“I build the operating system for AI inside hyperscale marketplaces.”
I’m Product Owner for Supply GenAI at Agoda. Engineering owns how it’s built; I own the problem, the priority and the outcome — agentic AI for hotel supply, especially the long tail of partners humans cannot cover. Usage is not the scoreboard. Impact is. Before Agoda: 11+ years in hyperscale e-commerce, 8+ at Amazon — a $200M clean-sheet business build and GenAI in production for 5,000+ operators.
Agoda · ex-Amazon
I take the messy, high-stakes problem with no playbook — and leave behind an operating system the organisation keeps using.
Eight-plus years at Amazon, the world's largest third-party marketplace, taught me to work where strategy meets execution: identify the inefficiency or control gap through data, author the business case, design the operating model, drive end-to-end delivery from problem definition through impact measurement, and codify it as SOPs and governance so the next vertical or market inherits the framework rather than starting over.
Four of those years were onsite in Munich, running Pan-EU operations across DE, UK, FR, ES and IT with Tier-1 brand portfolios (J&J, Reckitt, P&G). I've partnered with Director-level strategy boards on the high-ambiguity calls, led a 15-person team in Munich and three in Bengaluru — and I care about shipped outcomes, not decks.
Today I’m based in Bangkok. At Agoda I own what Supply GenAI ships: the problem, the priority, and the outcome — agentic AI for the long tail of hotel partners.
Frame the ambiguous problem, find the unit economics and the real bottleneck through data, size the opportunity.
Author the business case and design the operating model — SOPs, decision rights, governance, the cost structure.
Drive cross-functional delivery across Product, Engineering, Commercial, Data, Finance and Legal — into production, across regions.
Turn the win into a repeatable, multi-market playbook so it scales — not a one-off.
Representative builds — each a clean-sheet problem turned into a shipped, measurable outcome.
Spotted the whitespace, authored the business case (ROI model, investment framework, P&L forecast), hired 15 specialists and onboarded 25 Pan-EU vendors on a ~$100M portfolio. Scaled to 3 regions and 50+ partners — a $200M annual run-rate with $1M annual savings, in under 12 months.
Led an AI productivity charter end-to-end — POC → production → monitor-and-iterate with Engineering and Data Science — embedding a GenAI toolkit into the daily work of 5,000+ operators across 5 regions in 6 months (0→98% adoption). Recovered ~50,000 operational hours/week at ~8.6x first-year ROI. Authored the prompt-engineering and LLM-evaluation guidelines that became the org's production standard.
Led a 90-day Commercial Excellence deep-dive ($70M annual, 500+ SKUs): diagnosed unit economics, restored 190 products to healthy margins, drove 65% QoQ growth (6M→9.9M/quarter) vs ~7% category. Codified the playbook across 10 EU partners for $12M annual uplift, and closed $16M+ in long-term agreements.
Designed a global operating model — SOPs, decision rights, certification-gated readiness, and weekly/monthly/quarterly steering routines that surfaced ~$20M/week of risk before P&L impact. Outcomes: escalations down 91%, KPIs up 1,000 bps, onboarding cycle halved from 8 to 4 weeks.
Architected and launched role-based competency frameworks for 8,000+ managers and above — personalised learning paths and coaching ecosystems. Outcome: learner NPS of 85, knowledge-satisfaction up 1,500+ bps, and a Continuous Learning Program that lifted vendor satisfaction 1,000+ bps. Coached one IC to manager in 12 months — fastest in the org.
Delivered $4.3M in peak-season operational campaigns (Prime Day, Spring Deals) through cross-functional coordination across supply, marketing and partner teams. Resolved 610 operationally-impacted SKUs in 5 days pre-launch, recovering $4.1M in incremental revenue.
Two of the recommendations on my LinkedIn profile — from someone I managed, and someone I delivered alongside.
Soham led multiple enterprise-scale programs supporting a large, high-revenue vendor ecosystem, bringing together strategic clarity and strong execution. He designed governance and operating frameworks that reduced partner ramp-up time by 25% while improving satisfaction metrics by 1,000 basis points … He doesn’t just deliver results — he elevates the people and organisations around him.
His ownership, analytical skills and drive to deliver results were instrumental for the success of those projects. He was really active fostering collaboration and teamwork, which boosted team effectiveness and created an excellent work environment.
When I want to understand an operating model, I build it — a working prototype beats a slide about one. Two recent examples, each a single-agent Claude tool-using loop with grounded retrieval — answer only from the source, cite it, escalate rather than invent — the same discipline behind the production RAG rollout I led at Amazon (5,000+ operators, ~50,000 hours/week recovered). The human owns the goal and the judgment; the agent does the prep.
An agentic copilot for Key Account Managers on a flat team: it does the account prep so the manager spends the hour on the commercial conversation — and steers “what good looks like” across markets from one central engine.
~3–4 hrs of prep → ~2 min — pushing a flat team toward the 80%-commercial-time target.
An account-excellence assistant for Strategic Account Managers: pull a partner's commercial signals, diagnose the top issues, and draft the partner outreach — grounded in SOPs, not generic model opinion.
The orchestrator pattern: the human curates intent and judges the output; the agent does the retrieval and drafting.
Both are deliberately small, learn-by-doing prototypes — built independently, before I joined Agoda, on synthetic data — single-agent loops on the merchant/commercial side only (they make delivery reliability visible; they don't own logistics). Honest proof of building, not just sponsoring.
The third one is the page you're on. I designed, wrote and coded this site myself, directing Claude — one self-contained file, no framework, no CMS, no build step. The Ask Soham assistant is a deterministic intent matcher rather than a language model, which was a deliberate choice: it costs nothing to run and cannot invent a fact about me.
Bangkok, since August 2026. Product Owner for Supply GenAI at Agoda: agentic AI for hotel supply, especially the long tail of partners humans cannot cover.
Always glad to compare notes with people working on the same problem — the note below is where my thinking started.
Amazon noteMost enterprise AI pilots die somewhere between an impressive demo and a Tuesday morning. What actually moved 5,000+ operators at Amazon.
Read the note →If you work on applied AI in marketplaces or hotel supply — in Bangkok, Singapore, Kuala Lumpur or anywhere else — I’m interested in how you tell operating leverage from theatre.