Handling ambiguous problem statements
Part of the AI Engineer Roadmap
In short
"Use AI to improve our app." That's the whole brief. Interviewers say vague things on purpose, to see if you'll rush to build or first turn the fog into a clear, measurable plan.
Real AI problems rarely arrive well-specified — interviewers test whether you can turn a vague ask into a concrete, scoped, measurable project.
Think of it like this
The tailor's measuring tape
Ask a tailor for "a kurta" and a good one won't start cutting. Occasion? Fabric? Loose or fitted? By when? Measurements come before scissors. With an ambiguous AI problem, clarifying questions are your measuring tape, and they save you from stitching the wrong thing.
Watch it happen
From vague ask to a plan
1/6"What problem hurts most right now? Is it support load, churn or something else?" Say it back in one sentence.
The big ideas
What you need to know
Ask before assuming
Spend the first few minutes on questions. If they won't answer, state your assumptions clearly and carry on.
Make success measurable
Turn "improve" into a number with a target and a time frame.
Shrink to a first version
Pick the smallest version that tests the idea. Human-in-the-loop is a great safe start.
Say what's out of scope
Naming what you won't do in version one shows judgement and keeps the plan honest.
See it in code
Walk through the code
1Problem: Support team is overloaded.2Goal: Cut first-reply time, not headcount.3Users: Support agents (not customers) for v1.4Metric: Median first reply under 2 hours; satisfaction stays flat.5Limits: No customer data leaves our cloud. 6 weeks. 1 engineer.6v1: AI drafts a reply; an agent edits and sends it.7Not in v1: Auto-sending replies, voice support.8Risk: Wrong refund promises, so an agent must approve every reply.1/3The vague ask becomes a specific goal for specific users.
A one-page scope you can say out loud or write on the whiteboard. Each line answers one of the questions you asked.
Save yourself the pain
Common mistakes
Oops
Jumping straight to "I'd fine-tune an LLM" before knowing the actual problem.
Do this instead
Clarify goal, users and success metric first. The right solution might not even need a model.
Oops
Asking 20 questions and never moving forward.
Do this instead
Ask the few that change the design, state assumptions for the rest, and keep going.
Your mission
Scope three vague asks
Quick check
The interviewer says "Build something with AI to help our sales team." What's the best first move?
Go deeper
- 1Practice a structured approach: clarify the goal, define success metrics, scope the first version.
- 2Prepare an example where you had to make assumptions explicit before starting work.
- 3Avoid jumping straight to a solution — show the clarifying questions you'd ask first.
More behavioral at this stop
Explain your projects
Data & modeling trade-off discussions
Team collaboration
Communicating results to non-technical stakeholders
Stuck on this topic? Ask an Insider
Get 1:1 guidance from people who've walked this exact path — on the InsideEdge app.
