I'm Saurav, a product leader who builds things people come back to.

Building every service vertical in-house (car rental, laundry, home cleaning, pet grooming, gardening, e-bikes) would never pay back. Careem's vision was one app for everything, in every market.

Open the platform instead of funding verticals. Any provider, once vetted by Careem and agreed on commercials, could sell through Careem. Fulfilment stays with the partner.

• Partners plug into Careem Pay, discovery, the design system and the existing customer funnel.
• Vetting and commercial agreement as the gate to listing.
• Rolled out market by market: Dubai, then Jordan, then Egypt.
• Built lasting partnerships beyond Swapp, including Justlife in home services (announced by Careem) and Washmen in laundry.

Swapp had powered car rental on Careem since 2022, but by the time I joined the partnership was strained. I took it over, rebuilt the joint roadmap with them, and in six months the business looked very different. Once a proof of concept on the new model was confirmed, I took the partnership to an exclusivity deal with equity for Careem: Careem's first stake in one of its partner apps, announced in 2025. Careem now drives about 96% of the partner's GTV.

20+ partners in Dubai, 15 in Jordan and 7 in Egypt within six months. Careem went from about 5% of partner business at launch to about 96% of a car-rental partner's GTV and about 92% of a home-services partner's. Customers who adopted a partner service tripled their spend on the app (45%+ of adopters).

While running a care-automation programme, the share of agent-only tickets rose to about 70% at month end, driven by invoice requests, in the UAE, KSA and Jordan alike. Careem issued receipts, not tax invoices.

Interviews with 15+ users across three markets found employer reimbursement of 20 to 40 trips and orders a month. The surprising segment was real-estate agents. The idea had been shelved before, so the case had to prove demand first.

A fake-door page I built on Lovable, shown only to the cohorts asking for invoices. It drew 200+ forms from about 70 companies in two weeks, mostly 100 to 400 staff, with typically only 20 to 30 employees per company already on Careem. That gap was the expansion case.

• An admin workspace where HR or finance set spend policies per service and keep the employee list current.
• A company-funded, preloaded wallet with allowances that expire daily, plus rules such as dinner only after 10 PM in the office.
• A "pay as business" toggle at checkout, so the employee never files an expense.

Launched lean: a Lovable site, an ops dashboard and spreadsheets until traction was proven. I sold the first accounts myself, with no sales, TPM or commercial support. After the business crossed $200K a month, three months in, it got its first sales rep, then moved to the commercial org: build, operate, transfer. It now runs with its own sales, invoicing and compliance team, and grew beyond the UAE and KSA into Jordan and Egypt.

New business accounts as the north star. Revenue is the output, not the target.

Quick commerce trains people to top up, not stock up. Raising order value was the one goal every vertical declined at the company hackathon I created. My team took it.

• Evaluated about six form factors, including saved carts and category-paged flows. Chose a single-page list on a cognitive-load argument, and won the internal debate with a prototype I built myself.
• Refused the obvious lever, a big-basket discount, so customers would not learn to wait for discounts. Used country-of-origin tags to anchor quality instead.
• Shipped image input first and deferred voice, because conversational input cost more per interaction at the time.

Users uploaded list photos and internal agents transcribed them by hand. The LLM replaced the humans only after it matched that baseline. Out-of-stock items are shown up front with a "back in stock tomorrow" tag, to protect trust.

22% week-one adoption against a 5 to 7% target, 65% of adopters repeating within two weeks, adopter order value from about 70 to about 250 AED. A falsified hypothesis (people kept browsing aisles after finishing their list) became a second product: a "people also buy" strip at checkout. Uber and noon shipped equivalents within about two months.

A retrieval assistant over Confluence, Notion and Drive, inside Slack, that raises Jira tickets and can run a workflow on click. Agent actions were tested against a shadow replica of the production stack before launch. Documentation regenerates into the index whenever a feature ships, and thumbs up or down feedback runs after rollout.

Rebuilt the merchant platform after benchmarking Keeta and DoorDash, for four roles: owner, manager, restaurant staff and kitchen staff. AI menu scoring, categorisation and ranking, image quality checks and a dynamic pricing engine lifted merchant revenue 24% and cut onboarding from 6 days to 1.

Built an analysis skill (cross-analysis, dependencies, leakage detection) that is published on Careem's internal skill marketplace and used across the company.

A company-wide hackathon where each vertical submitted its own top problems against the company's goals, so priorities came from the businesses rather than from me.

Patients carried paper files between hospitals that would not accept each other's reports, so tests were repeated and no doctor saw the full history.

Five partner groups, each with a different reason to join:

• Government: one of the first five companies admitted to the national ABHA sandbox.
• Insurers: the payer side of the network.
• Hospital chains: 8 hospitals in the pilot.
• Diagnostic labs: reports delivered straight into the patient's record instead of over WhatsApp.
• Private clinics: 40 clinics, with strong reach into tier 2 and tier 3 cities from Jamshedpur.

For patients, one longitudinal record with time-limited consent per doctor. For doctors and labs, a patient-history and reach-out tool. Zealth was deliberately not a provider, payer or insurer, but the neutral layer a health marketplace could be built on.

Doctors and labs adopted it, and the pilot earned revenue. When the government deprioritised ABHA, I shut Zealth down rather than spend limited capital waiting. The lesson I carry: a platform built on someone else's rails needs a plan for when those rails move.

Piloted formats against each other (real-money, hyper-casual and strategy, web3 and non-web3) before committing. Hike's strengths in matching, relays and concurrency pointed to player-versus-player hyper-casual games.

Web3 tokens worked as game mechanics, but interviews showed players did not care about owning NFTs. We kept the mechanics (collect, upgrade and morph pawns) and dropped the web3 cost.

Two currencies, a pay-to-play loop for the freemium tier converting 8% of new users to paid each month, and Rush VIP tiers that raised spend 30% and D30 retention from 12% to 30%. Made A/B testing and pricing governance the default path to ship, improving margins 20%.

Led a 230-person cross-functional team through the pivot in 6 months to $100M in recurring revenue, profitable from day one, reporting to the CEO, Board and investors.

As single-threaded owner of seller success, I set the goal as doubling onboarding success, measured on a seller's first 100 orders.

• Mobile-first seller onboarding and gamification for the micro-seller base behind 80% of India volume, later scaled globally.
• "Top Brands", a product-page trust signal (buyers this month, positive reviews, return rate) so buyers could tell real D2C brands from fakes. Brand revenue rose 200% at 3% higher conversion.
• A quality engine on 30+ signals that blocked fraudulent sellers and listings, cutting returns 25%.

80 to 85% of peak orders were within about 3 km, and serviceability rules switched off long-distance restaurants at peak. Requiring a motorcycle was an artificial filter on supply, while milk and newspaper delivery riders on cycles sat idle at lunch and dinner.

Higher rider incentives, surge-style customer pricing, restaurant self-delivery, or new supply. I chose new supply, where money had an inverse relationship with partners.

• Order assignment, made multimodal and aware of order size, volume and intensity.
• Partner profiles, which assumed a driving licence and vehicle.
• Rate cards and payouts (see the next case).
• Guardrails: no night orders for women riders, no multi-floor buildings for riders with disabilities. Delivery-partner insurance, introduced for the first time because no underwriter covered cyclists.

A customer-facing "cyclist or biker" choice raised complaints while measured ETAs differed by only 30 to 45 seconds, so I kept it hidden. The first assignment rule angered bikers waiting outside restaurants, so a fast-follow made assignment concentric: cyclists first close in, bikers otherwise.

• Phase one enabled multimodal payouts. Later phases added incentive tiers by partner type, then cohort-level rate cards.
• Part-time rate cards: a rider online for two or five hours never sees a 14- or 50-order milestone they cannot reach. Full-day riders are not capped at 10 orders.
• I wrote the requirements; the payments team, then in the finance org, built against them, and payouts later moved into fulfilment.
• Riders engaged more, earned more and stayed longer.

Built Zomato's first promotion platform in 2018 and 2019. Bank, wallet, partner and Gold offers stacked automatically in one cart with a single total, instead of making customers do the maths. The same platform set the structure for platform fees, which could be added, removed and targeted by cohort.

Deployed the Karma Score ML fraud system, cutting platform abuse 90%, and a dynamic pricing engine that improved margins 18%. Unit economics on a 250-city P&L moved from -20% to +16% EBITDA on a $20M budget.

Travellers, the local travel agents who sold and planned the trips, and the on-ground providers who delivered them. I structured all customer feedback by agent, location and itinerary, so we could nudge travellers before known on-ground problems hit, and set expectations where nothing could be fixed.

When one person in a group marries, many more follow over the next few years. We built a referral reward mechanism and a loyalty rewards programme, and made trips generated per converted traveller the north star. One conversion translated into about 20 trip conversions.

Weekend getaways near Mumbai, Pune, Bengaluru and Delhi gave travellers a reason to come back every year, and a place to redeem referral points. At one point we sent more than 8,000 people on offbeat trips such as Rishikesh.

A retargeting and lead-scoring engine on 25+ profile attributes, and a 40-person CX team, improved lead-to-conversion 23%.

New long-tail shippers moved to a prepaid-only model until they reached a set number of orders, through a closed B2B wallet. It recovered $3M in unrealised revenue and cut leakage 95%.

Indian addresses are free text. AddFix clustered variants of the same address by learning from successful deliveries, giving routing and capacity planning a structured location. It was later offered to Amazon as a service.