Case Study · Product Design
SideDoor
From scattered DMs to a referral platform built on trust, structure, and visibility.
💭Have you ever sent 40 LinkedIn messages asking for a referral , and heard back from 3?
That's not a you problem. That's a systems problem. The referral process runs on cold outreach, hope, and WhatsApp. It's broken for everyone: candidates get ghosted, referrers get spammed, recruiters can't trust the quality. SideDoor is my attempt to fix it.
Candidates find relevant referrers ranked by role match, alumni signals, and response rate , not a random list of strangers.
Referrers evaluate candidates with structured summaries and graded confidence levels , not a blank DM and a gut feeling.
Both sides track every stage after the referral is submitted , no black box, no repeated follow-ups.
👉Short on time? Here's the Figma prototype.
The problem, lived
Meet Ishaan. 2 years into his career as a product designer. He finds a role he's genuinely excited about: Product Designer at CRED. He opens LinkedIn, searches for CRED employees, and starts messaging.
He messages 15 people. 2 reply. 1 agrees to help. Then begins the back-and-forth: "Which role exactly?" "Can you send the JD link?" "What's your notice period?" "Send me your resume as a PDF please."
Three days later, his contact submits the referral. Then silence. For two weeks, Ishaan has no idea if anything actually happened. He checks his email obsessively. Sends a follow-up. Gets a "will check" reply. Nothing more.
"I don't even know if they submitted it. I have no way to track it."
— Candidate, interview
Ishaan's not an edge case. He's every candidate. The employees he's messaging? They're getting 10–50 requests like his every week. One PM I interviewed got 40–50 DMs a day whenever their company posted a public listing. They'd turned off LinkedIn notifications just to work.
Why this matters
Referrals are the highest-converting hiring channel that exists.
The channel works. The experience around it doesn't. That's the gap SideDoor fills.
Who I was designing for
Candidates
0–5 yrs exp
They know referrals work. They've tried cold applying. They just can't reach the right people.
Referrers
Mid-level, tech
Every referral puts their reputation on the line. The ask feels social. The risk is professional.
Recruiters
Secondary
They value referrals in theory, but bonus gaming has made them skeptical. Trust needs rebuilding.
The solution, in five screens
Now Ishaan opens SideDoor.
Here's what changes.
These five problems aren't isolated: they're a system. Fix one without the others and you just move the problem somewhere else. Every solution had to work as part of the whole.
01Referrer Discovery
➡️How can a candidate find the right person to approach, not just any random employee?
Ishaan clicks "Get Referral" on the Product Designer role at CRED. Instead of a search bar, he sees a curated, ranked list of CRED employees.
Diya Sharma is at the top: 92% match, Same College, responds in ~8 hrs, has referred 12 people before. That's not a stranger anymore.
🎯Ishaan isn't messaging into the void. He's reaching out to the most relevant person for this exact role, with clear evidence they'll actually respond.
Referrers carry real professional risk when they put their name behind someone. A bad referral doesn't just waste their time — it reflects on their judgment with their recruiter, their manager, their team. Ranked signals — role overlap, shared alumni, past referral activity — do two things at once: they help candidates find the most relevant person to reach out to, and they help referrers recognise which requests are worth taking seriously. Trust has to flow both ways before either side commits.
Ishaan had been playing a numbers game. Then he saw the ranked list: Diya Sharma, 92% match, same university. For the first time, a name felt like a real lead. 🎯
02Structured Request
➡️How can a candidate send a request a stranger actually wants to respond to?
Ishaan doesn't write a DM. He fills a guided form: role is pre-filled, fit points are suggested from the JD, he writes a short pitch. Preview screen. Send.
Diya receives Ishaan's request and understands the fit in under 30 seconds. No back-and-forth. No chasing for details.
🎯Every request is structured the same way: clear, scannable, easy to evaluate. No more "refer me anywhere" messages.
JD auto-attached, fit points pre-suggested from the JD, preview screen before sending. Less effort for the candidate, higher quality for the referrer.
Diya opened it on her lunch break. No wall of text, just a clean card: 85% match, three fit points, a short pitch. She read it in under 30 seconds. 👀
03Referrer Evaluation
➡️How can a referrer decide confidently without risking their own reputation?
Diya opens Ishaan's request. She doesn't see a resume dump. She sees a structured evaluation: 85% overall match with breakdown, key strengths (green), potential concerns (orange), and Ishaan's own pitch.
Instead of "Refer or Don't Refer," she has four options: Decline / Review Later / Refer / Strongly Recommend. She clicks Strongly Recommend. Adds a private note for the recruiter.
🎯Diya didn't guess. She made a confident, informed decision in under 2 minutes.
Referrers aren't unwilling: they're uncertain. Building an evaluation assistant instead of just a profile view reduces that. The private note adds context for the recruiter without the candidate ever seeing it.
Diya clicked "Strongly Recommend." Ishaan's phone buzzed a minute later. Request accepted. For the first time, the silence didn't feel like being ignored. 🥹
04Shared Pipeline
➡️How can both sides know what's happening after the referral is submitted?
Ishaan gets a notification. He opens SideDoor and sees a shared timeline: Request Accepted → Referral Submitted → Application Under Review → Screening → Interview → Outcome.
At each stage, he can see who owns the next action: "Recruiter reviewing your application." No ambiguity. No need to follow up.
🎯Ishaan doesn't check his email 20 times a day anymore. He knows exactly where things stand, and so does Diya.
Both sides see the same timeline. People tolerate slow hiring. What they can't tolerate is uncertainty. This fixes that, without ATS dependency.
Ishaan tapped it expecting another follow-up to chase. Instead: a timeline. Request Accepted. Referral Submitted. Under Review. He put his phone down. Not once did he pick it back up. 😌
05Quality Over Volume
➡️How do we stop spam without making it harder for serious candidates?
Candidates mass-message because response rates are low. Response rates are low because messages are generic. It's a vicious loop. SideDoor breaks it at three points.
Structured requests
Filling fit points and a short pitch takes effort, which filters out people who aren't serious. Low-intent senders drop off naturally.
Ranked referrer matching
Candidates see a curated list, not every employee. This nudges targeted outreach over shotgun behavior.
Request limits & inbox prioritization
The referrer's inbox is ranked by fit score and quality signals, not chronology. Referrer attention is treated as a limited resource and protected.
Ishaan sent one request. Got one referral. Watched it move through screening without a single follow-up. Diya saw the same thing. No one was left guessing. 🤝
Who gains what
Candidates
Higher conversion from outreach to referral. Less time wasted messaging people who won't reply.
Referrers
Less effort, less risk to their reputation. Participating feels normal, not like a personal favour.
Recruiters
Better-quality signals. Less time screening bad referrals, more confidence in the ones that come through.
The design process, unfiltered
That's the product. Now here's how I got there.
None of this was obvious at the start. A lot of it was wrong before it was right.
Starting with assumptions, not answers
Before talking to anyone, I wrote down 14 assumptions: candidates don't know who to ask, referrers ignore cold DMs, nothing gets tracked, it all feels transactional. Then I went out to validate or kill each one.
Constraints I worked within
No ATS access
Couldn't depend on Workday or Greenhouse for anything — had to design fully independent of company systems.
Mobile-first, India
WhatsApp-heavy users with low tolerance for friction or long flows. Every tap had to earn its place.
Zero usage data
0→1 with no existing product. Every scope call had to be made on research and judgment, not behaviour.
Work alongside existing tools
LinkedIn and WhatsApp weren't going away. SideDoor had to fit into existing workflows, not replace them.
MVP scope only
Incentive systems, AI matching, and recruiter tools were real — but they had to wait for v2.
What 10 people told me
I interviewed 10 people: 2 recruiters, 4 referrers, 4 candidates. All in India's tech ecosystem. The same things kept coming up.
"Whenever a higher referral bonus is announced, suddenly the system gets flooded with random resumes."
— Recruiter
"If I'm doing a favor for somebody, why would I go through all this back and forth for someone I don't even know?"
— Referrer
"I messaged 40 people. 9 replied. 3 actually submitted. I have no idea what happened after."
— Candidate
What the numbers confirmed
The numbers confirmed what people were telling me.
Mapping the current experience
I mapped the AS-IS journey for both sides, step by step. Every stage was broken in a different way.
Discovery was manual and random. Outreach was anxiety-inducing. Evaluation was guesswork. Post-referral was a black box for both sides. Then I mapped where each problem should become a design solution, the TO-BE.
What competitors got right, and what none of them solved
I analyzed 11 platforms: LinkedIn, GetMeReferred, Instahyre, Jumbl, EasyRefer, Cutshort, Wellfound, TalentPool, Naukri, ReferMe, and Reddit/Discord communities, across 6 dimensions: discovery, referral flow, tracking, trust signals, spam control, and end-to-end coverage.
Discovery is largely solved
LinkedIn and AI matching are strong. The problem isn't finding jobs, it's everything that happens after.
Tracking is universally broken
Zero post-referral visibility on every platform. Not one solved this, not even referral-specific ones.
Trust exists but is shallow
Profiles and ratings exist, but no platform helped referrers actually decide whether to refer.
No one owns the full journey
Discovery → LinkedIn. Communication → WhatsApp. Submission → ATS. Tracking → nowhere.
The core insight: no one owned the end-to-end experience. Every platform solved a piece. None solved the whole. That became the strategic bet for SideDoor.
What I built, what I cut, and why
One filter for v1: does this solve the core trust and workflow problem, or does it add complexity? If the latter, it waited.
P0 · Built
Trust layer (structured profile, context signals), decision framework for referrers, unified referral flow, shared pipeline visibility
P1 · Next
Auto-fill referral submission, in-app messaging, notification system, status updates
Cut
Deep ATS integration (too much company dependency), AI-heavy matching (black-box trust problem), complex incentive/payout system (a separate product problem entirely)
Design principles, behind every screen
What every screen had to answer to
North Star
"Enable high-quality referral matches through trust, relevance, and clarity."
Key decisions that shaped the final design
I got several things wrong. Here's what broke and how I fixed it.
Every decision below had a wrong version before it. I'll show what that was and why it failed.
👉Ranked discovery instead of open search
My first version was an open search: type a company, find any employee, message them. Peer review killed it fast. It recreated the exact problem I was trying to solve: candidates would mass-message everyone. Referrers would get spammed. We'd be back to LinkedIn.
People trust things they can see: shared college, response rate, how many people they've referred before. That's what gets surfaced upfront.
Trade-off accepted: Candidates see fewer referrers, meaning lower discovery freedom. But quality interactions matter more than volume in referral systems. That was a conscious choice.
👉Structured request form instead of free-form messaging
Every referrer I talked to said the worst requests were vague. "Refer me anywhere." No role, no context, no reason. My early versions had a free-form text box. It felt flexible, but it was wrong. Free-form preserved the exact behavior I was trying to eliminate.
Structured input wasn't about removing freedom. It was about making it easy to send something good. Pre-suggested fit points and a preview screen before sending helped with both.
Intentional friction is a feature, not a bug. A small amount of effort acts as a commitment filter that discourages low-intent behavior without blocking serious candidates.
👉Graded recommendations instead of binary Refer / Don't Refer
My first evaluation screen had two buttons: Refer or Don't Refer. Simple, clean. Wrong. Testing revealed the real issue: referrers who thought a candidate was "probably good but not 100% sure" had no place to land. They'd either over-commit (risky for their reputation) or bail entirely. We were losing all the middle cases.
Human confidence isn't binary. A spectrum (Decline / Review Later / Refer / Strongly Recommend) gives referrers room to be honest about their confidence. Less pressure, more participation. The private note lets them share context with the recruiter that the candidate never sees.
👉Lightweight shared pipeline instead of ATS integration
Early on, I explored full ATS integration: connecting SideDoor to Workday and Greenhouse so tracking would be automatic and real-time. I dropped it fast. ATS integration requires company cooperation, IT approvals, and months of enterprise sales work. Completely unrealistic for MVP.
Users don't need real-time ATS data: they need enough visibility to stop the anxiety. Lightweight stages (Submitted → Screening → Interview → Offer) solve the emotional problem without the technical dependency.
Perceived clarity matters more than perfect automation.
Designing for the edge cases
SideDoor doesn't just handle the good days.
Every referral flow has a moment where something slips. The referrer gets busy. The ATS drops the submission. A company goes quiet for three weeks. Here's what the design does when that happens.
Referrer accepts, never submits
Nudge after 48 hrs. If still no action, candidate can re-route to another referrer.
ATS submission fails
Mark as 'Pending confirmation.' Referrer can manually confirm.
Company goes silent for weeks
Show 'No update yet, this is normal at this stage.' Sets expectations before anxiety spikes.
Candidate hits request limit
Show cooldown. Prompt to improve request quality, not just wait it out.
Target metrics that tell the real story
If this ships, here's what moves.
Ishaan's story is what good looks like. But design doesn't ship on stories. It ships on metrics. Here's what every decision was tied to.
These are design-time targets — what success looks like if this ships, not measured outcomes.
North Star Metric
Accepted referral requests per active user
The goal was never more requests. Better ones. When acceptance rates go up, trust is working and both sides are getting real value.
Referral funnel — Before Vs SideDoor
Target Business Outcomes
Referral-to-hire rate
One good request. One hire.
~1% cold apply
SideDoor target
Time to hire
Ishaan didn't wait 6 weeks.
44–55 days portals
SideDoor target
1-year retention
Better matches stay longer.
33% job boards
SideDoor target
What's next
There's still a lot on the table.
Honest reflection
If I started over, here's what I'd do differently.
I spent too much early time on the candidate's side, they were easier to reach. But referrers were the harder design problem. The hesitation, the fear of referring someone bad, the pressure of a yes/no choice: I only really understood those after building and scrapping my first evaluation screen.
The ATS constraint initially felt like a wall. It turned out to be a forcing function. Removing the dependency pushed me toward a simpler, lighter pipeline model that's more resilient — it works regardless of what ATS a company uses. I've started thinking of constraints that way: not as blockers, but as the thing that stops you from overbuilding.
My research sample was skewed. The four referrers I interviewed had all opted into the conversation, which means they were probably already more thoughtful about referrals than average. The real resistance — the employee who ignores every request — I never got to talk to. That's a gap. You should design for the skeptic, not just the willing.
If I started over, I'd prototype the referrer evaluation flow first and test it with real employees before anything else. Starting with the harder side would've made everything else clearer.
Ishaan didn't need to message 40 people. He needed one good match, one structured ask, and the ability to see what happened next.
That's what SideDoor was built around: not making referrals faster, but making them work the way they were supposed to. The channel works. The experience doesn't. This was my attempt to close that gap, for Ishaan, for Diya, and for everyone doing this the hard way.
If you made it this far, thank you. Always happy to talk. ❤️