Somesh Nayak — Business Analyst | Product Operations

I work on messy problems before they become obvious products.

I'm Somesh Nayak, a Business Analyst with 2.5 years at a B2B SaaS startup. I ran customer discovery, sold and delivered the first paid projects before any software existed, led product demos, and tested AI outputs on real construction-procurement workflows.

Somewhere between product, customers and execution.

My experience has rarely fit inside one function. I started in business analysis and gradually moved closer to customer problems, product decisions, GTM and day-to-day work with the founders.

I spent approximately 2.5 years at Ninety North Software Pvt. Ltd., a B2B SaaS startup where the product evolved from Enfund to Bilderly and eventually ywork.ai.

My work moved across market research, customer discovery, product validation, customer demos, GTM, AI product evaluation, CRM, operations, hiring and internal documentation.

The part I enjoy most is the stage where the answer is not obvious yet: understanding what customers are actually struggling with, testing whether the problem matters, structuring what needs to happen next and then helping execute it.

My Civil Engineering background also gives me useful construction-domain context, particularly around procurement, supplier quotations, BOQs and takeoff workflows.

Since May 2026 I have done freelance GTM and product-research work for two B2B and AI companies, and built Treelync, an AI mock-interview platform, on my own.

Customer → Product

Discovery, workflow mapping, validation, feedback synthesis and product iteration.

Product → Market

Product demos, early GTM, customer objections, CRM and commercial feedback.

AI → Business Value

Testing AI outputs on real documents: output quality, context handling, consistency and customer usefulness.

Founders → Execution

Cross-functional execution across product, GTM, operations, hiring, documentation and internal initiatives.

Selected Work

Four problems that shaped how I think.

Not just what I worked on — how I approached the problem, what we tested and what changed as a result.

Customer Discovery · Zero-to-One

Finding a problem customers would pay to solve.

During UK construction-market discovery, the same problems kept coming up: scattered project documents, manual RFQ chasing and quotations that were hard to compare like for like. Before building software, the team tested whether the workflow was painful enough that customers would pay for it.

Discovery

Ran 50+ discovery conversations with UK contractors, developers and suppliers, each booked through my own cold-email, LinkedIn and cold-call campaigns.

Validation

Within a £2,000 company discovery budget (not spent), sourced, sold and manually delivered 4 paid projects: supplier sourcing, RFQs, follow-ups and comparison of 50+ supplier quotations.

Documentation

Logged every delivery day in an Excel sheet (tasks, sources, supplier emails, comparison logic). That log became the basis for chain-of-thought prompts tested for the AI workflows.

Outcome

4 paying customers before the software existed. It showed that customers would pay, and which steps were worth automating first.

Demos · Customer Feedback · GTM

Turning demo conversations into product changes.

As the product moved from Bilderly to ywork.ai, demos became more than a sales activity. Each call was also a source of workflow evidence, objections and feature requests.

Demos

Led 100+ online demos across the UK and US, about 70 independently and about 30 alongside the founders.

Feedback loop

Recorded requests, objections and recurring questions after each call and shared the patterns with founders and engineers.

Commercial

Closed about 15 early paying subscriptions (£100–£500) on live calls, including the budget discussion and payment by Stripe link.

Example: from customer request to feature

A demo customer wanted to check AI-generated quantities directly on the drawings. I researched and specified a drawing-measurement tool with AI-suggested scale, reviewed the UI and UX, and coordinated testing with the product team.

AI Product Testing

Testing AI on real BOQs, not on the model name.

When sales on the existing features slowed, a co-founder and I explored whether AI could generate BOQs from construction drawings. My part was product-side testing and prompt iteration, not ML engineering.

Testing

Compared AI-generated BOQs with BOQs from past projects, trade by trade (groundworks, electrical, HVAC), and iterated the prompts.

Model comparison

Compared OpenAI and Claude on real procurement and BOQ scenarios for output quality, context handling, consistency and usability.

Human reference

Proposed recording a freelancer's manual BOQ workflow as a step-by-step reference for how the AI should follow the process.

Result

Internal accuracy rose from about 89% to 93% on our reference set (an internal measure, not an independent benchmark), and the tech team adopted the final prompt after my testing.

Process · Automation · Testing

Cutting manual data work with a supplier-data pipeline.

Loading supplier data into the product meant a slow process of gathering and cleaning records by hand. I proposed a one-week build and coordinated two developers and a designer.

Workflow

Apollo extraction, deduplication, website enrichment, company credit-data lookup and an AI-ready Excel output.

My role

Proposed the idea, defined the steps, split the work between design and development and supervised the build. The developers wrote the Python.

Tracking

Kept an Excel test tracker (feature, test case, owner, status, pass or fail, bug evidence) and ran scrum-style task planning with the founders from August to October 2025.

Result

About 80% less manual effort (internal estimate).

Understand → Validate → Structure → Execute → Learn

Understand

Start with the customer, workflow and business context before jumping to a solution.

Validate

Look for behaviour, urgency and willingness to pay rather than relying only on stated interest.

Structure

Turn an ambiguous problem into hypotheses, priorities, owners and measurable next steps.

Execute

Work across teams and functions instead of stopping once the analysis is complete.

Learn

Bring customer and market feedback back into the next product or GTM decision.

Worked closely with founders. Here's what they said.

Recommendations from the leadership team I worked with at Ninety North / ywork.ai.

“

He spoke to 50+ UK real estate professionals, identified a real market pain, and came back with 4 paying clients — before we had built a single line of code. That model became the foundation of ywork.ai. What sets him apart is his ability to identify what needs to be done and execute without being told.

Nishant Singh

Founder & Director — ywork.ai · Ninety North Ltd UK

View signed recommendation ↗

“

He was not just a support function. He was a thinking partner. Somesh partnered with me on designing our target customer segments, building outreach strategies, and executing our product launch. He approached every challenge the same way — understand the problem deeply, build a clear plan, and execute without waiting to be told.

Nisha Singh

Co-Founder & Director — ywork.ai

View signed recommendation ↗

Broad enough for early-stage teams. Focused enough to execute.

My experience spans multiple functions, but the common thread is understanding problems, creating structure and moving work forward.

Product & Strategy

Customer Discovery Market Validation Workflow Analysis Requirements Clarification Competitor and ICP Research Feedback Synthesis Zero-to-One Validation Product Documentation

GTM & Commercial

B2B SaaS GTM Product Demos Early-Stage Sales HubSpot CRM Apollo.io Hunter.io LinkedIn Outreach Webinars

AI & Analytics

Prompt Testing LLM Output Evaluation BOQ Accuracy Testing OpenAI Claude Excel SQL Tableau Python (foundational)

Cross-Functional Execution

Jira, Confluence, Mixpanel (working knowledge) Exploratory Testing Test Tracking Scrum Coordination SOPs Hiring Interviews Onboarding Events

Treelync

An AI mock-interview platform for BPO roles that I built on my own: 10 role categories with 1,200+ questions, 20-question timed sessions, speech transcription and AI scoring on accuracy, fluency and tone. Built with Next.js, Supabase, Groq Whisper and the OpenAI API.

Looking for problems worth solving.

I'm looking for product-focused Business Analyst, Product Operations and Associate Product Manager roles in B2B SaaS and AI. Open to relocation.