Product Manager
I build the supply side of marketplaces.
Most people call me Abhi, short for Abhivarmmen /əˈbɪ-vər-mən/. I work on inventory and partner platforms at Swiss Marketplace Group in Zurich, after earlier years on enterprise SaaS and regulated data systems at Roche and analytics at Shell in Australia.
I build products by working three sources in parallel: what the data says, what users tell me, and what the team can imagine. Skipping any one of them shows up in the result.
The shorthand
Most PMs build what users see. I work behind that.
Every marketplace has two halves. The buyer side is the part with the photos and the search bar. The supply side is the part where partners post inventory, APIs ingest it, and someone makes sure the data is good enough to feed search. That second half is where I've spent most of my decade.
What users see
Search, listing pages, filters, the photo carousel. The part with screenshots in the press.
- Search & discovery UI
- Listing detail page UI
- Buyer messaging UI
- Pricing & recommendations UI
What partners depend on
Where 7,000 dealers post 250,000 listings a day. Boring unless it breaks. Then nothing else works.
- Partner / dealer APIs API
- Listing ingestion pipelines PIPE
- AI-generated descriptions AI
- Data quality & VIN services DATA
Principles
Four things I optimise for.
The framework I use to decide what to ship, what to cut, and what to push back on.
AI & Automation
I use NLP, ML, and automation where the data is rich enough to move buyer engagement, content quality, and operational cost.
From the work
At SMG, I built AI-generated listing descriptions from scratch with no existing AI infrastructure or playbook. The result: measurably higher buyer engagement and lead conversions across the marketplace.
Read the case studyData-Driven Decisions
I build A/B tests, analytics, and user-interview loops into the workflow so the team has data on each call before the next release.
From the work
I embedded a structured A/B testing framework into marketplace listing workflows, moving the team from opinion-based releases to measurable iteration. One set of experiments alone lifted listing completion by 18%.
Read the case studyScalable Platforms
I design APIs, re-platform legacy systems, and automate workflows so the platform handles more partners without more engineering hours per partner.
From the work
I led the re-platforming of partner APIs serving 7,000+ dealers, turning a fragile integration layer into a scalable platform that drove 55% higher API adoption.
Read the case studyCustomer-Centric Strategy
I start the roadmap from the user problem. Small wins on real pain compound into product strategy.
From the work
When dealer workflows were slowing inventory velocity, I started with the seller pain points, automating repetitive tasks and improving data quality. Time-to-market dropped 35% and seller drop-offs fell 12%.
Read the case studyOperating
Shorthand for how I work.
Less philosophy, more operating preference. Four lines that hold up across roles, teams, and org size.
Write more, meet less.
I default to async, written-first. Meetings are for decisions and conflict; writing is for thinking. The team's documentation is the team's memory.
Owners before plans.
I push for who-decides on a call before what-gets-decided. Roadmaps that try to skip this step die in execution, not in planning.
Strong opinions, real evidence.
I bring a clear point of view to every call, and I am comfortable letting the data disprove it. A/B tests where time allows; best judgment where it does not.
Ship to learn.
A real thing in production beats a perfect thing in a deck. I trade some polish for cycle time and refactor on the signal that comes back.
Perspective
How I read a partner platform.
Most platform PMs inherit a single number, uptime or adoption or NPS, and a single number rarely tells the whole story. Five dimensions I read in parallel. The shape below is roughly where my current platform sits this quarter; the next bet usually lives where two of them disagree.
Reliability
Does the platform do what partners expect, every time?
Signals · uptime · error rates · breaking-change cadence
Adoption
Are partners using what you built?
Signals · active monthly partners · endpoint depth · feature utilisation
Data quality
Can downstream teams trust the output?
Signals · validation pass rate · attribute completeness · accuracy vs. source of truth
Onboarding velocity
How fast can a new partner go live?
Signals · time-to-first-call · dev-to-prod cycle · self-serve onboarding rate
Support load
How much human help does the platform demand?
Signals · tickets per partner · share resolved without engineering · MTTR
What this shape tells me
Reliability is healthy. Adoption is climbing faster than data quality, which is the classic platform fragility, so the next investment goes into validation and attribute completeness, not new endpoints. Support load is the lagging cost of that gap.
Stack
What I work with.
The toolkit I reach for, day-job and off-hours. Tools change; the reason for picking them does not: rich data, fewer manual steps, faster feedback.
- Data & experimentation
- SQL, BigQuery, Looker, Tableau, Mixpanel, Amplitude, Optimizely, Split.io, Datadog, Sentry
- AI & platform APIs
- NLP / LLM pipelines, Claude / OpenAI APIs, OpenAPI, GraphQL, MuleSoft, Apigee, Postman
- Discovery & delivery
- Figma, FigJam, Miro, Notion, Linear, Jira, Confluence, Loom
- Building on the side
- Swift / SwiftUI, TypeScript, React, Astro, Next.js, Tailwind CSS, Firebase, Vercel, Node.js, Python See what I’ve shipped →
A year off · 2018 – 19
The year I learned what "shipped" actually means.
In 2018 I took a year out to co-lead a humanitarian project in eastern Sri Lanka, managing an LKR 100M budget to build 35 permanent homes for families displaced by decades of conflict.
Coordinating local contractors, negotiating supplier contracts, navigating government processes, then watching 140 people move into homes of their own. It taught me more about stakeholder management and the real cost of scope than any product role since. The skills we sharpen in tech matter far beyond a screen.
homes
permanent, earthquake-resistant, handed over
budget
15% under cost through procurement renegotiation
lives
families resettled in their own villages
Outside the office
Basketball is the through-line.
Eight years of club play between 13 and 21. I still play most weeks. The rest of the week is padel, tennis, cycling, and bouldering, and I stay involved with the Swiss chapter of my school's alumni network.
Currently exploring
Three open questions on my desk.
Open questions stay sharper than the answered ones. These are the ones I would rather talk about over coffee.
- Q1
Generative AI for marketplace content
Past the press-release stage. What evaluation actually looks like in production, and where buyers stop trusting model output.
- Q2
Computer vision for vehicle damage detection
The cost-to-buyer-trust ratio when the model is wrong 8% of the time, and whether you ever ship without a human in the loop.
- Q3
Pricing models that learn from supply-side behaviour
How dealers actually price themselves when the platform shows them the distribution, and whether transparency drives convergence or revolt.
Testimonials
Endorsements
"Abhivarmmen shows outstanding initiative and remarkable commitment. We regard him as an exceptionally flexible and resilient employee, exemplary in every respect."
"Abhi possesses a unique ability to bridge the gap between technical complexities and business objectives. His strategic thinking and collaborative approach make him an invaluable asset to any product team."
"I can confidently say that he is an outstanding professional. Throughout our time together, Abhi consistently demonstrated a deep understanding of product strategy, a strong commitment to delivering quality results, and an excellent ability to collaborate with cross-functional teams."
Showing 2 of 3. Refresh to read another perspective.