AI Use Cases in Consulting: How to Structure Them in Case Interviews and Client Work

In McKinsey's June 2023 research across 16 business functions, 63 generative AI use cases can add $2.6 trillion to $4.4 trillion in annual economic value, with about 75% of that value concentrated in customer operations, marketing and sales, software engineering, and R&D. Sizing an AI investment in a consulting case interview follows the same steps as scoping a client pilot: tie a specific business challenge to a measurable cost or revenue outcome, test whether the operating model can scale it, and account for accuracy, privacy, and IP risks. Practice a case free, no sign-up, to rehearse your structure and math out loud before your interview.

Updated

What a consulting AI use case actually is

McKinsey defines a use case as a targeted application of generative AI to a specific business challenge that results in one or more measurable outcomes. When tackling AI questions in interviews, you can avoid common pitfalls by not describing the technology in abstract terms. Focus directly on isolating the exact task, the affected cost base, and the operational outcome.

Consulting AI use cases cover two connected areas: how consulting firms use tools internally to deliver work, and how you structure AI adoption cases in interviews. Internally, McKinsey equips teams with training and tools like its proprietary platform Lilli and ChatGPT Enterprise, while BCG studies highlight how consultants use AI for task drafting, process redesign, and knowledge retrieval. On client projects, the work divides into function-specific use cases and broad knowledge worker productivity, such as querying stored corporate records in natural language.

  • Internal knowledge management: querying corporate libraries in natural language to reduce time spent gathering information. In 2012, the McKinsey Global Institute estimated knowledge workers spend about a fifth of their time, or one day each work week, searching for information.
  • Task-level versus process-level gains: BCG research in March 2026 separates single-task gains (such as drafting a post, up 26% to 36%) from full-process gains (such as coordinating a campaign, up 34% to 47%).
  • Generative AI versus traditional AI: McKinsey notes traditional analytics and machine learning still drive $11.0 trillion to $17.7 trillion in annual value because they handle numerical optimization, while broader generative AI worker productivity adds $6.1 trillion to $7.9 trillion net of overlap.

Where the value sits: 4 functions drive 75% of impact

When an interviewer asks where a company should look first for generative AI value, focus on unstructured data like text, code, and images. Across the 16 business functions in McKinsey's report, about three-quarters of total annual value sits in customer operations, marketing and sales, software engineering, and R&D.

By our calculation from McKinsey's report tables, the $2.6 trillion to $4.4 trillion organizational use case range represents about 15% to 17% of the total $17.1 trillion to $25.6 trillion global AI economic potential when combined with traditional machine learning and broader worker productivity. Supply chain and manufacturing rank lower for generative AI alone because their main value drivers rely on numerical optimization.

  • Customer operations: automating self-service inquiries with chatbots, retrieving customer history during contact, suggesting real-time responses to human agents, and summarizing calls for coaching.
  • Marketing and sales: drafting advertising copy, slogans, and personalized emails, translating campaigns, synthesizing unstructured social feedback, optimizing SEO tags, and personalizing search.
  • Software engineering: pair programming, code correction and refactoring, root-cause log analysis, automated test generation, and IT architecture drafting.
  • Product R&D: early research analysis, generative virtual design of candidate molecules or materials, and physical test planning paired with deep learning surrogates.

Functional benchmarks from McKinsey and BCG research

Functional benchmarks from McKinsey and BCG research
Function or areaMeasured productivity or cost impactKey operational metrics in the official sources
Customer operations (McKinsey)30% to 45% of current function costsReduces human-serviced contacts by up to 50%, where roughly half of North American banking, telecom, and utility contacts are already handled by machines. In one study of 5,000 agents, issue resolution rose 14% an hour, handling time fell 9%, and agent attrition and manager escalation requests fell 25%.
Software engineering (McKinsey)20% to 45% of current annual function spendDriven by initial code drafts, refactoring, and root-cause analysis. One cited study found developers using Microsoft's GitHub Copilot completed tasks 56% faster.
Marketing and sales (McKinsey)High impact on addressable marketing spendEvaluated against marketing spend excluding ad buying costs, because generative AI does not change media space buying rates.
Corporate affairs and comms (BCG)26% to 36% task-level, 34% to 47% process-levelRanks among the top two enterprise functions for GenAI transformation upside, with potential productivity gains of roughly one-third or more across tasks and processes.
Industrial operations and manufacturing (BCG-WEF survey)14% savings on addressed manufacturing costsSurvey of nearly 1,800 manufacturing executives across 7 industries and 15 nations: 89% plan to implement AI, 68% have started, and early adopters saw 14% savings. Amazon achieved 25% faster delivery, 30% more skilled roles, and 25% efficiency gains. Foxconn cut deployment time by 40% and operational costs by 15%.
Industry spotlights: banking, retail, pharma, and industrial operations

In a case interview, standard functional lists must be fitted to the client's sector. McKinsey stresses that there is not one logical solution and instructs candidates to listen carefully and share solutions built for the client rather than standard frameworks.

Official research highlights distinct industry applications:

  • McKinsey notes a European bank built an ESG virtual expert to extract answers from unstructured documents, while Morgan Stanley built an assistant using GPT-4 to help tens of thousands of wealth managers query internal research.
  • Industrial operations and robotics (BCG-WEF): layered automation combines rule-based robotics for repetitive tasks like welding, training-based robotics via reinforcement learning for variable tasks like adaptive kitting, and context-based robotics using foundation models for zero-shot execution through natural language.

What McKinsey, BCG, and Bain look for in a case interview

When an AI topic appears in an interview, firms evaluate your structured reasoning, not technical trivia about neural network layers. Review the firm-specific interview guides before testing yourself: the McKinsey case interview guide, BCG case interview preparation, and Bain case interview examples.

McKinsey calls its case the Problem-Solving Interview. Its goal is to see how you structure tough, ambiguous business challenges, identify important issues, handle facts and data, make a recommendation and explain your thinking in a fast-moving discussion. McKinsey warns candidates against relying on standard frameworks.

BCG describes the case interview as a realistic business challenge where you work through the problem step by step: structure your approach, ask thoughtful questions, analyze data, perform quick calculations, and identify key drivers. BCG notes there is not always a single right answer and that reasoning quality matters most. Its guidance advises candidates to listen actively, think structurally, clarify assumptions, communicate clearly, show their thinking, and avoid staying silent or rushing.

Bain describes the case interview as a structured conversation where you work through a real business problem. Bain looks for how you think, structure problems, and build on ideas as the conversation evolves, rewarding logical structure and creativity rather than a single set answer.

A 4-branch framework for AI use case questions

Checklist and framework written by our team, built from the evaluation lenses in McKinsey's and BCG's published AI research. Adapt these four branches when analyzing an AI investment, prioritizing use cases, or troubleshooting a stalled deployment:

  • 1. Targeted workflow and measurable outcome: Identify the exact task being automated or sped up. Measure cost savings against the addressable cost base (excluding unaffected expenses like media ad purchases) or project revenue gains. Verify task-level productivity before redesigning an entire process.
  • 2. Economic feasibility versus baseline labor: Compare software development, compute, and licensing fees against human labor costs for that occupation and location. McKinsey notes that technical lab feasibility does not guarantee economic feasibility if tool costs exceed baseline labor.
  • 3. Operating model and scaling foundations: Pinpoint why pilots stall. In BCG's industrial AI survey, 98% of companies faced scaling challenges and only 16% reached AI targets, with 88% citing tech infrastructure gaps and 92% citing people and organization gaps (39% digital skills shortage, 33% lack of strategy). In BCG's CCO survey, 68% called their function a laggard and 35% cited lack of operating model design capability. Check for workflow redesign, dedicated funding (leading comms teams are 2.3 times more likely to allocate at least 10% of budget to AI), and training (71% of leading teams upskilled at least a quarter of staff, versus 22% of laggards).
  • 4. Risk, governance, and human-in-the-loop controls: Address hallucinations by requiring data source citations and human reviews. Guard against intellectual property infringement, data privacy breaches, and algorithmic bias. In BCG's June 2026 research, uniform AI outputs reduced group diversity of thought by 41%, and 14% of surveyed US workers using AI reported mental fatigue ('AI brain fry'), ranging from 6% in legal to 26% in marketing.

Worked example: sizing a customer service AI case

Worked example written by our team. The client and figures are invented for practice, modeled on the customer operations mechanics in McKinsey's report. For more practice formats, review our consulting case study examples.

Prompt: Your client is NorthStar Telecom, a North American operator with 2,000 customer care agents. Fully loaded agent compensation is $50,000 per year. NorthStar receives 16 million customer contacts annually. Half (8 million) are handled by machines, and 8 million are handled by human agents. Management is reviewing a generative AI package that provides a self-service chatbot and a real-time agent assistant. The vendor charges a $15 million upfront implementation fee and $5 million per year in licensing and compute. NorthStar requires technology investments to pay back within 12 months. Should NorthStar invest, and how should it deploy the tool?

Step 1: Clarify scope and baseline. Ask what drives agent capacity and how the software performs. The interviewer explains: (a) the chatbot deflects 25% of contacts currently handled by humans, and (b) the assistant reduces handle time so an agent resolves 20% more contacts per year. However, a pilot showed the assistant improves junior agents (60% of staff) but slows senior agents (40% of staff) if they must follow its prompts.

Step 2: Contact volume calculation. Today, 2,000 agents handle 8,000,000 human contacts, which is 4,000 contacts per agent annually, representing $100,000,000 in total labor (2,000 x $50,000). The chatbot deflects 25% of those 8,000,000 contacts (2,000,000 contacts), leaving 6,000,000 human-serviced contacts.

Step 3: Agent capacity with segmented rollout. McKinsey's cited research demonstrates that generative AI assistants benefit lower-skilled workers while occasionally reducing metrics for experienced staff. Segment the rollout: senior agents keep handling 4,000 contacts per year without the assistant, while junior agents use the assistant to increase output by 20% to 4,800 contacts per year. Keeping all 800 senior agents (40% of 2,000) covers 800 x 4,000 = 3,200,000 contacts. The remaining 2,800,000 contacts require junior agents: 2,800,000 / 4,800 = 583.33, or 584 junior agents. Junior headcount drops from 1,200 to 584, a reduction of 616 agents.

Step 4: Net savings and payback. Reducing junior headcount by 616 agents saves 616 x $50,000 = $30,800,000 per year in gross labor. Subtract the $5,000,000 annual software fee to reach $25,800,000 in net annual savings. Dividing the $15,000,000 upfront cost by $25,800,000 gives a payback period of 0.58 years (about 7 months), well within the 12-month requirement.

Rules on using AI during your consulting prep and interviews

When you use AI to prepare, understand firm policies regarding live assessments. McKinsey states that it welcomes curiosity about AI and views it as a valuable preparation tool when used responsibly, comparable to getting feedback from mentors or friends. Misrepresenting yourself or using tools to generate live responses violates its Assessment Integrity Expectations.

Official dos and don'ts from McKinsey's guidance:

  • Polishing your resume: Encouraged.
  • Practicing interview questions: Encouraged.
  • Asking AI to explain concepts or frameworks during prep: Encouraged.
  • Exaggerating or creating achievements that are not yours: Not allowed.
  • Generating real-time answers during an interview: Not allowed.
  • Using AI during assessments unless specifically permitted: Not allowed.
  • Virtual interview setup: have pen, paper, charger, and water ready, keep your camera on without virtual backgrounds or blur, and disable AI note-taking tools or virtual assistants so they do not record or assist.

How to rehearse AI cases out loud before your interview

Reading framework notes will not prepare you for live interview dialogue. McKinsey notes that many candidates find it helpful to practice aloud with family or friends to build a conversational style and speak clearly in fast-moving discussions.

Rehearse both technical case logic and personal experience questions. For McKinsey's Personal Experience Interview (PEI), prepare two personal examples across Personal Impact, Entrepreneurial Drive, and Inclusive Leadership, focusing on your specific role and leading with your reasoning. You can also review consulting fit and behavioral questions to practice out loud.

You can see a real sample report or practice a case free, no sign-up, on Soreno to rehearse structuring AI business cases and answering interviewer follow-up questions.

  • Step 1: Pick an industry (such as banking or retail) and give yourself two minutes to sketch a four-branch structure on paper.
  • Step 2: Deliver your opening structure aloud in under 90 seconds, covering workflow, measurable impact, operating foundation, and governance controls.
  • Step 3: State your initial hypothesis and explain where you want to drill down first to test economic feasibility.

From McKinsey & Company: The economic potential of generative AI: The next productivity frontier (June 2023), McKinsey & Company: Interviewing at McKinsey, McKinsey & Company: Getting ready for your interviews, BCG Careers: Case Interview Preparation, BCG Careers: Consulting Interview Process, BCG-WEF Project: AI-Powered Industrial Operations, BCG: Corporate Comms Is Playing Catch Up on AI (March 23, 2026), BCG: AI for CEOs: Amplifying Time and Judgment at the Top (June 23, 2026), Bain & Company: Case interview preparation and Bain & Company: Interviewing and Hiring Process.

1 practice question and sample answer

Say your answer out loud first, then read the sample answer.

Practice question

  1. Your client is the CEO of a large retail bank. She has received 20 different generative AI project proposals from across the bank and wants to know how to pick two or three use cases to fund this quarter and how to make sure they actually scale. How would you structure this for her?

    Sample answer. Example answer written by our team. The applicant is invented. I would structure this decision around three areas: first, value potential by function, second, technical and economic feasibility, and third, the operating model needed to scale beyond a pilot. First, on value potential, I would group the 20 proposals by the business challenge they solve and the measurable outcome they target. In retail banking, the biggest generative AI gains usually sit in customer operations, marketing and sales, software engineering, and risk documentation. For example, I would look closely at an internal knowledge assistant for frontline advisors, legacy code migration, or automated IVR for lost credit cards. Second, on feasibility, I would compare the implementation and compute costs against the specific baseline labor spend each tool addresses, and check data privacy and regulatory compliance so customer data never leaks. Third, to make sure the winners actually scale, I would check the operating model before writing a check: we need task-level productivity proof first, human-in-the-loop source verification to catch hallucinations, and dedicated budget for employee upskilling and workflow redesign. If you agree, I would love to start by looking at what functions the 20 proposals sit in.

    It avoids generic textbook lists and gives a three-part structure built for a retail bank CEO. It names concrete banking use cases from McKinsey's research, separates task-level proof from full-process scaling as highlighted in BCG's research, builds in human-in-the-loop risk checks, and ends by proposing where to take the conversation next.

Practice the case interview, out loud

Soreno's AI interviewer runs a full case one question at a time, checks your math and scores your structure, any time you have 30 minutes.

Questions people ask

What is the official definition of an AI use case in consulting research?

In its report on generative AI, McKinsey defines a use case as a targeted application of generative AI to a specific business challenge that results in one or more measurable outcomes, such as lower production costs or higher revenue.

Which corporate functions hold the majority of generative AI value potential?

Across the 16 business functions analyzed by McKinsey, about 75% of total annual value sits in customer operations, marketing and sales, software engineering, and R&D.

Should you rely on pre-packaged frameworks during consulting interviews?

No. McKinsey advises candidates to listen carefully and share solutions built for each client rather than standard frameworks. BCG and Bain similarly emphasize that there is no single right answer and that interviewers evaluate the quality of your reasoning.

Can I use AI tools to prepare for a McKinsey or BCG case interview?

Yes, for prep. McKinsey encourages candidates to use AI to polish resumes, practice questions, and clarify concepts, but strictly prohibits unapproved AI tools, real-time response generators, or note-taking bots during live interviews.

What core operational risks must be accounted for when evaluating AI use cases?

McKinsey and BCG highlight hallucinations requiring human verification, IP infringement, customer data privacy leaks, training data bias, reduced diversity of thought (by 41% in BCG research), and cognitive fatigue ('AI brain fry', reported by 14% of surveyed US workers).

Why do most corporate AI implementations fail to scale beyond pilots?

BCG studies show that scaling bottlenecks stem primarily from people and organization gaps, such as digital skill shortages and undefined strategies, alongside tech infrastructure deficits and missing operating model designs.

Keep going