The 'Improve This Product' Question, Reverse-Engineered

When an interviewer asks how you'd improve this product, they aren't fishing for clever feature ideas — they're testing whether you can think like an owner. The strongest answers pick one user segment, name that segment's goal, isolate the single biggest friction blocking it, and only then propose ideas tied to a measurable outcome.

Quick Answer: Don't list features. Pick a user segment, state their goal, find the friction blocking it, then propose 2-3 ideas — each mapped to one target metric. Structure beats cleverness every time.

The Anti-Pattern: Why Feature Lists Fail

Most candidates fail this question by rattling off features — dark mode, a referral program, faster checkout — with no shared thesis connecting them. Interviewers hear a list like this as evidence of random-idea generation, not prioritization, because nothing in it explains why these ideas and not fifty others.

This question shows up constantly in PM interviews, from APM screens to director loops, precisely because it is cheap to ask and brutally revealing to answer. It is a close cousin of the broader product sense interview framework, and candidates who've internalized that framework's discipline — segment first, problem second, solution last — rarely fall into the feature-dump trap here.

What the Feature-Dump Answer Signals

An unstructured answer doesn't just sound weak — it actively signals specific gaps to a trained interviewer:

• No segmentation instinct. Treating "users" as one undifferentiated blob instead of segments with different goals and pain points.
• Solution-first thinking. Jumping to features before naming the problem those features would solve.
• No prioritization logic. Presenting five ideas as equally good, with no criteria for ranking them.
• Vague success criteria. Justifying ideas with "it would improve engagement" instead of a specific, falsifiable metric.

The table above isn't academic — it's a checklist. If your draft answer matches column one anywhere, rewrite before you say it out loud. For a fuller map of how this question sits alongside other advanced PM interview formats, the complete guide to advanced PM career questions is worth reading end to end.

The Winning Structure: Segment, Goal, Friction, Then Ideas

The structure that consistently scores well has four steps, in a fixed order: pick a user segment, clarify their goal, find their biggest friction, then ideate and prioritize. Skipping straight to step four — even with brilliant ideas — is the single most common reason a technically strong answer still lands flat.

1. Pick a user segment. Not "all users." Choose a specific, nameable slice — new sign-ups in week one, power users who've plateaued, lapsed users who churned after a bad first session. Say why this segment matters (size, growth potential, or strategic value) in one sentence.
2. Clarify their goal. What "job" is this segment hiring the product to do? This is exactly the territory of Jobs to Be Done — a framework Clayton Christensen popularized at Harvard Business School — and if you haven't internalized JTBD's push/pull/anxiety/habit framing, the complete guide to Jobs to Be Done is the fastest way to borrow that vocabulary before an interview.
3. Find the biggest friction. Walk the segment's actual journey and find where it breaks down — not where you assume it breaks down. A structured walk of the customer journey, stage by stage, is how you avoid guessing.
4. Ideate, then prioritize. Generate 3-5 ideas that address the friction directly, then rank them using an explicit framework like RICE (Reach, Impact, Confidence, Effort — a prioritization method popularized by the product team at Intercom) or Kano (basic, performance, delight). Present only your top 1-2 in depth; mention the rest exist and why they ranked lower.

Why Order Matters More Than Content

Interviewers can usually tell within thirty seconds whether you're following this order or improvising. An answer that opens with "let me pick a user segment first" immediately signals structured thinking, even before the content of the answer is any good. That framing device alone raises the floor of the response.

Conversely, an answer that opens with an idea and retrofits a user and metric onto it afterward reads as post-hoc rationalization — because it is. Interviewers have heard hundreds of these; the seams show.

Connecting Ideas to Metrics, Not Vibes

Every idea you propose needs one target metric it would move, chosen before you pitch the idea — not invented afterward to sound rigorous. A metric with no idea is a wish; an idea with no metric is a guess. The discipline is picking the metric first, then checking whether your idea plausibly moves it.

Two real, well-known frameworks make this easier than inventing metrics from scratch:

• HEART (Happiness, Engagement, Adoption, Retention, Task success), developed by Kerry Rodden and colleagues at Google, maps cleanly onto UX-flavored improvement ideas — a friction fix usually targets Task success or Engagement specifically, not "UX" broadly.
• Leading vs. lagging indicators. Retention is a lagging metric — it moves slowly and is hard to attribute to one change. Pair it with a leading metric closer to your idea (e.g., time-to-first-value, or completion rate of a specific step) so you can argue the idea would show results in weeks, not quarters.

Avoid metrics you cannot plausibly explain the causal chain for. If your idea is "simplify onboarding step 3," the metric is "step 3 completion rate" or "time-to-first-value" — not "overall revenue," which is three inferential leaps away and impossible to attribute cleanly.

Directionally, research bodies like the Baymard Institute have long documented checkout and onboarding abandonment in the range of 60-70% for many flows — a reminder that the biggest friction is often earlier and more boring than the exciting feature you first thought of.

The Most Common Metric Mistake

The most common mistake at this step isn't picking the wrong metric — it's picking a metric one level too far from the idea. A "North Star metric" like weekly active users is a fine company-wide compass, but it's a poor evaluation criterion for a single friction fix, since dozens of other factors move it too. Anchor your idea to the nearest metric it directly touches, and mention the North Star only to show you understand how that near-term metric ladders up.

Worked Example: Narrowing Scope Before Proposing Anything

The fastest way to lose points on this question is starting broad ("let's improve retention for all users") and never narrowing. A strong answer narrows aggressively in the first thirty seconds, then stays narrow through every subsequent step. Below is a full walkthrough using a note-taking app as the product.

Step 1 — Segment. Not "all users of the note-taking app." Specifically: people who signed up, created 2-3 notes in their first week, and then stopped opening the app entirely. This segment is valuable because they've already shown intent — they're not tire-kickers, they tried and bounced.

A Note on Segment Selection

A tempting alternative segment is "power users who want more advanced features" — but that segment is already retained, so improving their experience moves a smaller, noisier lever. Early churners are earlier in the funnel, larger in raw count, and their drop-off is a clean, attributable signal. Naming why you rejected the alternative segment, in one sentence, is itself a signal of judgment.

Step 2 — Goal (JTBD framing). You don't need a large study to name this job with confidence. Nielsen Norman Group's long-standing usability research suggests that a handful of target users — sometimes as few as five — surfaces most of the friction points that matter. Talking to even that many suggests the job is "capture a fleeting thought fast, before I lose it, and trust I can find it again later without effort." That's a job, not a feature request.

Step 3 — Friction. Walking their journey stage by stage surfaces a likely dip right after the first note is created: the app hands them a blank, unstructured canvas with no default organization, so the second and third notes feel effortful rather than automatic — and the habit never forms.

Step 4 — Ideate and prioritize. Three ideas, each mapped to a segment, a friction, and one metric:

Ranked with RICE: the daily capture prompt likely scores highest on Reach and Effort (cheap to build, hits the whole segment), while auto-tagging scores higher on Impact but lower on Confidence, since it depends on model quality. Present the prompt as your lead recommendation and the others as sequenced follow-ups — not simultaneous asks.

Why This Version Scores Well

Notice what's absent: no mention of dark mode, no "add AI," no unranked list of ten features. Every idea traces back through friction to a named goal for a named segment, and each has exactly one metric it's accountable to. That traceability — segment → goal → friction → idea → metric — is what a rubric-driven interviewer is actually scanning for, not idea count or novelty.

How Interviewers Score This Question

Interviewers scoring this question are rarely grading "is this a good idea" — they're grading whether you demonstrated a repeatable process for finding good ideas, since that process is what you'll apply to problems they can't anticipate. Seniority shows up in scope and evidence, not in idea flashiness.

This scoring pattern isn't unique to this question — it's the same ladder used across advanced PM interviews, decoded in detail in the PM leveling rubric. If you're using this question as practice for a promotion conversation rather than an external interview, the same segment-goal-friction-metric discipline is exactly what underpins a credible promotion case built on strategic impact — both are, at bottom, arguments that you can find and size the right problem before anyone hands you one.

The Three Axes Behind Most Rubrics

Underneath the level-specific language, most interview rubrics for this question grade three underlying axes: structure (did you follow a repeatable process, not improvise), evidence (did you ground friction in something observable rather than assumption), and judgment (did you defend a prioritization call instead of hedging on all options equally). Weakness on any one axis caps the score, regardless of how strong the other two are.

Practicing This Without a Real Interview Loop

You can rehearse this structure without waiting for an actual interview by deliberately mapping a product's journey and marking where the experience dips, rather than brainstorming features from memory. Seeing the dip before you ideate is what keeps the exercise evidence-led instead of vibes-led.

Key Takeaways

• Never open with a feature. Open with a user segment — specific and nameable, not "all users" — before you say a single idea out loud.
• Borrow Jobs to Be Done framing to state the segment's goal in terms of the job they're hiring the product for, not a feature they've asked for.
• Find friction by walking the journey stage by stage, not by guessing where you assume the product is weak.
• Every idea needs exactly one target metric, chosen before the idea, with a causal chain you can defend in one sentence.
• Prioritize explicitly with RICE or Kano rather than presenting a flat list of equally-weighted ideas.
• Narrow scope aggressively and stay narrow — the most common failure is starting broad and never coming back to specifics.
• Seniority is shown through evidence and scope, not idea count — one well-traced idea beats five shallow ones.