed what exhibits said without a “so what”
Math	Were calculations fast and correct under time?	Accurate untimed, error-prone when rushed
Decision	Did I commit to a clear recommendation?	You analyzed thoroughly but hedged the answer

A five-dimension score after each case turns this review from a vague feeling into a tracked number. When your Analysis score climbs from session to session, you know the hypothesis-driven drilling is working. When Structure stalls, you know where to point the next loop. Reviewing against a consistent rubric is what converts practice hours into a rising score.

Practice Hands-On Decomposition: Chat with Case

The single best exercise for the reasoning Solve investigates is working a real case the way a consultant actually would — not answering trivia about it, but reasoning through it stage by stage. CasesCoach’s Chat with Case walks you through a real case in five stages:

1. 背景 (Context) — understand the situation and the question being asked
2. 选框架 (Choose a framework) — build the structure that fits this specific problem
3. 填数据 (Populate with data) — pull the relevant numbers into your structure
4. 分析 (Analyze) — test hypotheses against the evidence
5. 结论 (Conclude) — commit to a recommendation and defend it

That progression is a direct rehearsal of Solve’s investigative reasoning: you decompose, you gather evidence, you interrogate it, and you decide. Because you work against a real case from the library rather than an abstract prompt, the decomposition muscle you build transfers straight to Ecosystem Building and Red Rock — and to the live case interview that follows.

A Realistic Weekly Schedule

A sustainable rhythm for someone preparing over four weeks, roughly five hours a week:

Day	Session	Duration	Focus
Mon	Targeted drill	45 min	Weakest reasoning muscle from diagnosis
Tue	Mental math	20 min	Speed estimation, timed
Wed	Chat with Case	45 min	Five-stage decomposition on a real case
Thu	Guided Mode case	40 min	Build structure with AI hints
Fri	Expert Mode case	45 min	Full-pressure simulation
Sat	Two full cases + review	90 min	Simulation, then five-dimension review
Sun	Review + plan	20 min	Read your score trends, re-diagnose

Notice the loop repeats across the week: diagnose on Sunday, drill early, simulate mid-to-late week, review on Saturday. Candidates who hold this rhythm for three to four weeks typically see their weakest dimension come up to the level of their strongest — which is exactly the balanced profile Solve’s process-based scoring rewards.

Where Solve Prep Meets the Case Interview

Solve is a gate, not the finish line. The candidates who clear it well are usually the ones who prepared for it as reasoning practice rather than test-cramming — because that same reasoning is what the live case interview and the PEI behavioral round demand next. The four muscles in this guide (decomposition, data investigation, mental math, decision-making under time) do double duty.

For the full picture of the assessment itself, start with our McKinsey Solve preparation hub. To go deeper on the individual reasoning skills, see problem-solving strategy by question type and data interpretation techniques — both translate directly to Red Rock’s investigative demands.

Key Takeaways

• You cannot drill the exact Solve mini-games, but the reasoning they measure — systems thinking and data investigation — is fully learnable
• Run every practice session as a loop: diagnose your weakest muscle, drill it in isolation, simulate under time, then review against feedback
• Use a five-dimension score (Structure, Analysis, Communication, Business, Math) to make your weakness concrete instead of guessing
• Build structure cleanly in Guided Mode, then prove it holds under pressure in Expert Mode
• Use Chat with Case to rehearse hands-on decomposition through five stages on a real case — the closest thing to Solve’s investigative reasoning
• The reasoning you build for Solve carries straight into the case interview and PEI round that follow

Start Practicing Today

The gap between candidates who pass Solve and those who don’t is rarely talent — it’s whether their practice built the right reasoning and gave them a way to measure progress. You can start closing that gap now.

Create a free CasesCoach account and you get 3 full cases plus AI Mock Interview to run your first diagnose-drill-simulate-review loop and see your five-dimension scores. When you are ready to build a real routine, upgrading to Pro unlocks our full library of 835+ real cases and multiple AI mock sessions — enough to run the weekly loop above for as many weeks as your timeline needs.

Ready to build the reasoning Solve tests? Start a free AI Mock Interview now, or compare plans on the pricing page to see what Pro unlocks.