Using AI in Research, Synthesis & Slide Drafting
A blank slide deck at 11:40 p.m. feels harmless until the AI gives you a polished-looking answer built on weak sources, vague logic and one invented market fact. The real skill is not “using ChatGPT” - it is knowing where AI speeds up thinking, where it corrupts thinking, and how to keep human judgment in control.
- Use AI after the problem is defined, not before. A bad question produces confident noise faster.
- The safe workflow is: scope - source - extract - synthesize - storyboard - verify.
- Research output is not insight. Insight explains what matters, why it matters, and what to do next.
- Use AI for first drafts, comparison tables, theme clustering, counterarguments and slide titles.
- Do not use AI as the final source of truth. Every specific claim needs a primary or credible source.
- The best slide prompt gives context, audience, decision, constraints and output format.
- Before presenting, check: source coverage, factual accuracy, MECE structure, actionability and slide clarity.
Big Picture
AI is most useful in consulting-style work when it sits between raw information and your final recommendation. Think of it as a tireless analyst that can read, cluster and draft - while you own the problem definition, judgment, trade-offs and final answer.
The Core Model: Use AI as an Analyst, Not as the Partner
In a consulting team, the analyst does not decide the client recommendation alone. The analyst gathers facts, cleans inputs, creates exhibits, tests hypotheses and drafts pages. AI should be treated the same way.
That means you give AI bounded work: “summarize these three annual report extracts,” “cluster these customer complaints,” “draft three possible slide titles,” or “challenge this recommendation.” You do not ask it, “What strategy should this company follow?” and copy the answer.
Definitions You Should Be Able to Say in One Breath
- AI-assisted research: using AI to locate, summarize and compare information while humans decide relevance, truth and implication.
- Synthesis: turning many observations into a smaller set of judgment-led insights that answer the business question.
- Slide drafting: converting the answer into a structured storyline, page messages and visual exhibits for decision-making.
- Prompt: an instruction containing context, task, constraints, output format and quality criteria.
- Grounding: tying AI output back to supplied documents, cited sources or verified data instead of model memory.
The Six-Step AI Workflow for Research, Synthesis and Slides
If you remember only one process, remember this one. It prevents the two classic failures: starting with AI before the problem is clear, and ending with AI before facts are checked.
This workflow pairs especially well with case-solving because it forces you to define the problem before analysis. If that first step feels weak, revise defining the problem before solving it before trying to automate the rest.
What AI Is Good At - and What You Must Still Do Yourself
The trap is to treat all thinking tasks as equal. They are not. AI is excellent at high-volume language work. It is weaker when the task needs accountability, fact ownership, client judgment or trade-off decisions.
The AI Task-Risk Matrix
Use this simple 2x2 before assigning work to AI. Low-consequence, low-ambiguity tasks are safe to automate heavily. High-consequence, high-ambiguity tasks need human judgment first and AI second.
Prompt Templates That Produce Consulting-Style Output
A weak prompt says, “Make slides on the industry.” A useful prompt gives the AI a role, context, source boundary, output structure and quality bar.
Role: Act as a consulting analyst. Context: We are advising a mid-sized Indian consumer brand on whether to enter quick commerce. Task: Synthesize the attached notes into 3 strategic implications. Constraints: Use only the provided notes; flag missing data; do not invent numbers. Format: Give a one-line answer, 3 supporting bullets, and a slide title for each implication. Quality bar: Make the points MECE, decision-oriented and evidence-backed.
Notice the difference: the prompt does not ask AI to “be smart.” It gives AI the ingredients needed to produce usable consulting output.
Practical Quality Measures Before Output Leaves Your Laptop
You do not need a complicated AI governance dashboard for interview prep or classroom cases. But you do need visible checks. Use these measures before turning AI-assisted work into a submitted deck or spoken recommendation.
Example - Using AI for an Indian Market Scan
Suppose you are studying Blinkit, Zepto and Swiggy Instamart for a market-entry case. A good AI workflow would summarize public articles, app observations and company disclosures into themes such as assortment, delivery promise, dark-store density, unit economics and customer occasions. The strategic “so what” is not “quick commerce is growing”; it is which operating capability creates advantage and what a new entrant would need to match.
The primary driver in this kind of analysis is the operating model behind speed - inventory placement, store density, picking process and delivery routing. Supporting drivers include consumer habit formation, category mix, supplier terms, app experience and funding discipline. If your AI output gives only a demand-side story, it is incomplete.
Case Study - Infosys Topaz and the Industrialization of AI-Assisted Work
Infosys launched Topaz as an AI-first set of services, solutions and platforms in 2023, making it a useful Indian example of how AI is being embedded into professional services delivery (Infosys press release, 2023).
Situation: Large technology and consulting firms were seeing clients experiment with generative AI, but isolated pilots were not enough. Enterprise clients needed reusable methods, platforms, governance and domain-specific use cases - not just a chatbot demo.
The move: Infosys positioned Topaz as a structured AI-first offering rather than a one-off productivity tool. The primary driver was embedding generative AI into service delivery and client solutions. Supporting drivers included reusable platforms, domain knowledge, engineering capability, data foundations and responsible-use guardrails.
The lesson: The winning move in AI-assisted consulting work is not “use AI everywhere.” It is standardizing where AI helps - research extraction, code or content generation, workflow automation, knowledge reuse - while keeping accountability with humans. For an MBA student, the same principle applies at a smaller scale: build a repeatable workflow, not a random prompt habit.
How AI Changes Research, Synthesis and Slide Drafting
By 2026, AI is changing this workflow in three concrete ways.
Practical student workflow: Load your case facts, company annual report extracts and class notes into NotebookLM. Ask it to generate: 1) five likely interviewer questions, 2) a source-grounded issue tree, 3) three possible recommendations with risks, and 4) a list of claims that still need verification. Do not upload confidential client data, placement test material under NDA or private company documents.
If you want the broader career implication, revise how AI is changing consulting roles, pyramids and pricing after this topic.
Interview Relevance
“Suppose you have to prepare a competitor landscape and 5-slide recommendation deck by tomorrow morning. How would you use AI, and how would you control quality?”
In a consulting interview, say explicitly: “I would not let AI be the source of truth. I would use it to accelerate extraction and drafting, then verify claims and own the recommendation.” That one sentence signals maturity.
Common Mistake
The single biggest mistake is asking AI for the answer before defining the problem. It costs candidates because they produce a generic, confident recommendation that ignores the client decision, constraints and economics. One-line fix: write the decision question, success metric and source boundary before you prompt.