Accenture Case Interview Examples: 10 Practice Cases and the Official Workbook (2026)

Accenture case interview examples need to cover three formats that rarely appear together at one firm: the standard candidate-led case, the group case in the assessment centre, and the Potentia discussion used mainly on Strategy processes. The case content itself carries a heavy technology layer, so cloud migration payback, platform consolidation and AI deployment economics show up alongside classic profitability and market entry. Aggregated candidate reports put the process at 2.82 out of 5 for difficulty with a 68.4% positive experience rate across 12,362 submissions and an average of 29 days from application to offer, and the commonly recommended preparation load is 15 to 20 full cases before an assessment centre. Accenture also publishes its own case interview workbook, which teaches a six step solving loop and sorts every case into three shapes.

This page gives you 10 worked Accenture-style examples, a full group case walkthrough, a Potentia practice prompt, and a decoded summary of what that workbook actually asks you to do. For the process overview, rounds and track differences, see the Accenture case interview guide.

What should an Accenture practice case actually show?

A useful practice case is not a transcript. It shows the prompt, the structure you would open with, the evidence you would request, the arithmetic that decides the answer, and the recommendation with a number attached. Anything less leaves you rehearsing recognition instead of performance.

Work each example in this order: state the decision, build the tree, request the highest-value evidence, run the decisive calculation, then close with a recommendation and a risk. If your answer ends at "it depends on the data", you have described a framework rather than solved a case.

How Accenture cases differ from McKinsey, BCG and Bain

Candidates who prepare exclusively on MBB material arrive with the right analytical toolkit and the wrong expectations about format. The analytical bar is comparable in Strategy roles, but three structural differences change what you should rehearse.

The practical consequence is that an Accenture case rewards a structure that survives contact with implementation reality. When the client is migrating a core platform, "increase revenue, decrease costs" is not a structure, it is a category. The tree that scores names the migration decision points: what moves, what stays, what it costs to run both during the transition, and what breaks if the sequence slips.

Structure is also the one thing you cannot grade for yourself, because the tree you would have built always looks reasonable to the person who built it. Get one scored before you read the examples.

The five question types inside an Accenture case

An Accenture case is a sequence of separately scored asks rather than one continuous conversation. Naming them tells you what to drill in isolation, which is far faster than running full cases and hoping the weak spot surfaces.

Accenture's own materials lean on the same split. The workbook's three case shapes map almost one to one onto these asks: the great unknown tests structuring and estimation, the parade of facts tests triage, and the back of the envelope tests estimation arithmetic.

The Accenture group case interview, worked end to end

This is the format most candidates have never rehearsed, and the one where preparation pays back fastest because so few people practice it. In the assessment centre, four to six candidates receive the same business scenario, read individually, discuss as a group, and present a recommendation to assessors who are watching the whole time.

The shape of the exercise:

• Individual reading of the case pack, typically around 10 minutes
• Group discussion, typically 20 to 30 minutes, where the team structures the problem and converges on a recommendation
• Group presentation to assessors, typically 10 to 15 minutes, followed by questions
• Assessors score your individual analysis and your team behaviour at the same time

What actually earns points. Assessors are scoring contribution quality, not contribution volume. The highest-scoring moves are the ones that change the state of the discussion: proposing the structure the group then uses, catching a number nobody checked, resolving a stalled disagreement by naming the tradeoff, and summarizing where the group has landed before time runs out. The lowest-scoring behaviour is not silence, it is repeating a point someone already made in different words.

A worked group case scenario. A national grocery retailer with 480 stores is losing share to a delivery-first competitor. It has 90 days to decide whether to build its own delivery capability, partner with an aggregator, or defend in store. Your group has 25 minutes.

Here is how a strong candidate plays each phase:

1. Minute 0 to 2, claim the structure, not the airtime. "Before we debate options, can I propose we split this into three tests: can we serve the demand, can we make money on it, and can we do it in 90 days? Then we take each option through all three." A structure offered as a question gets adopted; a structure announced as a verdict gets argued with.
2. Minute 2 to 10, take the unglamorous branch. Volunteer for the economics rather than the strategy narrative. Whoever owns the numbers ends up owning the recommendation, because the group cannot conclude without them.
3. Minute 10 to 18, resolve rather than restate. When two people disagree about build versus partner, name the actual crux: "We are disagreeing about whether 90 days is real. If it is a hard deadline, build is off the table regardless of economics. Can we agree to treat it as binding and revisit if we have time?"
4. Minute 18 to 22, guard the time. "We have seven minutes and we still owe the presentation a recommendation and one risk. Should we lock the answer now and spend the last four minutes on how we present it?" Assessors notice who prevented the group from running out of time.
5. Minute 22 to 25, synthesize out loud. "So we recommend partnering with an aggregator for the first year: it hits the 90 day window, it costs a fraction of the build, and it buys us demand data. The risk we would watch is margin dilution, and we would revisit build if delivery passes 15% of volume."

That last move is the single highest-return behaviour in a group case, and it is the same skill an interviewer tests at the end of a solo case: compress a messy discussion into a decision, a reason and a risk. For the broader dynamics of participating without dominating, see the group case interview guide.

The Potentia interview, and a prompt to practice on

Potentia appears mainly in Accenture Strategy processes and is the format that breaks the most prepared candidates, because everything they trained on is unavailable. Reports describe roughly 45 to 60 minutes, about five minutes of solo preparation on a prompt, then a discussion. There is no arithmetic and no single correct answer. The interviewer is watching how you open a problem you have never seen, how you defend a position, and how you update when pushed.

What it is not. It is not a market sizing question with the numbers removed, and it is not a brain teaser. Treat it as a structured argument: a position, the two or three tests that would prove or break it, and an honest statement of what would change your mind.

A practice prompt. "A national government is considering requiring that all AI-generated media be labelled at the point of publication. Should it?" Give yourself exactly five minutes, then talk for 20.

A structure that works on almost any Potentia prompt:

1. Define the objective in one sentence. What is the policy actually trying to protect, and for whom? Naming trust in public information as the objective is a different case from naming consumer protection.
2. Name two or three tests. Does labelling change behaviour, can it be enforced at reasonable cost, and does it create a worse second-order effect such as implying that unlabelled content is verified?
3. Take a position early and hold it under pressure. Committing at minute three and defending gives the interviewer something to push on. Fence-sitting until minute eighteen gives them nothing to score.
4. Say what would change your mind. "If enforcement cost more than the harm it prevents, or if labelling measurably increased trust in unlabelled fakes, I would drop this and regulate distribution instead."

Because Potentia carries no math, the temptation is to prepare for it by reading. That does not work. It is a spoken format, and the gap is almost always delivery rather than ideas. Rehearse it out loud with a partner, or use the same discipline in a behavioral rehearsal where you have to hold a position and answer follow-up pressure in real time. Accenture Strategy processes usually pair it with a fit round, so build the story bank alongside it using the Accenture behavioral interview guide.

Accenture's official case interview workbook, decoded

Accenture publishes a case interview workbook that ranks at the top of search results for good reason: it is the firm's own statement of what it is looking for. It is worth working through in full, but here is what it actually teaches, because most candidates download it and skim the frameworks page.

The six step loop. Listen to the case, clarify the problem, decompose it, state hypotheses, test them, summarize findings. The two steps candidates skip are clarify and summarize, which are also the two the workbook spends the most time on.

The three case shapes. This is the most useful idea in the document and the part almost nobody quotes.

The soft criteria. Alongside analytical skill, the workbook names poise (handling pressure without unravelling), communication (making your reasoning audible), flexibility (updating when new information arrives), and intangibles such as energy, initiative and time management. These are not decoration. In a candidate-led format, an interviewer who cannot follow your reasoning has to score what they can see, which is your delivery.

The frameworks it lists are the classic set: profitability decomposition, the four P's, Porter's Five Forces, SWOT. Treat them as vocabulary rather than as answers. A recited framework is exactly what the workbook's "there is one right answer" misconception section warns against. If you want the version that adapts rather than recites, work through the case interview frameworks guide.

The practice cases inside it are largely estimation and diagnosis prompts of the era it was written in: a furniture retailer whose profits fall while sales grow, a bread division deciding whether to exit, a fast food franchise weighing an airport location, plus estimation prompts such as counting dry cleaners in a city, valuing a taxi medallion, and sizing savings on airline beverages. The business context is dated. The shape of the ask is not, and the estimation prompts in particular are still exactly what a back-of-the-envelope opener feels like.

Estimation is the skill that decays fastest without reps, and it is the one the workbook drills hardest. Build one driver chain against the clock and see where the assumptions get flagged.

10 Accenture-style practice cases with worked solutions

Work each one the same way: read only the prompt, cover the worked approach, write your structure for two minutes, then compare. The numbers in these cases are authored teaching data built to behave like real Accenture case material, not published client figures.

Case 1: Cloud migration payback for a regional bank

Prompt: A regional bank with $1.4B in annual revenue runs its core banking platform in two owned data centres. A new CIO wants to migrate to public cloud within three years. The board asks whether the migration pays for itself, and if so when.

How to drive it: This is a payback case wearing a technology costume. Propose the structure yourself: current run cost, target run cost, one-time migration cost, and the transition period where you pay for both. The fourth branch is the one candidates forget and the one that decides the answer.

The facts you would request:

• Current infrastructure run cost: $92M per year across the two data centres
• Target cloud run cost at steady state: $68M per year
• One-time migration cost: $140M, spread across three years
• Dual-running overlap: both environments live for 18 months at roughly 70% of legacy cost

Worked approach:

1. Steady-state saving is $92M minus $68M, so $24M per year.
2. Dual-running is the hidden cost. For 18 months you carry 70% of $92M, about $64M per year, on top of ramping cloud spend. Call the overlap penalty roughly $96M across those 18 months.
3. Total cost to get there is therefore about $140M of migration plus about $96M of overlap, roughly $236M.
4. At $24M per year of steady-state saving, simple payback from the point of completion is close to ten years. That is the answer the board did not expect.
5. Recommendation: the migration does not pay for itself on infrastructure economics alone at this scale. It is justifiable only if it unlocks revenue or risk benefits the CIO can quantify, such as faster product launch cycles or retiring a platform with a known regulatory exposure. Recommend re-scoping to migrate only the workloads with the worst run cost per transaction, which typically captures a large share of the saving for a fraction of the migration spend.

The trap in this case is not the arithmetic, it is remembering that the overlap exists at all. Technology cases are decided by run-rate and payback rather than by totals, and the candidates who lose lose on a missing branch rather than a slipped multiplication. Rep the same shape of calculation until the units stay attached.

Case 2: Platform consolidation for a global manufacturer

Prompt: A manufacturer operating in 14 countries runs 9 separate ERP instances after a decade of acquisitions. Finance close takes 21 days. Leadership wants a single instance. Where do you start, and what is it worth?

How to drive it: Resist the urge to structure this as a technology project plan. The question is what the fragmentation costs, so build the tree around the cost of the current state: direct IT cost of running nine systems, the labour cost of reconciliation, the working-capital cost of a slow close, and the decision cost of not having comparable data.

Worked approach:

• Direct cost: nine instances at roughly $4M each per year in licence, hosting and support is about $36M. A single instance at scale might run $16M, so the visible prize is around $20M per year.
• Labour: if 140 finance staff spend a third of their time reconciling across systems, at a fully loaded $85K that is roughly $4M per year of avoidable effort.
• Working capital: a 21 day close means the business steers on data that is three weeks stale. The value here is real but hard to quantify, so treat it as a qualitative argument rather than inflating the case with an invented number.
• Sequencing: consolidate the three largest instances first, since they typically carry the majority of transaction volume, and leave the smallest markets on their existing systems until the template is proven.
• Recommendation: consolidate in waves rather than in a big bang, target the three largest instances in the first 18 months, and hold the smallest four markets until the template is stable. Full consolidation as a single programme carries execution risk that the $20M annual prize does not justify.

What Accenture is testing here: whether you can hold implementation risk and financial value in the same answer. A recommendation that ignores sequencing reads as a candidate who has never seen a transformation programme slip.

Case 3: Market entry into embedded insurance

Prompt: A European property insurer is considering entering embedded insurance, selling cover inside the checkout flow of retailers and travel platforms rather than direct to consumers. Should it?

How to drive it: Standard market entry structure, with one Accenture-flavoured addition. Alongside market attractiveness, competitive position and economics, add a capability branch: does the client have the integration and pricing infrastructure to sit inside somebody else's checkout in under 200 milliseconds?

Worked approach:

• Attractiveness: embedded distribution converts far better than direct because the customer is already transacting. Assume the addressable premium pool is growing well above the core market.
• Competitive position: the incumbents here are technology-first insurers and the platforms themselves. A traditional insurer competes on balance sheet and regulatory licence, not on distribution.
• Economics: acquisition cost collapses because the platform owns the customer, but so does the margin, since the platform takes a commission that can reach a large share of gross premium. The question becomes whether volume at thin margin beats current volume at fat margin.
• Capability: this is the branch that usually kills it. Embedded requires real-time quoting APIs, automated underwriting, and claims that resolve without a call centre. A carrier with a 30 year old policy administration system cannot do this without rebuilding it.
• Recommendation: enter, but through a small number of deep partnerships rather than broad distribution, and treat the first partnership as an infrastructure investment rather than a revenue play. Revisit at 18 months against a single test: loss ratio on embedded business compared with direct.

The transferable skill here is deciding an entry mode out loud, under pressure, with the capability constraint in view.

Case 4: Profitability decline at a telecom operator (a parade of facts)

Prompt: A national mobile operator's EBITDA margin has fallen from 34% to 28% over three years. Subscribers are up 4%. The interviewer then gives you nine facts in ninety seconds.

The facts, as delivered: average revenue per user down from EUR 21 to EUR 18.40; handset subsidy spend up 30%; network capex up 22% on 5G rollout; a low-cost competitor launched 26 months ago; churn up from 14% to 19% annually; customer service cost per subscriber flat; roaming revenue down 40% after a regulatory change; retail store count down 12%; enterprise segment revenue up 9%.

How to drive it: This is the workbook's parade of facts. The scoring move is to say out loud which facts you are setting aside. "Store count and customer service cost look like consequences rather than causes, and enterprise growth is the one bright spot, so I want to park those three and work the revenue-per-user decline against the cost increases."

Worked approach:

1. Revenue side: ARPU fell 12.4%, from EUR 21 to EUR 18.40, against 4% subscriber growth, so revenue is down on a like-for-like basis despite more customers. That alone does not explain a six point margin fall.
2. Test the competitive story. Churn rising from 14% to 19% starting around the competitor's launch means the operator is buying customers back with price. ARPU decline and churn increase are the same event seen twice.
3. Cost side: handset subsidy up 30% is the operator paying to hold the subscribers it is losing. That is a retention cost masquerading as an acquisition cost.
4. Separate the regulatory hit. Roaming revenue down 40% is not a management failure and not addressable, so it should be quantified and set aside so the rest of the diagnosis is honest.
5. Recommendation: the margin problem is a defensive pricing response, not a cost control failure. Stop subsidising retention on the lowest-value segments, where the subsidy exceeds the lifetime value being defended, and redirect the spend into the enterprise segment that is still growing at 9%. Watch churn on the abandoned segment as the risk, and set a threshold at which you re-intervene.

For the general version of this diagnosis, see the profitability framework.

Case 5: Estimate the market for AI customer service agents (the great unknown)

Prompt: "How large is the annual US market for AI agents that handle customer service conversations?" That is the entire prompt. There is no exhibit and no follow-up until you ask for one.

How to drive it: Ask three clarifying questions before building anything: are we sizing software spend or the labour it displaces, are we counting only voice or also chat and email, and are we sizing today or at some future adoption level. The workbook's great unknown shape is scored on the questions, not on the number.

Worked approach, assuming software spend on voice and chat today:

• Start from the workforce. Assume roughly 3 million US customer service representatives.
• Not all of that work is automatable. Assume 40% of contact volume is routine enough for an AI agent to handle end to end, so the addressable pool is equivalent to about 1.2 million roles.
• Convert to spend. If a human role costs roughly $45,000 fully loaded and vendors price to capture around 20% of the saving, that is about $9,000 per displaced role of annual software value.
• 1.2 million times $9,000 is roughly $11B at full penetration.
• Apply adoption. At an assumed 15% penetration today, the current market is around $1.6B, growing quickly as penetration rises.

Sanity check: state one. A $1.6B software category growing toward $11B is plausible for an early-stage enterprise software market. If your number had come out at $200B you would be sizing the labour market rather than the software market, which is exactly the ambiguity your first clarifying question was supposed to resolve.

The driver-chain method behind this estimate is the same one the workbook's back-of-the-envelope cases assume you already own: a chain of assumptions each of which you can defend on its own, and a check at the end that the result is the right order of magnitude.

Case 6: Shared services productivity for a consumer goods company

Prompt: A consumer goods company runs a shared services centre handling accounts payable, payroll and vendor management for 22 markets. Cost per transaction is 40% above the benchmark its board was shown. Fix it.

How to drive it: Decompose cost per transaction into its parts: volume, labour hours per transaction, cost per hour, and rework. Then ask which of those the benchmark actually measured, because a benchmark comparing different scopes is the most common trap in an operations case.

Worked approach:

• Volume and mix: if the centre handles a long tail of low-volume markets with bespoke processes, the average is being dragged by markets that should not be in scope at all.
• Hours per transaction: check automation rate. If 55% of invoices are touched manually against a benchmark of 20%, that single gap can explain most of the difference.
• Rework: invoices that fail matching and return for manual handling can be counted twice in cost and once in volume, which inflates cost per transaction twice over.
• Cost per hour: location arbitrage is usually already exhausted. Chasing it is the answer candidates give when they have not found the real driver.
• Recommendation: attack automation rate and rework before headcount. Standardising the top five vendor formats typically lifts straight-through processing sharply, which reduces both hours and rework at once. Only after that does a location or headcount conversation make sense.

For the general structure, see the operations case interview guide.

Case 7: Repricing a managed services contract

Prompt: A technology services provider signed a five year managed services contract at a fixed annual fee of $60M. Two years in, the client's transaction volumes are 45% above the modelled baseline and the contract is losing money. What do you do?

How to drive it: Structure around three questions: how much money is it actually losing, what does the contract permit, and what is the relationship worth. Candidates who jump straight to "renegotiate" skip the question of whether they are allowed to.

Worked approach:

• Quantify: if delivery cost has risen from $52M to $71M against a fixed $60M fee, the contract is running at an $11M annual loss, and there are three years left, so roughly $33M of exposure.
• Contract terms: most managed services agreements contain a volume band and a change control mechanism. If actual volume sits outside the modelled band, the provider usually has a contractual route to reprice rather than a commercial argument to win.
• Relationship value: if this client also buys $90M of other services, an aggressive repricing that wins $11M can lose far more elsewhere.
• Recommendation: open the change control clause rather than a renegotiation, propose a volume-banded fee that restores margin above the modelled baseline while holding price flat within it, and offer an automation investment that lowers unit cost so the fix is not purely a price increase. That reframes the conversation from "we want more money" to "the contract assumed a volume that turned out to be wrong".

What is being tested: commercial judgment, not arithmetic. Accenture sells contracts like this one, so a candidate who understands that a contract is a set of levers rather than a fixed price stands out immediately.

Case 8: Digital customer experience for an airline

Prompt: An airline's mobile app has 12 million downloads but only 18% of bookings run through it. The chief digital officer wants that at 40% within two years. The interviewer hands you a funnel exhibit.

The exhibit, described: app sessions 100%, search performed 62%, fare results viewed 58%, passenger details started 24%, payment page reached 19%, booking completed 16%. Web funnel for comparison: search 71%, results 68%, details 41%, payment 34%, completed 31%.

How to drive it: Read the exhibit in the standard order: describe, interpret, quantify, then state the implication. The description is that both funnels lose people steadily, but the app loses a disproportionate share at one specific step.

Worked approach:

1. Describe: the app converts 16% of sessions to bookings against 31% on web, so the app is roughly half as effective per session.
2. Locate the break: from fare results to passenger details, web holds 60% of users while the app holds 41%. Every other step is broadly comparable. The gap is concentrated in one transition.
3. Quantify: if the app matched web's results-to-details rate, app completion would rise from 16% toward roughly 24%, which is half the distance to the 40% target from a single fix.
4. Interpret: a drop at passenger details usually means data entry friction, missing stored profiles, or a login wall appearing at the wrong moment.
5. Recommendation: fix the passenger details step first with stored traveller profiles and autofill, and treat the remaining gap to 40% as a separate acquisition problem rather than a conversion problem. State the risk: if the drop is caused by price shock rather than friction, autofill will not move it, so test before building.

Exhibit reading is a timed skill, and the failure is almost always spending the first 40 seconds describing rather than locating the break. Rep it against a chart you have not seen.

Case 9: AI deployment economics for an insurer

Prompt: An insurer wants to deploy an AI model that pre-assesses motor claims from photographs. The pilot handled 8,000 claims. Should it scale to all 900,000 annual claims?

How to drive it: This is a benefit-versus-risk case where the risk is quantifiable, which is unusual and is exactly why it is a good rehearsal. Structure it as: what does it save per claim, what does it cost to run, and what does it cost when it is wrong.

Worked approach:

• Saving: if the model resolves 35% of claims without a human assessor and an assessor visit costs roughly $180, the gross saving is 900,000 times 35% times $180, about $57M per year.
• Run cost: model inference, integration and monitoring at, say, $6M per year leaves roughly $51M.
• Error cost: this is the branch that decides it. If the model over-pays by an average of $90 on the 4% of automated claims it gets wrong, that is 315,000 automated claims times 4% times $90, roughly $1.1M. Small relative to the saving.
• The real risk is not the average error, it is the tail: systematic bias on a claim category, or a regulatory challenge to automated decisions. Say so explicitly.
• Recommendation: scale, but with a confidence threshold that routes low-confidence claims to humans, a sampled human audit of automated decisions, and a category-level bias review before full rollout. The economics are clearly positive; the governance is what makes it deployable.

Why this case is Accenture-shaped: the recommendation is not "yes" or "no", it is "yes, with the controls that make yes defensible". Implementation-aware answers score here in a way they might not at a pure strategy firm.

Case 10: A Potentia-style discussion prompt

Prompt: "In twenty years, will most people work for organisations at all?" You have five minutes to prepare. There is no data and no correct answer.

How to drive it: Do not hedge. Define what "work for an organisation" means, take a position, name the forces that would prove you right or wrong, and say what would change your mind.

Worked approach:

• Define: distinguish employment (a contract with one entity) from coordination (working inside a firm's system). Platform drivers are not employees but are heavily coordinated by a firm. That distinction is most of the answer.
• Position: employment as a legal form declines, coordination by firms does not. Firms exist because coordinating work through markets has transaction costs, and AI lowers those costs unevenly rather than uniformly.
• Tests: where do transaction costs fall fastest, does capital intensity still favour firms in the sectors that employ the most people, and what do regulation and benefits provision do to the boundary.
• Counter-position, stated by you before the interviewer raises it: healthcare, benefits and legal liability are sticky, and in most countries they are attached to employment, which slows the shift regardless of economics.
• What would change your mind: if benefits provision decoupled from employment at national scale, the decline would be much faster than I am predicting.

Notice there is no arithmetic anywhere in that answer, and it still has a structure, a position, a test, and a falsification condition. That is the whole job.

Self-diagnostic: are you practicing like an Accenture candidate?

After working the ten cases, grade yourself honestly on the five questions below. Each one maps to a rep you can run today rather than to vague advice.

1. Did your structure survive the technology detail? If your cloud, platform or AI cases collapsed into generic revenue and cost trees, the gap is structuring, not knowledge. Rep issue trees until the branches name decisions rather than categories.
2. In the parade-of-facts case, did you say out loud which facts you were dropping? Silent triage reads as missing the point. Naming it reads as command of the material.
3. Did every recommendation carry a number and a risk? "Consolidate the ERP estate" is a topic. "Consolidate the three largest instances in 18 months, worth about $20M a year, with sequencing risk as the thing I would watch" is an answer.
4. In the group case walkthrough, would your contributions have changed the discussion? Count how many of your imagined interventions proposed structure, corrected a number, resolved a disagreement or synthesized. Anything else was airtime.
5. Could you talk through the Potentia prompt for twenty minutes without notes? If not, the gap is spoken delivery, and reading more will not close it.

Official and free Accenture practice resources

Work the firm's own material first, then move to reps under pressure. Accenture publishes more usable preparation content than most candidates realise.

If your process includes an online assessment before interviews, that is a separate track with its own preparation, covered in the Accenture assessment test guide. For the firm context behind the cases, what Accenture does and what Accenture is are worth twenty minutes before a fit round, and the Accenture salary guide covers the numbers you will eventually negotiate against.

How to use these Accenture practice cases

Do not run all ten in a weekend. The value is in the diagnosis loop, and that requires a gap between attempt and review.

Execution checklist

• ✓

  Confirm your track and formats from the invitation. Potentia prep is wasted on a Consulting process, and skipping it is fatal on a Strategy one

• ✓

  Work cases 1, 4 and 5 first. One technology payback case, one parade of facts, one great unknown covers the three shapes the workbook names

• ✓

  Write your structure before reading any worked approach. Comparing your tree with the worked one is the only part of this that teaches anything

• ✓

  Grade one structure and one math rep with AI feedback. The two skills you cannot self-assess, available in the free drills

• ✓

  Run the group case walkthrough out loud with three or more people. Group dynamics cannot be rehearsed alone; recruit classmates or a prep community

• ✓

  Rehearse the Potentia prompt for twenty minutes with a partner pushing back. The failure mode is spoken delivery, not ideas

• ✓

  Run a full AI-graded case end to end. Reading examples proves recognition; a scored case proves performance, in the case library

• ✓

  Grade your resume before the round opens. Fix it once with the consulting resume grader rather than after a rejection

Then wire each weakness to a specific rep rather than running more generic cases:

• Structure came out template-shaped: run free structure drills until the tree names decisions.
• The estimate wobbled: work the market sizing method, then rep free sizing drills.
• Arithmetic slipped under time: use the math drill tool.
• The exhibit took too long: work reading charts and exhibits.
• The recommendation rambled: tighten it with the synthesis guide.
• Behavioral answers ran long: rehearse in the behavioral simulator and tighten the story bank.
• Application materials are not ready: the Accenture cover letter guide and the Accenture internship guide cover the rest of the funnel.

For worked examples across other firm formats, see case interview examples.

If your application is not in yet, the Accenture resume guide covers what the screen reads before you ever see a case.

Sources and further reading (checked July 31, 2026)

• Accenture careers, tips for the case-study interview: accenture.com/us-en/blogs/blogs-careers/secrets-to-a-successful-case-study-interview
• Accenture careers, pro tips for your job application: accenture.com/us-en/careers/explore-careers/area-of-interest/pro-tips
• Accenture careers, how to prepare for a behavioral interview: accenture.com/us-en/blogs/blogs-careers/how-to-prepare-for-a-behavioral-interview
• Accenture case interview workbook (FY19 edition, contents and practice cases): slideshare.net/slideshow/accenture-fy19caseworkbookoneaccentureconsulting/250424284
• Accenture case interview workbook (archived copy): scribd.com/document/554907549/Accenture-Case-Interview-Workbook
• Accenture interview process, aggregated Glassdoor statistics and assessment centre format: finalroundai.com/blog/accenture-interview-process
• Accenture round structure and Potentia timing reports: casebasix.com/pages/accenture-case-interview
• Accenture case interview preparation overview: joinleland.com/library/a/how-to-prepare-for-accenture-management-consulting-case-interviews