I did not recruit for consulting at Kellogg, but a lot of my close friends did, and I watched the whole thing from up close. Most of them are at Bain, McKinsey or BCG now. When I asked what the casebook was like, one of them gave me a line I have never forgotten: reading a casebook is like learning to drive in a parked car. You can learn where everything is. The real thing happens when the car is moving.
The moving car was the problem. A live case has a rush to it: the interviewer slides a chart across the table, your framework from minute three starts to wobble in minute twenty five, you drop a factor of twelve and have to find it out loud while the clock runs. You can only learn that by doing it, and the reps were always the scarce thing. Case partners have schedules and are not calibrated against a McKinsey scoring sheet. Coaches are, and cost enough that you get five sessions where you needed thirty.
AI should have fixed this, and on volume it has. On quality, most of it has not, because a voice chat with no exhibits that ends at fifteen minutes is still a parked car. We have run product management interviews on Tough Tongue AI for over a year and could have shipped a chat-style case mock any time. We did not, because it would not have felt like driving.
Instead we spent the year with candidates, coaches, and my consulting friends, getting the messy parts right: exhibits that appear on screen when the interviewer says so, an interviewer that makes you find your own arithmetic error, a push-back built from your own words, a report graded against the case’s answer key. The people who tested it told us it finally felt like the real thing, and that is why it is ready to share now.
This post covers what a case interview actually tests, how the current AI tools compare, and 29 full-length MBB cases you can run tonight.
What a case interview actually tests
A first-round case at McKinsey, Bain or BCG runs twenty five to forty minutes. The interviewer gives you a prompt, you ask a few clarifying questions and lay out a structure. Then the data starts arriving as exhibits: a chart or table the team “pulled together,” and a question about what you see. You read it, do the arithmetic out loud, get probed. There is usually a second quant block, a brainstorm, and a recommendation to the CEO that the interviewer pushes back on to see whether you hold your ground.
Three things about that format matter for how you practice.
The exhibit is the interview. Reading a chart under time pressure, catching the trap in a footnote, converting per-member-per-month to annual without dropping a unit. That is where candidates separate. A case with no exhibits has the hard part removed.
The arc is long. Nobody fails in minute five. They fail in minute twenty five, when the second calculation lands on a shaky framework, or in minute thirty five, when the interviewer says “Miami has fewer competitors and faster growth, why not Miami?” and they abandon a correct answer. A fifteen minute mock never reaches those places.
Feedback has to be calibrated against something. Firms score against a rubric, and every casebook case ships with an answer key. “That went fine” is not feedback. “Your commission expense came out at $15 million, the key says $150 million, you dropped a factor of ten going monthly to annual” is feedback.
Where the current AI case interview tools fall short
Almost every AI interview tool launched in the last two years is the same product underneath: a voice or text chat. That works for behavioral interviews and framework drills. It breaks for consulting on exactly the three points above.
No exhibits, or exhibits pasted into the transcript as text. The tool ends up narrating data at you (“revenue in 2019 was 50 units, up from 5 in 2015”) and the chart-reading skill never gets exercised.
Short sessions. Ten to fifteen minutes is the norm, partly because a pure conversation gets tedious past that without exhibits to carry it. Candidates tell me they get bored. I did too.
Generic feedback. With no answer key, the model grades frameworks generously, occasionally invents case rules, and corrects your arithmetic on the spot instead of making you walk through the calculation until you find the error yourself.
None of this makes the tools useless. It means you should know what each one is for. Here is how the main options compare, based on their own sites at the time of writing. Prices change, so treat that column as approximate.
| Tool | Format | Exhibits on screen | Full-length case (30+ min) | Feedback | Price (per their site, Sept 2026) |
|---|---|---|---|---|---|
| ChatGPT or Claude with a prompt | Text (voice mode optional) | No | Only if you push it | No rubric, grades generously | Free to $20 per month |
| PrepLounge | Case library, peer meeting board, AI casebot, human coaching | Casebook PDFs; casebot is text-first | Peer mocks yes; casebot no | Peer feedback; coach feedback paid | Free tier; premium and coaching extra |
| CaseCoach | Video course, sample interview videos, coaching | In the course materials | With a human coach | From ex-consultant coaches | Coaching from $265 per session |
| RocketBlocks | Timed drills (structure, math, charts) | Yes, as drills | No full case simulation | Drill scores | Subscription |
| CasePrepared | Voice AI | Some charts and calculations | Cases listed at about 10 minutes | Structure, quant, insight, communication | $26.99 to $39.99 per month |
| Case Study Prep AI | Voice AI | Not stated | Not stated | Structure, math, synthesis | $35 for 5 cases, $50 for 20 |
| Soreno | Text AI with drills | Yes, in text | Question bank, not timed full mocks | Rubric scores plus time-stamped notes | $199 per month |
| Tough Tongue AI, MBB collection | Voice AI interviewer driving a live slide deck | Yes, revealed per phase | Yes, 30 to 45 minutes with a time-managed arc | Graded against the case’s answer key, hire band, percentile | See the collection page |
A few honest notes on that table. PrepLounge is still the best place to find a human case partner, and CaseCoach’s video course is a good way to learn the twelve skills before you start casing at all. RocketBlocks’ chart drills are worth doing if exhibit reading is your weak spot. Soreno’s text feedback is more specific than most. What none of them do is the thing in the last row: an interviewer that runs a full-length case, controls the exhibits, and grades you against the key.
What we built instead
I sat down with the same friends, who have now done these interviews on both sides of the table, and we set out to build the moving car rather than another chat. Each case is a voice interview where the interviewer also drives a slide deck. When the case reaches the data, the interviewer says “let me pull up Exhibit A, you should see it on your screen now,” and the slide flips to the chart. The deck is gated per phase, so you cannot read ahead.
Every case is written from a real casebook and runs 30 to 45 minutes: an opening fit question, clarifying information you only get if you ask, one or two quant blocks, a brainstorm, and a recommendation with a push-back built from your own analysis. A conductor watches the clock, and with five minutes left it steers you to the recommendation whether or not you are ready.
Three behaviours that make the practice calibrated rather than merely available.
It does not correct your math. Get a number wrong and it asks you to walk through the calculation aloud until you find the error. It hands you the figure only after two failed nudges, and the report records that it had to.
It probes correct answers too. Land the right number and it still asks why, or what would change your ranking. Depth is scored separately from correctness.
It pushes back on the recommendation. Every case ends with one objection drawn from what you said earlier. Holding a correct position is part of the rubric; caving is noted.
Afterwards you get a report graded against the case’s actual answer key, the way firms grade: structure, quant accuracy, business judgment, communication, recommendation and risks. Your numbers sit next to the key’s numbers, each score is backed by a quote of what you said, and you get a hire band (strong hire, hire, borderline, no hire), your top three improvement areas, and a percentile against everyone else who has run the case. That is the calibration question answered.
29 full-length MBB cases you can practice now
Every case below is interviewer-led with live exhibits, and each one tells you the case type so you can pick what you need to work on. The collection lives at MBB Case Interview on Tough Tongue AI. Difficulty, where the source casebook lists one, is shown on the case page.
| Case | Source casebook | Case type | Industry | Practice |
|---|---|---|---|---|
| Avalon | Kellogg 2024 | Opportunity assessment | Energy and utilities | Run |
| Busch’s Barber Shop | Kellogg 2024 | Market entry | Hospitality | Run |
| Bubbly LLC | Kellogg 2024 | Product mix and market entry | Retail and CPG | Run |
| Chic Cosmetology | Kellogg 2024 | Market entry, breakeven | Education, hospitality | Run |
| Chicouver Cycle | Kellogg 2024 | New product, market entry | Transportation | Run |
| Dark Sky | Kellogg 2024 | New product, pricing | Aerospace and defense | Run |
| DigiBooks | Kellogg 2024 | Market entry | Media, retail | Run |
| Events.com | Kellogg 2024 | Profitability, pricing | Technology | Run |
| Garthwaite Healthcare | Kellogg 2024 | Profitability, cost reduction | Healthcare payer | Run |
| Health Coaches | Kellogg 2024 | Operations, program economics | Healthcare payer | Run |
| Healthy Foods | Kellogg 2024 | Growth strategy, chart interpretation | Food wholesale | Run |
| High Q Plastics | Kellogg 2024 | Profitability, plant P&L rebuild | Automotive parts | Run |
| Kellogg Capital | Kellogg 2024 | Growth strategy | Asset management | Run |
| Electro Chargers | Wharton 2024-25 | Market entry | EV infrastructure | Run |
| Happy Bellies | Wharton 2024-25 | Revenue growth | Food and beverage | Run |
| Smart Radiators | Wharton 2024-25 | Acquisition | Consumer electronics | Run |
| Aquador | Others | Market entry | Consumer goods | Run |
| Swift Enterprises | Others | Revenue decline and growth | Industrial | Run |
| Get-A-Ride | Others | Market entry, capacity | Mobility | Run |
| GDS Systems | Others | Cost reduction | Technology | Run |
| AG&E | Others | Compliance cost comparison | Utilities | Run |
| Hens Co. | Others | Acquisition screen | Agriculture | Run |
| Surfside | Others | Capacity expansion, operations | Leisure | Run |
| Polystore | Others | Program evaluation, chart interpretation | Retail | Run |
| Bushel Co. | Others | Outsourcing, operations | Agriculture | Run |
| Prytown | Others | Investment, valuation | Real estate | Run |
| Telcast | Others | Market entry | Telecom | Run |
| Burger Corp | Others | Acquisition | Fast food | Run |
| Pink Cross | Others | Profitability | Healthcare | Run |
If you want a suggested order: start with Aquador or Swift Enterprises to learn the rhythm, move to Electro Chargers or Avalon for a proper exhibit-driven market entry case, do Health Coaches when you want a units test (per member per month is where most people drop a factor of twelve), run Healthy Foods or Polystore for pure chart interpretation with almost no arithmetic, and save High Q Plastics and Surfside for when you want to be hurt a little.
A practice loop that works
The tool matters less than the loop you run it in. This is the one I give students:
- Run one medium case end to end. Full thirty minutes, no stopping when you get stuck. The first run is for finding out where you actually lose the interview, and it is rarely where you think.
- Read the whole report, not just the score. Your numbers next to the key’s numbers, the quoted evidence under each category. The top three improvement areas are your assignment for the next session.
- Redo the same case the next day. This is the step people skip and the one that produces most of the improvement. It tells you whether the fix stuck or whether you just knew the answer.
- Then one new case per day, mixing types. Alternate quant-heavy with chart interpretation, and interviewer-led with candidate-led. McKinsey and Bain will not feel the same in the room.
- Book a handful of human mocks for the last two weeks. An AI interviewer cannot read your body language or model the partner who will actually interview you. Use it to get to the bar first, so the scarce human hours go to polish rather than to discovering you cannot convert PMPM.
If you want a case that is not in the collection yet, you can connect Claude or ChatGPT to Tough Tongue over MCP, paste in a case from any casebook, or a real project from your own industry, and generate an interviewer for it. The setup is on the agents page.
Where AI case practice still has limits
I want to be plain about this because the enthusiastic version overclaims and the consulting-prep crowd will notice.
An AI interviewer cannot judge presence. Whether you seem like someone a partner would put in front of a client is a human judgment, and it is a real part of the decision. Practice with people for that.
Firm-specific cadence is only partly captured. The cases in this collection follow the interviewer-led McKinsey style, converted from casebooks that include candidate-led originals. If you are recruiting for Bain or BCG, practice driving the case yourself as well, and use the McKinsey-style sessions for the exhibits and the math.
And a report against an answer key is only as good as the key. Casebook cases sometimes have quirks (Garthwaite Healthcare uses 500,000 agents in one part and 500,000 members in the next, and the interviewer is told to let it go), and the report will follow the casebook where the casebook is odd. When you disagree with a score, the transcript and the quoted evidence are there so you can check.
None of that changes the main point. For the parts of a case interview that are learnable by repetition, which is most of it, the bottleneck has been calibrated practice at volume, and that bottleneck has moved.
Start here
Pick a medium case from the table above, block thirty five minutes, put pen and paper next to you, and run it like it counts. It will feel messier than reading the casebook did. That is the point. Then read the report and run it again. If you are prepping for a specific firm, McKinsey, Bain or BCG, tell me in the comments on the video which one, and I will point you to the right cases first.
The full collection is at MBB Case Interview on Tough Tongue AI, and the rest of the interview courses are on the courses page. For how the platform handles interactive tools beyond a conversation, the post on AI interview platforms goes deeper, and the interview prep solutions page covers the other interview types.
