I ship AI products at work. I build them by hand on weekends.

The setup: 4 clients wanted non-employees to raise Helpdesk tickets. The CTO rejected it — “Helpdesk is for employees.”

Instead of dropping it, Arjun interviewed 14 clients and found the real need:

• Ex-employee background-verification requests were running on untracked email threads
• Reframing the persona from “external stranger” to “ex-employee” changed the conversation
• The CTO raised spam and pricing concerns — the spam one Arjun admits he’d missed
• He came back with whitelisting, rate limits, and credit-based pricing

The result: approved, shipped, now one of the module’s most-used features — and a new revenue channel.

An org-wide UX revamp reached Helpdesk: 3 personas, 100+ configurations, 7 pages — and a real cost to getting it wrong.

Instead of restyling, Arjun mapped jobs-to-be-done per persona and tested each against usage data. The data overturned his assumptions:

• 30% of “admins” actually worked as agents too → one unified interface, not two
• 95% of agent time sat on 2 of 7 pages → a discovery problem, not a page problem
• The top job — catching about-to-breach tickets — had no UI support at all

Three months after launch: breach rate 10% → 7%, average resolution 70 → 58 hours.

The service-delivery SKU had stagnant ARR — and no owner for the problem. Arjun went looking.

The find: Employee Relations — grievances, investigations — running on spreadsheets and email at enterprise clients, with only 3–4 focused vendors in a market expected to double by 2035.

• Ran 11 client interviews and a competitive teardown of HR Acuity, Dovetail, AllVoices, ServiceNow
• Vibe-coded a clickable prototype with AI tools — stakeholders used it, not slides
• Took the case through CPO, sales, and finance
• Descoped anonymity himself to halve the release timeline

Greenlit at $500K ARR potential; engineering is building it now.

The module’s largest-ever client engagement: Arjun visited clients across 11 cities, surfacing 100+ product gaps from 39 organisations varying wildly in size, industry, and HR maturity.

The discipline was in what happened next — prioritising by usage metrics, not by who asked loudest.

The loudest request is rarely the most valuable one; the data usually knows better.

Weekly clinic retention was stuck at 10% despite 12,000 doctors onboarded.

Arjun visited 40+ clinics in 3 cities and found the real problem: 80–85% of patients were walk-ins — and the product only did week-ahead bookings.

• The engineering lead pushed back on the architectural change
• Arjun moved the conversation from “is this worth the effort” to “what’s our plan for the retention goal you own”
• Brought him into scoping the MVP instead of handing him a spec

A 40-clinic pilot, then 300 clinics. Retention: 10% → 65%.

A month into Phable, Arjun needed the health-programs lead for knowledge transfer — and got a cold shoulder.

Rather than escalate a complaint, he did the digging himself: delivery managers, the actual spreadsheets, customer communications, customer interviews. Then a findings brief to the CPO and a joint meeting.

Work moved; a three-sprint roadmap came out of it.

His honest read: the initial friction was partly self-inflicted — he’d leaned on one person instead of doing his homework first.

The brief: validate a full-stack program for fresh graduates in 10 business days, with no engineering support.

Shipped in 3 days, all no-code: an Unbounce landing page he built himself, GoToWebinar demos, Zapier gluing registrations into Google Sheets, a Razorpay payment page.

• First month: ~47 enrollments at ₹10K each
• Registration conversion 8% → 13%, attendance 30% → 42%, payment 2% → 3.1%
• Six months later: ~₹50L a month — roughly 10×

Only then was it rebuilt properly, on real infrastructure.

The AI/ML postgraduate program page was underperforming, and sales kept fielding the same questions.

Before redesigning anything, Arjun did the homework:

• 15 customer interviews
• 300+ sales call recordings across 70 buyers
• 7 sales and ops rep interviews

The finding: ~90% bought because of the university collaboration — which the page barely mentioned. The rebuilt page led with credibility, validated in a 30-day 50/50 A/B test.

Traffic-to-lead 3.6% → 4.8%; hot-lead-to-sale time down 30 → 20 days.

A freemium self-learning portal looked promising: sign-up conversion nearly doubled (5.2% → 9.9%) in a 4-week experiment, and product logins jumped from 19% to 60%.

But the truth was underneath:

• Week-1 retention: 20%
• Demo bookings: under 1%, vs 30% in the regular flow
• A 300-parent pilot said it plainly — parents wanted a human tutor matched to their child, not self-serve software

Arjun scrapped the model he’d championed. Numbers that flatter acquisition can still describe a product nobody needs.

A partnership with Brainly targeted North American math learners — 15–20 leads a day sounded like a win.

Reality:

• ~60% junk rate
• Poor downstream demo conversion vs every other channel

Arjun shut it down with a rule attached: acquisition channels without an upfront qualification step just relocate the cost to sales. The lead count was never the metric that mattered.

Construction: a domain Arjun didn’t know, users who’d never used software, no baseline metrics.

He started with head-office ERP — approvals from weeks to days, ₹400+ crore of monthly orders processed — then took a team of 10 to the real problem: the sites.

• TrackZ: 40% of received material tracked through weighbridges
• KwartZ: lab-test calculations cut from 30 minutes to 3
• Safety Eye: 2,100+ QR-reported issues resolved, resolution down from 25 days to 5
• Fatalities: zero

Honest footnote: site-team adoption and external developers’ timelines were a constant grind.

Tech Mahindra had booked rotating batches of freshers across 30–45 days when a power outage knocked out rooms mid-stay. Their Admin Manager questioned whether Treebo could handle corporate business at all.

Arjun’s response, same evening:

• Learned the outage would last 3+ days — no waiting it out
• Got approval for transport, moved guests to better rooms at another hotel
• Stationed a sales manager on-site for the logistics
• Apologised in person the next morning

The account stayed. Trust isn’t rebuilt with explanations — it’s rebuilt with speed.

Managing 1,500+ enterprise leads, Arjun bucketed them by potential, bottleneck, and history — and found a theme: dozens of accounts stalled on the lack of Bill-to-Company credit.

Rather than lobby, he built the case:

• Quantified the total revenue trapped behind the constraint
• Proposed tight terms to cap the risk — next credit extension only if the previous payment cleared in 15 days

Exception approved; stalled pipeline moved. His sales years are why every product pitch he makes comes with a business case attached.

Arjun’s own startup is his most instructive failure — and he doesn’t dress it up.

• Spent 8 months building a hangout-planning app before real validation
• Research was designed to confirm the idea, not test it
• Post-launch retention collapsed; he doubled down on features — the wrong diagnosis, because the problem was never acute enough to change behaviour

The rule he kept: if people aren’t already using some imperfect workaround, the pain won’t pull them to a new tool. Every feature request he triages today goes through that lens.

Six months in, the team kept polishing and postponing. NIT Warangal’s tech fest was coming — real users, real events, founders physically present to watch people use the product.

No data supported launching early; the blocker was fear of looking bad.

Arjun reframed it: the real failure isn’t looking bad — it’s making a bad product and not learning.

They launched at the fest, and that weekend’s lessons triggered the pivot that defined the company’s second act.