# Michael Novack: Director of Product, Lyft Business > Michael Novack is Director of Product at Lyft Business, based in Toronto. Before Lyft he spent fifteen years in product at Hims & Hers, Dropbox, Agtonomy, Serve Robotics (Postmates), Just Eat Takeaway, PwC and WWF. Open to Head of AI Product and VP Product roles at AI companies and AI-native consumer products. Full-time only. *Updated 2026-10-07. Canonical: https://builtbymikey.com* This is the long version of `/llms.txt`. Everything here comes from the same data as the website, so the facts match the pages. ## Identity - **Name:** Michael Novack (also Mike Novack, Michael J. Novack) - **Current role:** Director of Product, Lyft Business, since December 2025 - **Looking for:** Full-time Head of AI Product or VP Product roles at AI companies and AI-native consumer products. Not taking fractional, consulting or advisory work. - **Location:** Toronto, Ontario, Canada. Commutes YYZ–SFO regularly and is open to relocating. - **Email:** hi@builtbymikey.com - **Book a call:** https://www.builtbymikey.com/schedule - **LinkedIn:** https://www.linkedin.com/in/michaeljnovack/ - **GitHub:** https://github.com/michaelnnovack - **Resume (markdown):** https://builtbymikey.com/resume.md - **Structured profile (JSON):** https://builtbymikey.com/api/profile - **Now:** https://builtbymikey.com/now - **Notes:** https://builtbymikey.com/notes ## Career summary Michael has worked on both new products and products at scale. On the new side: founding team product at Agtonomy (the TeleFarmer operator app and the TrunkVision computer vision spec) and the Level 4 sidewalk autonomy product spec at Serve Robotics. At scale: the courier network behind 1B+ Just Eat Takeaway deliveries, the GLP-1 launch at Hims & Hers ($100M+ incremental revenue in its first year) and multi-product adoption at Dropbox ($8M+ expansion revenue). Much of that work involved models whose output varies from run to run. On those products he owns the product side of the model: what counts as a correct answer, the eval set that checks it, which mistakes block a release, and what the user sees when the model is unsure or wrong. At Hims he owned the labels and F-beta target for an intake classifier and built the offline test set and release check. At Serve he designed the regression test sets, the safety thresholds and the release check at the depot. He works with data science and engineering on the model itself and does not claim engineering scope. He also builds. Field Kit and Ask Enzo on this site are his own work, built with the Anthropic Claude API (Sonnet 5.5 and Haiku 4.5) and the Vercel AI SDK. ## How he works, in his words ### Product Strategy *Turning user problems into product decisions that actually matter* Strategy starts with what users actually struggle with, not what we assume they need. The hard part is saying no to good ideas so the team can execute the great ones. ### AI Product Development *The eval set is the product spec* In probabilistic systems you can't write a deterministic spec, so the eval set defines what the model must get right and which failures you refuse to ship. Trust is the product, which means users should understand what happened and fix it cheaply when it misses. ### Growth & Scaling *Building repeatable systems that compound over time* Growth isn't a hack. It's finding the moment users get value and removing every obstacle in front of it. Durable growth comes from products people genuinely want to use, not clever tricks. ### Team Leadership *Getting out of the way so teams can do their best work* Set direction, not tactics. Make the problem clear, explain why it matters, then let the team find the how. Trust is built by surfacing problems early and planning the recovery, not by projecting false confidence. ### Business Operations *Removing the friction that slows teams down* Most operational problems are communication problems: too many meetings, unclear ownership, decisions that need five approvals. I find where work gets stuck and simplify until the path is obvious, because the best process is the one nobody notices. ### Strategic Finance *Using numbers to make better decisions, faster* Finance isn't spreadsheets. It's understanding tradeoffs: this feature or that market, the real cost of moving faster. I build models that answer those questions clearly, so teams can decide with confidence. Clarity beats precision. ## Domains - **AI and machine learning:** LLM products, classification, recommendations, computer vision, Level 4 autonomy, evals, structured output. - **Marketplaces:** two- and three-sided marketplaces, balancing supply and demand, scaling across countries. - **Payments and fraud:** payment flows, fraud rules, pricing and incentives. - **Autonomous systems:** fleet tools, remote operator experience, real-world deployment. - **Product-led growth:** self-serve onboarding, activation, analytics that tie usage to revenue. ## Case studies Each case study matches the page at `/work/{id}` and is written in his own words, in the first person. ## Lyft: Business Travel as a Competitive Advantage **Role:** Director of Product Management · **Updated:** 2026-05-01 · **Link:** https://builtbymikey.com/work/lyft-business-travel · **Company:** https://www.lyft.com/business > Architecting a global B2B platform, and the AI tooling that lets a small team run it ### The problem Lyft Business spans a global portfolio of B2B use cases, business travel, enterprise rideshare, and corporate transportation benefits, but the platform underneath wasn't built to support that breadth. Every new vertical risked becoming its own one-off build instead of an extension of a shared foundation: - Ground transportation spend was hard for finance and travel teams to see or control, so companies defaulted to fragmented booking tools and ad hoc ride choices - Each enterprise use case, business travel, corporate rideshare, event transportation, threatened to duplicate the same pricing, provisioning, and dispatch logic instead of sharing it - The product organization needed to cover a widening portfolio of markets and verticals without scaling headcount at the same rate ### What I did Leading product strategy and team management for Lyft Business, architecting a platform that serves the whole portfolio rather than one vertical at a time: - Platform architecture: building the core capabilities, pricing, provisioning, benefits administration, once, then layering in the specific functionality each vertical actually needs on top of it, rather than rebuilding shared logic per use case - Portfolio and team leadership: managing a team of product managers across the portfolio, setting scope and strategy for business travel, enterprise rideshare, and adjacent enterprise use cases while keeping the underlying platform coherent - AI-enabled team velocity: built the team a shared AI toolkit, a central repo of reusable skills, plugins, and automated workflows, that PMs use to research, spec, and ship, so the team covers more scope without adding headcount ### What happened Currently in progress. Early focus has been on the platform foundations and team capabilities that compound as the portfolio scales: - **Platform foundation:** shared core capabilities now underpin multiple B2B verticals instead of duplicated, one-off builds - **Team enablement:** the shared AI toolkit is in active use across the product management team, expanding the scope each PM can own - **Portfolio strategy:** set the scope and roadmap for business travel, enterprise rideshare, and adjacent enterprise use cases under a single platform strategy ### At a glance - Leading product strategy across Lyft's global B2B portfolio - Managing a team of product managers spanning business travel, enterprise rideshare, and more - Built a shared AI toolkit (skills, plugins, automated workflows) adopted team-wide - Architecting one platform to power multiple enterprise verticals ### Methods `Platform Architecture`, `B2B Portfolio Strategy`, `Team Leadership`, `AI Tooling & Automation`, `Product-Led Growth`, `Enterprise Solutions` ## Hims & Hers: GLP-1 Launch & Platform Modernization **Role:** Senior Director of Product · **Link:** https://builtbymikey.com/work/hims-glp1-platform · **Company:** https://www.hims.com > Launching weight loss GLP-1s while rebuilding payment and personalization systems ### The problem Hims & Hers was growing quickly, initially focused on hair loss and erectile dysfunction medication, but was starting to lean into GLP-1s. Our platform wasn't built to support this yet. The demand was high, but we faced serious challenges: - Fraud was skyrocketing as treatment costs jumped from $30 to $1,500, threatening our ability to serve legitimate customers at scale - Payment conversion was tanking because our fraud rules were too aggressive, blocking real customers who wanted treatment - We couldn't personalize experiences, so every customer got the same generic flow regardless of their needs or likelihood to convert - Every pricing experiment required custom engineering, making it impossible to move fast and find the right price points for different customer segments ### What I did I led three platform initiatives that ran in parallel: - Payment funnel optimization: Split the payment flow into two separate paths. One optimized for fraud detection on high-risk transactions, another for conversion on everything else. This let us catch fraudsters without punishing legitimate customers. - ML-powered personalization: Partnered with DS/ML to ship intake-time intent classification. Owned the label rubric (high-intent vs. exploratory) and tuned the F-beta target so the funnel weighted precision higher than recall — false positives push exploratory users into a fast-checkout flow for a regulated medication, so the cost asymmetry is real and quantifiable. Designed the cold-start fallback for users with too little signal to classify, and built the offline regression eval set + online deployment gate that blocked any model iteration that regressed conversion, LTV, or downstream clinical-appropriateness signals. The UX translated model confidence into shorter or longer onboarding — high-confidence users went straight to checkout, low-confidence users saw more educational content to build trust. - Dynamic pricing infrastructure: Launched a CDP-powered incentives system that automatically served targeted discounts based on customer data. This unlocked rapid pricing experiments across different segments without needing engineering for each test. ### What happened The platform transformation enabled Hims & Hers to successfully enter the GLP-1 market while building infrastructure that would support the company for years: - **Market entry:** The GLP-1 launch generated over $100M in incremental revenue in its first year, establishing Hims & Hers as a legitimate player in the weight loss medication space - **Payment economics:** We solved the fraud-vs-conversion tradeoff by reducing fraud 23% while simultaneously improving conversion 8%. The key insight was that you don't need to sacrifice one for the other with the right architecture - **Personalization:** Personalized onboarding lifted conversion 18% and lowered acquisition costs, but more importantly, it made the experience feel tailored rather than generic - **Pricing velocity:** Dynamic pricing drove 31% improvement in lifetime value. We could test new pricing strategies in days instead of months, running experiments 40% faster than before - **Platform foundation:** These weren't just one-time wins. The platform capabilities we built became the foundation for every future product launch at Hims & Hers ### At a glance - $100M+ incremental revenue from GLP-1 platform - 23% reduction in fraud while improving payment conversion 8% - 18% lift in onboarding conversion via ML personalization - 31% improvement in LTV through dynamic pricing experiments - 40% faster experiment velocity via incentives platform ### Methods `Payment Orchestration`, `Fraud Detection`, `Intent Classification`, `F-beta Tuning`, `Regression Eval Sets`, `Deployment Gating`, `Cold-Start UX`, `CDP Integration`, `Dynamic Pricing`, `A/B Testing` ## Dropbox: The Next Phase of Growth via Multi-Product Adoption **Role:** Director of Product & Strategy · **Link:** https://builtbymikey.com/work/dropbox-multi-product-adoption · **Company:** https://www.dropbox.com > Transforming single-product users into multi-product power users ### The problem Dropbox had become synonymous with cloud storage, but that perception was actually limiting growth. The company had acquired and built amazing products like DocSend, HelloSign, and Capture, yet users weren't discovering them. We faced critical challenges: - 90% of users only touched core storage, completely missing the adjacent products that could make their workflows more powerful - Our acquisition strategy had created siloed products with separate onboarding flows, so users never saw the full ecosystem - Retention was plateauing because single-product users churned at much higher rates than multi-product users - We had no way to intelligently recommend the right product at the right time based on user behavior and needs ### What I did I led a cross-functional initiative to transform how users discovered and adopted Dropbox products: - Smart onboarding redesign: Rebuilt the first-time user experience to intelligently showcase relevant products based on signup context. Sales professionals saw DocSend, while creative teams saw Capture and Replay. - Behavioral recommendation engine: Built a "right product, right user, right time" ML system that analyzed storage behaviors to surface product recommendations. For example, users frequently sharing large presentations got nudged toward DocSend for better analytics and control. - Seamless activation flows: Created lightweight in-app experiences that let users try adjacent products without leaving their current workflow. Someone could activate HelloSign to collect a signature without ever leaving a shared folder. ### What happened We successfully repositioned Dropbox from a storage utility to a complete content collaboration platform, unlocking a new phase of sustainable growth: - **Revenue expansion:** Generated $8M+ in incremental expansion revenue from multi-product adoption, proving the strategy could drive meaningful financial impact - **Adoption velocity:** Increased users adopting 2+ products by 35%, fundamentally changing how people thought about and used Dropbox - **Retention:** Multi-product users showed 25% better 30-day retention, validating that deeper product engagement creates stickier customers - **Discovery speed:** Users found value in new products 60% faster through intelligent recommendations versus hoping they'd stumble upon features organically - **Platform scale:** The activation framework we built scaled across 6 different product lines, becoming Dropbox's standard approach for launching and growing new products ### At a glance - $8M+ incremental expansion revenue from multi-product users - 35% increase in users adopting 2+ products - 25% improvement in 30-day retention for multi-product users - 60% faster time-to-value for new product discovery - Cross-product activation flows deployed across 6 product lines ### Methods `Product Recommendation ML`, `User Journey Mapping`, `Cross-Product Analytics`, `Onboarding Optimization`, `Activation Funnels` ## Agtonomy: Bringing Autonomy to Agriculture **Role:** Founding Team Member · **Link:** https://builtbymikey.com/work/agtonomy-autonomous-fleet-platform · **Company:** https://www.agtonomy.com > Founding-team product for autonomous farming — TeleFarmer operator UX and TrunkVision CV product spec ### The problem Specialty crop farming (vineyards, orchards, and citrus groves) was facing a perfect storm. Labor shortages meant farmers couldn't find enough workers during critical harvest windows, and the equipment needs were 3x higher than traditional row crops. We saw key challenges: - Farmers needed to trust unproven startups with their entire harvest, which wasn't realistic for established operations - Major equipment manufacturers had the farmer relationships and distribution but lacked the AI expertise to build autonomous systems - No one had solved the operator experience problem: how do you manage multiple autonomous tractors simultaneously across complex agricultural terrain? - Existing autonomous solutions required line-of-sight supervision, making the economics unworkable ### What I did Joined as an early founding team member to build the 0-1 product and help shape the go-to-market strategy: - TeleFarmer operator platform: Designed the mobile-first interface that lets a single operator manage 10+ autonomous machines simultaneously. The key insight was treating it like fleet management, not individual robot control. - Computer vision for specialty crops: Built TrunkVision, our CV system for centimeter-precision navigation between vine rows and around tree trunks. This was critical because GPS alone isn't accurate enough in these environments. - OEM partnership strategy: Rather than building tractors from scratch, positioned Agtonomy as the "AI Factory" that partners with the "Iron Factory" equipment manufacturers. This let us focus on autonomy while leveraging their existing farmer relationships. ### What happened Created the foundational technology and strategy that established Agtonomy as the autonomous agriculture platform: - **Operator efficiency:** The TeleFarmer platform achieved 1:10+ operator-to-machine ratios, fundamentally changing the economics of autonomous farming - **Autonomy performance:** 99.7% mission success rate across thousands of autonomous operations proved the technology was production-ready for real farms - **OEM validation:** Partnership platform now powers integrations with Bobcat, Kubota, and other major manufacturers, validating the "AI + Iron" collaboration model - **Commercial traction:** Deployed at premium vineyards including E&J Gallo, Treasury Wine Estates, and Silver Oak, showing enterprise customers would trust the technology - **Company trajectory:** The founding team vision and early product work helped raise $42.8M+ and scale pilot programs 500%, establishing Agtonomy as the category leader ### At a glance - 1:10+ operator-to-machine management ratio via TeleFarmer platform - 99.7% autonomous mission success rate in production deployments - Centimeter-precision navigation in specialty crop environments - Partnership integrations with 5+ major equipment manufacturers - Commercial deployments at E&J Gallo, Treasury Wine Estates, Silver Oak - $42.8M+ funding raised; 500% pilot program expansion from founding work ### Methods `Computer Vision (TrunkVision)`, `Fleet Management Platform`, `Hybrid Autonomy Systems`, `Real-time Telemetry`, `Mobile-first Operator UX`, `OEM Integration` ## Serve Robotics (Postmates): Creating the Future of Last-Mile Delivery **Role:** Director of Product & Business Operations · **Link:** https://builtbymikey.com/work/postmates-serve-robotics · **Company:** https://www.serverobotics.com > Owning the Level 4 sidewalk autonomy product spec, fleet platform, and customer/merchant experience ### The problem Short-distance deliveries were destroying unit economics and creating unsustainable delivery operations. Postmates faced fundamental challenges: - Deliveries under 1 mile were unprofitable, losing money on every order because sending a car and driver for a burrito made no economic sense - Driver churn was hitting 40% as couriers avoided short, low-paying trips, creating constant recruiting costs - No one had built a complete product experience for autonomous delivery. How would customers receive packages from a robot? How would merchants hand off orders? - The autonomy stack needed to work in real cities with real chaos, not just controlled environments ### What I did Pioneered the 0-1 development of autonomous sidewalk delivery, building the product and autonomy systems from the ground up: - Autonomy product spec: Owned the Level 4 sidewalk autonomy product spec — designed the regression eval sets (per-scenario perception, planning under occlusion, novel sidewalk geometries) and the safety-critical thresholding the perception team optimized against. Owned the depot-release deployment gate: no robot left the lot until eval coverage cleared, and the operator-intervention UX caught the long-tail edge cases the model would not handle on its own. The product bar was predictable and safe over fast and clever. - Operator platform: Designed fleet management tools enabling 1:many operator-to-robot supervision. Built remote intervention systems so one operator could monitor multiple robots across the city, making the unit economics viable. - Customer and merchant experience: Owned the end-to-end delivery product from merchant handoff to customer unlock. Designed intuitive robot unlock flows, merchant integration workflows, and optimized every touchpoint to make autonomous delivery feel seamless and delightful. - Delivery optimization: Redesigned the delivery experience through user research, reducing pickup time by 35% and cutting support interventions by 300%, proving the product could scale efficiently. ### What happened Built the foundational product and technology that established autonomous sidewalk delivery as a viable category: - **0-1 Launch:** Pioneered first commercial autonomous delivery robots, managing hardware partnerships and regulatory approvals to launch across 5 markets - **Delivery performance:** 1,000+ autonomous deliveries proving the product and autonomy systems worked in real-world conditions - **Operational efficiency:** Achieved 1:8 operator-to-robot ratios through smart fleet management, fundamentally changing the economics of last-mile delivery - **Product experience:** 35% faster pickup times and 300% reduction in support interventions showed customers and merchants could trust robots - **Company impact:** The product and technology foundation enabled Serve to spin out as an independent company, ultimately going public and partnering with Uber Eats for large-scale deployment ### At a glance - Pioneered 0-1 autonomous delivery across 5 markets - 1,000+ deliveries managed through product and autonomy systems - 1:8 operator-to-robot ratio via fleet management platform - 35% reduction in pickup time; 300% fewer support interventions - Platform enabled Serve spinout, IPO, and Uber Eats partnership ### Methods `Level 4 Autonomy`, `Regression Eval Sets`, `Safety-Critical Thresholding`, `Depot-Release Deployment Gate`, `Operator-Intervention UX`, `Fleet Management Platform`, `Robot UX Design`, `User Research` ## Just Eat Takeaway: Building a 3-Sided Delivery Marketplace **Role:** Group Product Lead · **Link:** https://builtbymikey.com/work/justeat-3sided-marketplace · **Company:** https://www.just-eat.com > Scaling from 0 to 1B+ deliveries through courier network and acquisition integration ### The problem Just Eat had built a successful 2-sided marketplace connecting restaurants with customers, but we were falling behind. Deliveroo and Uber Eats were offering faster, more reliable delivery with their own courier networks, and we were stuck relying on restaurants to handle their own delivery. We faced existential challenges: - Restaurant-managed delivery was slow and inconsistent, creating poor customer experiences that drove users to competitors - We had no control over the delivery experience, the most critical part of the customer journey - Venture-backed competitors were aggressively expanding with their 3-sided models (customers, restaurants, couriers), threatening our market position - We needed to add an entire third side to our marketplace (couriers) without breaking the existing business or alienating restaurant partners ### What I did Led the complete transformation from 2-sided to 3-sided marketplace, building courier operations from scratch while integrating a major acquisition: - Courier marketplace strategy: Formulated the marketplace strategy and platform architecture for our courier network. Built the 0-1 product including courier-facing mobile apps, real-time dispatch systems, and operational frameworks that could scale globally. - Skip acquisition integration: Executed the £200M Skip acquisition, overseeing technical consolidation and user migration. This wasn't just merging codebases. It was combining two different marketplace models and migrating millions of customers without service disruption. - Global expansion: Scaled from single-market proof-of-concept to 14-country rollout. Built the platform to handle diverse regulations, cultural differences, payment systems, and operational models. Each market needed localization while maintaining a unified platform. - Product development: Managed development of web and mobile apps with real-time tracking, dynamic routing, and payment systems. Built the tools restaurants needed to integrate with our courier network seamlessly. ### What happened Successfully transformed Just Eat into a full-service delivery platform, enabling the company to compete with venture-backed rivals while maintaining profitability: - **Platform scale:** Scaled from 0 to 1 billion+ deliveries, establishing Just Eat as a dominant player in food delivery across multiple markets - **Marketplace growth:** Connected 10M+ customers with restaurants through the new 3-sided marketplace, generating £50M+ in annual revenue - **Acquisition success:** The Skip integration drove 550% YoY order growth, proving we could successfully merge and scale acquired marketplaces - **Operational excellence:** Achieved 90%+ delivery accuracy across diverse global markets with 20%+ courier retention, showing the operational model worked at scale - **Network expansion:** Launched 14-country courier network in 18 months with 100K+ active couriers, building the infrastructure to compete globally - **Customer impact:** 25% faster delivery speeds increased order frequency by 15%, driving sustainable growth through better customer experience ### At a glance - Scaled from 0 to 1B+ deliveries through 3-sided marketplace transformation - £50M+ annual revenue; 10M+ customers connected via platform - £200M Skip acquisition integration driving 550% YoY order growth - 14-country courier network with 100K+ active couriers - 90%+ delivery accuracy; 20%+ courier retention across markets - 25% faster delivery driving 15% increase in order frequency ### Methods `3-Sided Marketplace Architecture`, `Real-time Dispatch`, `Route Optimization`, `Multi-Country Operations`, `Acquisition Integration`, `Mobile Development` ## PwC: Transforming Enterprise GTM Through Product-Led Growth **Role:** Management Consultant, Manager · **Link:** https://builtbymikey.com/work/pwc-plg-transformation · **Company:** https://www.pwc.com > Rebuilding go-to-market strategy and operational infrastructure for a Fortune 10 software company ### The problem A Fortune 10 enterprise software company was losing market share to nimbler competitors despite having a technically superior product. Their traditional sales-heavy approach was fundamentally misaligned with how modern buyers wanted to evaluate and purchase software: - 18-month sales cycles were industry standard, but customers wanted to "try before buy" and experience value before committing to six-figure contracts - Product usage was completely disconnected from the sales process. Sales didn't know which prospects were actively using the trial, and product teams had no visibility into revenue impact - Customer acquisition costs were ballooning as the sales team chased deals blindly, with no way to distinguish genuine product engagement from tire-kickers - Cross-functional chaos reigned: Product, Sales, Marketing, and Customer Success operated in silos with conflicting priorities and no unified strategy ### What I did Led a comprehensive transformation from enterprise sales to product-led growth, serving as management consultant focused on product strategy and operational excellence: - Value stream mapping: Conducted deep-dive analysis across 15 business units to identify bottlenecks, waste, and misalignment. Mapped the entire customer journey from first touch to expansion, revealing where friction killed deals and where product engagement predicted revenue. - Product-led growth framework: Designed a new GTM motion where the product itself drives acquisition, conversion, and expansion. Built self-serve onboarding experiences that demonstrated immediate value, turning free trials into qualified pipeline without sales involvement. - Revenue attribution system: Architected analytics infrastructure connecting product engagement metrics to revenue outcomes. Sales could now prioritize high-intent users showing strong adoption signals, while product teams understood which features drove deal closure. - Cross-functional alignment: Restructured organizational workflows to break down silos. Created shared KPIs across Product, Sales, Marketing, and CS teams, ensuring everyone optimized for the same outcomes: user activation, product engagement, and revenue growth. ### What happened The transformation established product-led growth as the company's primary revenue engine, fundamentally changing how they acquired and expanded customers: - **Cost efficiency:** Eliminated $50M+ in operational waste through process optimization and improved sales-product alignment, redirecting resources toward high-ROI initiatives - **Acquisition economics:** Reduced customer acquisition costs by 40% as product engagement pre-qualified leads, allowing sales to focus on high-intent prospects ready to buy - **Speed to market:** Accelerated time-to-market by 60% through product-sales alignment. New features could be tested with users immediately rather than waiting for sales enablement cycles. - **Adoption velocity:** Improved product adoption by 25% through redesigned onboarding that demonstrated value within the first session, not weeks later - **Platform expansion:** Scaled the PLG framework across 8 different product lines, creating repeatable playbooks for launching and growing products without proportional sales headcount increases - **Organizational impact:** The operational infrastructure and cross-functional workflows became the foundation for the company's growth strategy, enabling them to compete effectively against product-led disruptors ### At a glance - $50M+ operational waste eliminated through process optimization - 40% reduction in customer acquisition costs via product-led motion - 60% faster time-to-market through product-sales alignment - 25% improvement in product adoption through redesigned onboarding - PLG framework scaled across 8 product lines - Organizational transformation across 15 business units ### Methods `Product Analytics`, `User Journey Mapping`, `GTM Strategy`, `Process Optimization`, `Revenue Attribution`, `Cross-functional Workflows` ## WWF: Scaling Earth Hour Through Digital Engagement **Role:** Campaign Lead · **Link:** https://builtbymikey.com/work/wwf-digital-conservation-platform · **Company:** https://www.worldwildlife.org > Leading global campaign to drive conservation action through digital transformation ### The problem Traditional conservation campaigns generated awareness but struggled to convert that attention into sustained action. Earth Hour had massive reach (millions participated in the lights-out moment) but engagement was superficial and episodic: - Supporters engaged once a year during the event, then disappeared until the next Earth Hour - Awareness didn't translate to behavior change. People turned off lights but didn't adopt long-term conservation practices or donate - The digital experience was fragmented and outdated, failing to capture the energy of a global movement or make participation feel impactful ### What I did Led the digital transformation of Earth Hour, rebuilding the campaign infrastructure to convert awareness into sustained conservation action: - Campaign platform rebuild: Designed and launched a mobile-first digital platform that made participation tangible and shareable. Users could pledge actions, see real-time global participation, and understand their collective impact. - Behavioral engagement strategy: Architected a year-round engagement model beyond the single event. Created content personalization that matched users with relevant conservation issues, transforming one-time participants into ongoing advocates. - Data-driven storytelling: Built analytics infrastructure connecting individual actions to measurable environmental outcomes, making abstract conservation goals feel personal and achievable. ### What happened Successfully transformed Earth Hour from a one-night awareness event into a year-round digital conservation movement: - **Engagement growth:** 340% increase in website engagement through mobile-optimized experiences and personalized content that kept users coming back beyond the annual event - **Community building:** 180% growth in monthly active supporters, proving the platform could sustain engagement year-round rather than just during Earth Hour - **Conversion impact:** 250% improvement in donation conversion by connecting participation to tangible outcomes, showing supporters how their contributions drove real conservation work - **List growth:** 500K+ new email subscribers acquired through optimized signup flows, building a foundation for ongoing digital campaigns - **Revenue impact:** $1.2M+ in additional fundraising revenue generated through improved digital experiences and conversion optimization ### At a glance - 340% increase in website engagement via mobile-first platform - 180% growth in monthly active supporters beyond annual event - 250% improvement in donation conversion rates - 500K+ new email subscribers acquired - $1.2M+ additional fundraising revenue generated ### Methods `Campaign Management`, `Behavioral Analytics`, `Content Personalization`, `Mobile Optimization`, `Conversion Tracking` ## Notes Short essays on product and leadership at https://builtbymikey.com/notes. Each one starts from a real decision. ### Pushing teams without pushing people *Published 2026-10-03 · 4 min read · https://builtbymikey.com/notes/pushing-teams-conways-law* > Teams stall on ambiguous problems because they lack a shared picture, not shared information. Give them a vision concrete enough to rule things out, then organize the work on purpose, because the shape of the team becomes the shape of the product. ### Don't blame the model for your layoffs *Published 2026-05-15 · 3 min read · https://builtbymikey.com/notes/ai-layoffs-and-team-morale* > AI didn't fire anyone. Leaders did. Hiding a business decision behind the model disrespects the people leaving and wears down the trust the people who stay need to ship good AI products. ### Why I picked Haiku for Field Kit's either-or questions *Published 2026-05-08 · 3 min read · https://builtbymikey.com/notes/haiku-for-decision-simulator* > When the job is filling in a fixed structure, Haiku 4.5 starts answering well under a second sooner than Sonnet at a fraction of the price. Pick a model per task, not per product. ### Three AI product bets for 2026 *Published 2026-05-03 · 2 min read · https://builtbymikey.com/notes/three-ai-bets-for-2026* > Write the eval set before the spec, ship agents like teammates, and route each task to the right model instead of the biggest one. AI product work is starting to look like systems design. ## Field Kit Live at https://builtbymikey.com/field-kit. A visitor asks a product question. Claude Haiku 4.5 writes a short first read that appears in about a second. Claude Sonnet 5.5 writes the full answer and picks one of four shapes: a choice between two options, steps in order, a diagnosis, or a list of risks. Every line carries a source: one of Michael's principles, one of his projects, or words from the question. The answer also says what it is unsure about and names facts the asker did not give. If the visitor pushes back, the answer only changes when the pushback adds a new fact. How it is built: the answer is streamed as JSON that must match a Zod schema, and the page validates it again before showing it, so malformed output becomes an error message instead of a broken page. The full answer sees his playbook and a nine-entry case file of his projects. The first read sees only the playbook, so it cannot borrow a project detail it might get wrong. The instructions, playbook and case file form an identical prefix of about 11,900 tokens, cached for one hour, and a per-question ranking hint comes after it. Token counts, timings and cost under each answer are read from the API response, and the page shows a dash when a number is missing. How it is tested: 33 questions, each run 3 times. Code checks the format, the rules and every citation. A Haiku 4.5 grader scores what code cannot judge against a written rubric. A check that fails on every run blocks the release, and one that fails on some runs is flagged as flaky. Runs are recorded so they can be replayed for free when prompts change. What went wrong, as the page itself says: early cost figures were estimates, the prompt cache had never fired because the cached prompt was too short, an early version scored options out of five with no way to measure them, and one version invented details about his current role. Each was fixed and now has a check. ## Headline numbers - **Revenue:** $100M+ incremental GLP-1 revenue in year one at Hims & Hers; $8M+ expansion revenue at Dropbox; £50M+ annual revenue at Just Eat Takeaway; $1.2M+ additional fundraising at WWF. - **Scale:** 1B+ deliveries and a 14-country courier network with 100K+ active couriers at Just Eat Takeaway; 1,000+ autonomous deliveries in 5 markets at Serve Robotics. - **Conversion:** 23% less fraud with 8% higher payment conversion and 18% higher onboarding conversion at Hims & Hers; 35% more users on two or more products at Dropbox. - **Efficiency:** $50M+ operational waste removed and 40% lower acquisition cost at PwC; 99.7% autonomous mission success at Agtonomy. ## Common questions **What roles is he looking for?** Full-time Head of AI Product or VP Product roles at AI companies and AI-native consumer products. Not taking fractional, consulting or advisory work. **What does he own on AI work?** What a correct answer looks like, the eval set, the precision and recall tradeoff seen from the user's side, fallbacks when there is too little data, what the user sees when the model is unsure, recovery when it is wrong, and the logging that turns user behavior into improvements. **Where can I see his AI work?** Field Kit at /field-kit, including how it is tested and what went wrong. Ask Enzo, the chat assistant on every page, answers questions about his work using seven tools. **How do I reach him?** hi@builtbymikey.com or https://www.builtbymikey.com/schedule. He usually replies within 24-48 hours.