Senior Product Manager - AI Platforms - B2B SaaS

I turn customer insight into platform bets that pay.

12 years building B2B SaaS in complex, high-stakes markets. For the last 4 years I've led AI platform product at Granicus: 0 to 40+ enterprise customers, launch cycles cut from 14 weeks to 5, $2M in unlocked revenue.

Read my case studies Download CV

Platform bets in AI, enterprise SaaS, and growth, taken from discovery to launch, with the numbers to prove it.

Email Me +91-9560283111 LinkedIn

Bengaluru, India - Open to Remote

Career history

Experience

Senior Product Manager, AI Platforms - Granicus (B2B SaaS, GovTech)

Bengaluru (Remote), India - Nov 2021 - Present

Current

Modular agent platform scaled from 0 to 40+ enterprise customers

Cut AI use-case launch time from 14 weeks to 5 by modularising the AI agent (GXA) into reusable chunking, embedding, and access-flow components.
Pitched and shipped centralised SSO after discovery flagged sign-in friction as the top adoption blocker, unlocking $2M in incremental revenue.
Owned the AI roadmap for the Customer Data Platform, turning 30+ enterprise customer interviews into bets that lifted engagement 30% across 500K+ users.
Killed a fine-tuning track after evals showed retrieval drove 80% of failures, redirecting the squad to reranking and lifting containment from 20% to 41%.
Redesigned customer service routing across CRM, messaging, and knowledge-base surfaces, cutting manual ops 30% and resolution time from 3 weeks to 2 weeks.
Defined the consent, access-control, and audit-logging model for AI workflows, making the assistant sellable into FedRAMP and GDPR-bound enterprise accounts.
Built the AI quality framework tracking retrieval accuracy, deflection, and adoption with per-answer observability and human review on edge cases, the org's first shared definition of a working AI feature.

Associate Product Manager - Maropost (Marketing Automation)

Punjab, India - Jun 2020 - Oct 2021

+$3M ARR

Grew ARR $3M through 12 growth experiments

Grew ARR by $3M in a single quarter by running 12 A/B tests across onboarding and pricing, lifting signup-to-paid conversion 35%.
Shipped a conversational engagement product from concept to GA for 50K+ SMB customers, driving 30% more interactions and 15% less support volume.
Rebuilt PLG onboarding after 20+ discovery interviews, lifting activation and CSAT 60%.
Sequenced the roadmap across 3 squads, turning customer and sales signal into prioritised bets and cutting sprint carry-over 25%.

Product Analyst - RxAdvance PBM India (Healthcare)

Noida, India - Apr 2019 - Jun 2020

+30% engagement

Lifted patient engagement 30% on regulated platform

Lifted patient engagement 30% and renewals 20% by scoping a patient monitoring product around third-party risk-scoring models.
Delivered 5 features across 2 squads in 6 months on a HIPAA-regulated stack, while addressing all audit findings on data governance.
Uncovered workflow gaps through discovery with care managers and payers across 3 accounts, resequencing the roadmap around real clinician needs.

Product Analyst - OATI (Enterprise Energy & Utility SaaS)

Mohali, India - May 2016 - Apr 2019

WebEIM - 200+ accounts

WebEIM scheduling automation across 200+ utility accounts

Translated scheduler and dispatcher workflows on WebEIM (energy scheduling and imbalance-settlement automation for utilities) into roadmap input across 5 releases, lifting CSAT 50%.
Productised WebEIM onboarding into a standard configuration flow, shortening time-to-go-live 20% and support tickets 30%.
Translated compliance mandates into shippable roadmap scope for a regulated energy SaaS suite, across 3 engineering pods.

Associate Analyst - Google (via Binary Semantics Ltd)

Gurugram, India - Mar 2014 - May 2016

+40% velocity

Drove Scrum transition, 40% velocity gain

Accelerated delivery velocity 40% and time-to-ship 30% by moving a 30-person team from waterfall to Scrum.
Mined user behaviour data to surface content quality signals, formalised into guidelines adopted by a 200+ person org.

Selected work

Case studies

Seven shipped bets across AI platforms, enterprise SaaS, and growth. Each card shows the problem, what I did, and the measurable outcome.

Granicus, AI Platforms, 2021-Present

14 wks to 5 wks, 0 to 40+ customers

Modular agent platform

Problem: every AI use case was a bespoke build, so each launch took ~14 weeks and the roadmap could not scale.
Research: mapped repeat patterns across use cases with eng + design; identified reusable components (retrieval, orchestration, guardrails).
Solution: split the agent into modular components with shared evals and launch checklists; sequenced migration by impact.
Impact: new use cases ship in 5 weeks; scaled from 0 to 40+ enterprise customers.

Platform StrategyRoadmapModularisation

Granicus, Growth, 2021-Present

$2M incremental revenue

Centralised SSO

Problem: sign-in friction blocked adoption across enterprise accounts.
Research: discovery across enterprise admins flagged SSO as the top adoption blocker, ahead of feature asks.
Solution: pitched SSO over feature work; shipped centralised SSO with GTM and rollout plan.
Impact: $2M incremental revenue unlocked.

DiscoveryPrioritisationGTM

Granicus, Customer Data Platform

+30% engagement, 500K+ users

Interview-led CDP roadmap

Problem: scattered asks across CRM, messaging, and knowledge surfaces with no clear sequencing.
Research: 30+ enterprise customer interviews; sized opportunities by reach and impact.
Solution: sequenced roadmap across CRM, messaging, and knowledge; defined engagement success metrics up front.
Impact: +30% engagement across 500K+ users.

DiscoveryPrioritisationAnalytics

Granicus, AI Quality

Containment 20% to 41%

Evals over fine-tuning

Problem: AI answers were inconsistent; the default ask was expensive fine-tuning.
Research: built a golden set from past tickets and scored retrieval, grounding, and refusal quality separately; retrieval drove ~80% of failures, not the model.
Solution: killed the fine-tuning track; redirected the squad to chunking, reranking, and retrieval quality, gating launch below the eval bar; every answer logged with sources and confidence, edge cases queued for human review, new misses turned into new evals.
Impact: containment doubled from 20% to 41% without fine-tuning cost.

ExperimentationEvalsObservabilityTrade-offs

Maropost, Growth, 2020-2021

$3M ARR in one quarter

Onboarding and pricing experiments

Problem: signup-to-paid conversion lagged across 50K+ SMB customers.
Research: funnel and cohort analysis to find the highest-drop steps.
Solution: ran 12 A/B tests across onboarding and pricing; shipped winners with success metrics defined per test.
Impact: conversion up 35%; $3M ARR added in one quarter.

A/B TestingPLGFunnel

OATI, Energy SaaS (WebEIM), 2016-2019

CSAT +50% across 5 releases

WebEIM scheduling automation

Problem: utility schedulers managed energy needs across fragmented scheduling and settlement workflows; every WebEIM rollout became custom work.
Research: mapped scheduler and dispatcher workflows across 200+ accounts; NPS and ticket signal pointed at configuration, not features.
Solution: productised WebEIM onboarding into a standard configuration flow with compliance scope built in.
Impact: CSAT up 50% across 5 releases; time-to-go-live down 20% and support tickets down 30%.

WebEIMOnboardingEnergy SaaS

RxAdvance, Healthcare, 2019-2020

+30% engagement, +20% renewals

Patient monitoring on a HIPAA-regulated stack

Problem: patient engagement lagged and renewals were at risk across a HIPAA-regulated platform.
Research: discovery with care managers and payers across 3 accounts; scoped around third-party risk-scoring models.
Solution: scoped and sequenced a patient monitoring product around clinician needs; delivered 5 features across 2 squads in 6 months with audit findings on data governance addressed.
Impact: patient engagement up 30% and renewals up 20%.

DiscoveryHIPAARoadmap

How I work

AI PM toolkit: how I actually build

I work like an AI PM, not a feature manager. Discovery with users, then evals, then ship. I decide what the model should do, what it must never do, and how we know it worked — with observability on every answer and a human in the loop where stakes demand it. The cards below are my real stack. Tap any card to see how I use it on real work.

My default starting point

Evals before models

I write the test before I touch the model. Golden sets, grounding checks, containment and accuracy — with observability on every answer and human review on edge cases. At Granicus this call moved containment from 20 to 41 percent with zero fine tuning.

Tap to see my eval loop +

Where most AI fails

Retrieval and grounding

Most wrong answers are retrieval misses, not model misses. I work chunking, embeddings, reranking and citations until the answer can prove where it came from.

Tap to see the method +

What the model must never do

Guardrails and safe UX

I design abstain and route flows, confidence thresholds and human review for edge cases. In regulated accounts a clean refusal beats a confident wrong answer.

Tap to see the pattern +

From demo to platform

Agents and platforms

I turn one-off demos into reusable pieces. At Granicus I split the agent into chunking, embedding and access flow parts, which cut launch time from 14 weeks to 5.

Tap to see the split +

Proof, not vibes

Experimentation and analytics

I run growth like science. At Maropost 12 tests in a quarter added $3M ARR. Funnels, cohorts, SQL, and a log so we never retest a settled question.

Tap to see the playbook +

Why enterprise buys

Enterprise readiness

SSO, consent, access control, audit logs, residency. I shipped centralised SSO for $2M in revenue because sign in friction was blocking every big deal.

Tap to see the checklist +

Teardowns, click a card

Teardown: Intercom Fin, read as an outsider

A teardown of a product I do not own, in four parts. Each card opens the full breakdown: who it serves, where it breaks, what I would fix first and how I would measure it.

Read the problem +

1. Problem and persona

Product: Intercom Fin, an AI support agent sold to B2B SaaS support teams. Read from public docs and behaviour, not insider knowledge.

Persona: support ops lead at a 500+ seat B2B SaaS, owns CSAT and handle time, fears wrong answers more than slow answers.

Read the journey +

2. Journey and competition

Journey: ticket arrives to bot attempts answer to low confidence to human handoff to resolution logged to bot learns (or does not).

Competition: Zendesk AI optimises containment; Sierra and Decagon sell accuracy plus evals. My read: eval-gated answers with cited sources beat raw containment in regulated markets.

Read differentiation +

3. Differentiation and pricing

Differentiation: retrieval-grounded answers with confidence bands. Abstain-and-route beats confident-and-wrong where compliance matters.

Pricing: per-resolution pricing aligns cost with value; enterprise tier adds audit logs, residency, SSO.

Read my top 3 fixes +

4. Top 3 improvements I would ship

Eval-gated launch checklist (high impact / low effort) block launch below containment + accuracy bars. Metric: % answers with citations; containment.
Abstain-and-route UX (high / medium) show confidence, route gracefully. Metric: reopen rate, CSAT on bot conversations.
Closed-loop learning inbox (medium / medium) agents review edge cases weekly. Metric: week-over-week eval pass rate.

Method only: this is an outside-in read of a third-party product. My own comparable AI support work is written up under Case studies. Want this lens on your product? Reach out and I will walk you through it.

Foundations

Education

B.Tech, Electrical and Electronics Engineering

SSCET, Pathankot (Punjab Technical University), 2007 - 2011

Certifications

Certified Scrum Product Owner (CSPO)2023
AI and Agility Certification2024
AI Security Certification2025
FedRAMP and Cloud Compliance Training2026

Open to what's next

Let's build AI users can trust.

Senior PM roles in AI platforms, agents and enterprise SaaS. Bengaluru or remote.

Email Me LinkedIn CV

asambyal23@gmail.com - +91-9560283111 - Bengaluru, Open to Remote