Think of a courier who takes your parcel, gives you no receipt, no tracking number and no delivery promise. You would never use that courier twice — yet that is exactly what pressing Apply feels like on a job portal. The parcel leaves your hands and the product goes silent. My capstone brief for the AI-First PM cohort put a number on that silence: among seekers who apply to three or more roles in a week, only 21% attend an interview within fourteen days, and an interview counts only when the seeker confirms attending and the recruiter marks them interviewed. The target was 36% in six months, with a constraint that shaped everything: I could design only the seeker side. Recruiters pay for Naukri, seekers mostly do not, and nobody can order a recruiter to reply.
This first essay is about the diagnosis. The second, Three Taps and a Handshake, is about the cure.
Trace one parcel, not the whole courier
The research mistake I avoided was asking people what they think of the app. People have opinions about apps; they have memories of applications. So every conversation traced one real, recent application day by day: what the app showed, what arrived by call, WhatsApp or email, and what the person did next. Thirty conversations — twenty-two seekers across experienced professionals and freshers, six recruiters, a placement officer and a career coach — plus a teardown of Naukri as a seeker and a study of nine other channels, from LinkedIn and Indeed to referrals and Telegram groups.
A note on provenance, because it matters in product work: the interview log in my submission is a labelled pilot persona set built to the research guide so the analysis pipeline could be proven end to end; it is marked as such on every page and is being replaced with field respondents. The counts below should be read as the pattern the pipeline surfaces, not as a survey result.
| What the trace showed | How often | What it means |
|---|---|---|
| "Applied" never changed; nothing told the seeker whether a human had opened the profile | 19 of 22 seekers | Seen and unseen look identical, so the seeker cannot act |
| Could not name a single signal that a role was live or genuine | 24 of 30 | Post date is not liveness; agencies keep filled roles open |
| Had an offered interview that simply never happened | 11 of 22 seekers | Scheduling happens off-platform with no confirmation or reminder |
| No one-tap or bulk way to respond, and no upside if they did | 6 of 6 recruiters | Replying is a per-candidate chore with no reward |
Fourteen days in the life of an application
I then laid the application out on a calendar. Day zero it is sent. Over the first two days, roughly half are never opened — the role is stale, filled, or buried under four hundred others. Of those opened, most sit silent because the recruiter has no cheap way to say "not now". A small share get a proposal, usually by phone from an unknown number or a WhatsApp message with a day but no address. Of those proposals, a third lapse because nobody confirmed. Of those confirmed, a quarter are not attended, by either side. And of those attended, nearly one in five is never marked by the recruiter, so it does not count.
Per application, 2.7% becomes a counted interview. Per seeker, over about nine applications in two weeks, that compounds to 21%. The model reproduced the brief before I designed anything.
The six loss points are the useful output of this exercise: never seen, seen but silent, contacted off-platform, proposal lapses, no-show, never marked. Each one is a place a product can intervene — and, as it turned out, they pair neatly with two solutions.
Symptoms versus the one cause
Stale postings, fake recruiters, no-shows and spray-applying each look like a separate problem with a separate fix. First-principles thinking says otherwise. Every one of them is downstream of a single fact: nothing a recruiter does inside their applicant list is visible to the seeker, and nothing a seeker does is visible to the recruiter until a call happens. The application is a one-way message with no shared state.
That framing also explains the doom loop. Silence pushes seekers to apply more. More applications raise recruiter noise. More noise lowers reply rates. Lower reply rates mean more silence. This is why "send more alerts" and "pay for a boost" fail: they add volume to a system that is already drowning in it. Seventeen of the twenty-two seekers described boosts as paying to be ignored faster.
The tension I had to resolve
Seekers want visibility on every application. Recruiters want fewer, better applicants and zero extra steps. On a two-sided market where one side pays, you cannot give the seeker visibility by giving the recruiter homework. The resolution, which the next essay builds out, is to expose the recruiter behaviour that already happens — views, replies, activity — at zero effort, and to price the few manual actions that matter at one tap each, with a reward attached.
Interviewing the recruiters was the turning point. Not one of them was unwilling to respond. They simply had no button for it and nothing to gain. Give a manufacturing HR head a "Not now" that works on fifty applicants at once, and he will press it. The market is not cruel; it is unequipped.
Problem statement, lifecycle map, barrier analysis and the thirty-row log are in the case folder: github.com/Eswar5313/product-management-case-studies → 14-naukri-capstone.